PetAi Guide

This works in the main chat on the Chats page and in the PetAi Agent panel (the round 🤖 button), JARVIS included.

1. Formatted answers

   Answers in the Agent panel and in the 👥 Workers results now look like the main chat: lists, tables, quotes and links. In the panel a code block shows in a grey box that scrolls sideways. Only web addresses that start with http or https become links. Anything in an answer that looks like a web page's code is shown as plain text, so an answer cannot add a button to the page.

2. Copy a message

   Under each answer there is a small Copy button. Your own messages have one too: point at the message and it appears. Press it and the button says Copied for a moment. It copies the message as it was written, with its marks, so it pastes well into a note. The thinking box is never copied. If the browser blocks it, the button says "Not copied"; then select the text and copy it yourself. In 👥 Workers, each worker's output and the summary have their own Copy button.

3. Edit a question you sent

   Point at one of your own messages and press Edit, beside Copy. The message turns into a box with Cancel and Send. Change the words and press Send (or Enter; Esc cancels). That question and everything after it leave the chat, and a new answer is written from there. The old answers are gone, as with Undo. There are no versions to flip between.

   If the question had a file or a picture, the box says "The file or picture is not sent again. Attach it again if the new answer needs it." In the Agent panel, when a removed answer already did something on the PC, the box also says "Actions from the removed answers stay done. Use their Undo if you need to." In the main chat, while an answer is still coming, it says "The answer is still being written. Wait for it, or press Stop, then edit."

4. Read an answer aloud

   Each answer has 🔊 Read aloud beside Copy. It reads that answer, the whole of it, never the thinking. While it reads, the same place shows Pause and Stop. After Pause the button reads Resume. Only one answer is read at a time: Read aloud on another answer takes over.

5. The reading bar in the Agent panel

   In the Agent panel, a bar over the chat box says "Reading aloud" with ❚❚ Pause and ■ Stop. After Pause it says "Paused" and the button becomes ▶ Resume. The same bar shows when JARVIS reads its own reply. While a reading is paused, the JARVIS mic stays off. It listens again when the reading ends.

6. Stop really stops

   Before, stopping JARVIS's reading only cut the sentence it was on, and the next sentence started. Now, in the Agent panel, anything that stops the reading ends it at once: the Stop button, a new question, Clear, Undo, Retry, closing the panel or turning JARVIS off. The JARVIS mic then listens again. These stop only the panel's reading. A reading started in the main chat goes on until you stop it there, open another chat, or edit the question it answers.

Some browsers, and the desktop panel window, cannot pause speech in the middle of a sentence. Then Resume reads that sentence again from its start.

1. Formatted answers

   On the Assistant tab, answers show bold, lists, tables, code in a box that scrolls sideways, and links you can tap. Only http and https addresses become links. Workers results, Research reports and Compare answers look the same.

2. Select or copy

   Press and hold any message to select its text with the phone's own handles. Every message also has a copy icon (two sheets). Under an answer it sits beside the speaker icon; under your own message it is at the bottom right. Tap it and "Copied" shows for a moment. It copies the whole message, never the thinking.

3. Edit a message you sent

   Your own messages have a pencil icon beside the copy icon. Tap it, and the message becomes a box called "Edit your message" with Cancel and Send. Send gives a fresh answer from that point; the answers after it are gone. While an answer is still coming the pencil is dimmed, and a tap says "The answer is still being written. Wait for it, or press Stop, then edit." If the message had a picture or a file, the box says "The picture is not sent again." or "The file is not sent again."

4. Read aloud with Pause

   Tap the speaker icon under an answer. It now reads the whole answer, not only the start. A bar above the chat box says "Reading aloud" with Pause and Stop. After Pause it says "Paused. Resume starts the sentence again." with Resume and Stop. The phone cannot pause in the middle of a sentence, so Resume starts that sentence again. The same bar shows while JARVIS reads its reply, and the JARVIS mic stays off while a reading is paused.

Two small fixes on the phone. A line the phone writes itself when something fails, such as losing the PC, is no longer sent to the model, so the next answer does not react to it. And a chat on the phone's own model no longer stops acting on the phone after four actions: the limit of four now counts per message.

• Nothing to switch on. It works once the PC runs the new PetAi and the phone has the new app.
• Whether to edit or to ask again in a new message. An edit removes the answers after it for good.

1. Your old Agent chat moves once

   The first time you open the Agent panel after the update, the chat it kept in the browser is moved to the PC. The panel says "Your Agent chat is now kept on the PC." If the PC cannot take it just then, the panel goes on as before and tries again next time.

2. Open your chats

   Press 🗂 in the panel's top row. Its tooltip is "Chats: your earlier Agent chats, kept on this PC (the phone sees them too)". A list called 🗂 Chats opens, newest first, with the time and how many messages each holds. Click a chat to open it. ✎ renames it, and 🗑 deletes it after asking. Close hides the list. With no chats yet it says "No saved chats yet. Ask something and the chat is kept here."

3. Start a new chat

   Press + New chat in the list, or ➕ in the panel's top row, where 🗑 used to be. Its tooltip is "New chat: start an empty chat. This one stays under 🗂 Chats." The chat you had stays in the list.

4. A long chat

   Once part of a chat goes to the model as a summary, a line under the context meter starts "Remembers", gives a number of earlier messages and ends "as a summary." ⬇ Shorten now brings the summary up to date at once; every message stays in the chat.

5. Change the model

   Pick another model in the panel's Model list and ask your next question. Under the answer a line starts "Switched from", names the old and the new model, and says "it read the earlier chat". It shows once per change, and again when you reopen the chat.

6. The main chat remembers too

   On the Chats page, a very long chat no longer drops its early messages. The same two lines show under an answer when a summary was used or the model changed. With 🔧 Tools ticked, the chat now remembers your own questions, does not mix in another chat's answers, and is still there after a reload.

7. One answer at a time

   If the phone and the PC ask in the same chat at once, the second one is told "An answer is still being written in this chat. Wait for it to finish, then ask again."

The summary is written by the Agent's own models in the background, and in the main chat by the model you picked. It costs one extra request now and then, only in long chats. With 🕶 incognito on, nothing is saved as before. With 🔧 Tools on, an answer that read a web page earlier in the chat cannot trade, send or change anything in a later answer; it tells you what it would do instead.

1. Your chat comes back

   Closing PetAi, a phone restart or a PC restart no longer empties the Assistant tab. The chat you had opens again.

2. Open your chats

   Tap Chats in the Assistant tab's top bar. With a PC paired the list says "Kept on your PC. The PC's Agent panel lists the same chats." These are the same chats as the PC's 🗂 list, so a chat begun on the phone goes on at the PC. With no PC it says "Kept on this phone." Tap a chat to open it. Delete asks "Delete this chat?" with Delete and Keep. New chat starts an empty one, and Close closes the list. With nothing saved it says "No saved chats yet."

3. Change the model

   With a PC paired, change the Agent's model under Agent model in Settings. With no PC, change the phone's own model with Change model at the top of Settings. The chat stays, and the next answer has the line that starts "Switched from" and says "it read the earlier chat".

4. Retry, Undo and Shorten

   Retry and Undo take the question back in the saved chat too. If the PC cannot do that just then, the exchange comes back with "Your PC still has that exchange, so it is back in the chat." Shorten brings the summary up to date and keeps every message.

5. A chat deleted on the PC

   If you deleted the open chat on the PC, the next question on the phone says "That chat is gone. Start a new chat."

With the 🕶 chip on, nothing is saved and Chats is hidden. Turning it off brings the saved chat back. Chats kept on the phone are left out of Android's cloud backup.

• Which chats to keep. Delete old ones from 🗂 on the PC or Chats on the phone; a deleted chat cannot be brought back.
• Use 🕶 incognito for a chat that should not be kept at all.

Open the Agent with the round 🤖 button, then press 🧰 to open the Toolbox. It loads your notes each time it opens.

1. Keep memory on

   In the tick list at the top of the Toolbox, keep Memory (across sessions) ticked. It is on by default. With it off, the Agent does not save or look up notes.

2. Find the Memory section

   Scroll past 🔒 Permissions and 💻 Where commands run to 🧠 Memory. It says "What I remember about you, across sessions." and "Pin a note and it stays: the Agent can never forget it and Clear all leaves it. You can still edit it, or forget it yourself with ✕."

3. Read a note

   Facts from your profile come first, in bold. Then the notes, pinned ones first. Under each note, a small grey line says where it came from: "you added this", "from a chat" or "from the Agent". Notes saved before this update have no such line. Each note has three buttons: Pin, ✎ and ✕. With nothing saved it says "Nothing remembered yet."

4. Pin a note

   Press Pin. The note moves to the top with a 📌 and a light tint, and the button now reads Unpin. A pinned note stays until you unpin it or forget it yourself.

5. Fix a note

   Press ✎. The note becomes a text box with Save and Cancel. Ctrl+Enter saves and Esc cancels. An empty box says "A note cannot be empty - use ✕ to forget it." If it cannot be saved you see "That change was not saved." and your words stay in the box. Only the words change; the pin and the source line stay as they were.

6. Show one kind

   When your notes have more than one kind, a Kind list appears at the top. It starts with All, then each kind the Agent used, such as fact, preference or project. Pick one to see only those notes ("No notes of that kind." if there are none). Pick All to see everything again.

7. Forget one note

   Press ✕. A note that is not pinned goes at once, with no question. For a pinned note the browser first asks "This note is pinned. Forget it anyway?"

8. Clear all

   Press Clear all in the Memory heading. The browser asks "Forget everything the Agent has remembered about you? Notes you pinned are kept." After OK, every unpinned note goes. If some were pinned, the list says, for example, "Kept 2 pinned notes - unpin one first to let Clear all forget it." The bold profile lines stay too.

9. Add a note yourself

   There is no Add box. In the Agent's chat box, type something like:

   PetAi Agent

   Remember that my shop closes at 9 pm.

   Copy

   The Agent saves it, and the note reads "from a chat". The Memory list has no Refresh button, so close the Toolbox and open it again with 🧰 to see the note. Then press Pin if it should never be forgotten. Whether the Agent really saves it is up to the model, so check the list.

What the Agent itself cannot do: delete a pinned note, delete notes by a phrase, or open the memory file. With "Deleting what the Agent remembers" ticked under 🔒 Permissions, it asks you before it forgets any note, even with JARVIS on.

1. Find the card

   Open Settings and scroll to What the Agent remembers, below Permissions and above Floating avatar. It shows only while the phone is paired. It says "What the Agent keeps about you between chats. Pin a note and it cannot forget it by itself."

2. Read a note

   Each note shows its words and a small line such as "pinned - fact - from a chat": "pinned" only when it is pinned, then its kind, then where it came from, in the same words as the PC. Pinned notes come first. The first 20 show until you tap Show all N. With nothing saved it says "Nothing remembered yet."

3. Pin, fix or forget

   Each note has three icons with no text: a pin, a pencil and a bin. Tap the pin to pin or unpin. It is bright when pinned and faded when not.

   Tap the pencil to open Edit this note. Change the words and tap Save (greyed out while the box is empty), or Cancel.

   Tap the bin to open Forget this note?. It says "The Agent will no longer know this.", or for a pinned note "It is pinned, but you asked for this one: it goes anyway." Tap Forget it or Keep it.

4. Read the list again

   Tap Refresh at the bottom of the card.

5. Add a note from the phone

   On the Assistant tab, set Actions run on to PC and say something like "Remember that my shop closes at 9 pm." Then go to Settings, What the Agent remembers, tap Refresh, and pin the note.

The phone card has no Kind list, no Clear all and no profile lines. Those are on the PC only. If the PC runs an older PetAi, pinning or editing says "Update PetAi on your PC to pin or edit notes from here."

• Which notes to pin. Those are the ones the Agent can never forget and Clear all keeps.
• Whether the Agent keeps notes at all: the Memory (across sessions) box in the Toolbox.

This is on the Chats page. There is nothing to switch on: it is the first choice in the Purpose list.

1. Find it

   Above the chat there is a row of boxes. Open Purpose. The first choice is Auto detect. A new PetAi starts on it, and so does a browser where you never used the Purpose list. If you had "All models" saved, it now shows as Auto detect. Under it, the Model list starts with Auto (picks per message), then every model.

2. Send different kinds of message

   Leave Purpose on Auto detect and the Model on Auto (picks per message), and send these one at a time. The rules are simple, so a message may be read differently; they also know some Hindi and Punjabi words.

   • Fix this Python function: def add(a, b): return a - b. It should be read as coding.
   • What is the share price of Reliance today? It should be read as finance.
   • Draw a picture of a red fox. It should be read as an image.
   • Explain how rain forms. Nothing special here, so it is plain chat.
3. Read the line under the answer

   Under each answer Auto detect routed, a small line says what it did: the word Auto, the kind it chose, the model, and why. For the coding message it reads like: This looks like a coding request in Python. Next to it is a link, Change. Click it: the Purpose switches to the kind Auto detect chose, so the Model list shows only that kind's models, the list opens, and a note says "Pick one, or choose Auto detect again." Choose Auto detect in the Purpose list to go back.

4. The question before a slow one

   Auto detect asks first before it starts a video, and before it makes a picture on a paid provider. Send the message Make a video of waves. The chat shows a question under your message: that this looks like a video request, which model it would use, and that it can take several minutes. Two buttons: Use it sends it as picked, and Answer in chat instead sends the same words to a chat model. While the question waits, another Send shows "Answer the question above first: Use it, or Answer in chat instead."

5. Your own pick always wins

   Pick a model by hand in the Model list and press Send. Nothing is routed and no line appears; the chat stays an Auto detect chat. Pick a picture model by hand and send the message Draw a red fox: you get a picture, whatever the Purpose says. That is the bug fix.

6. If Auto detect cannot be reached or cannot pick

   The message goes to the first chat model that works, and a note says so. If no model is listed at all, you see an error asking you to choose a model first, and your words are put back in the message box so nothing you typed is lost.

7. See it in Settings

   Open ⚙ Settings & appearance and the Auto detect group. It explains the same in a few lines and tells you to use the Purpose list. It has no switch. The plan also had an optional smarter mode that asks a small model, but that was not built.

8. The Agent knows about it too

   In the Agent panel, the Model list starts with Auto (best available, with fallback). With that chosen, the Agent now tries first the models that fit what you asked (for a coding question, code models first) and keeps the old order for every other question. A model you pick by hand is never changed. You can also tell the Agent to set the Purpose to Auto detect. On the PC the Agent shows nothing extra; on the phone a short line says what it picked (see the phone group).

The phone has no Purpose list, so Auto detect is not there. It has the line "Auto picked ..." under Agent answers: see the phone group.

• Whether to use Auto detect at all: pick another Purpose, or a model by hand, and it steps aside.
• Whether a video, or a picture on a paid provider, goes ahead: you answer the question first.

This is on the Chats page, and on the AI Code page in the left menu. There is nothing to switch on.

1. Files keep their names and folders

   When a reply gives each file a name (path=src/App.jsx after the three backticks, a first-line comment with the name, or a bold file name in the line above the code), the pane Files from this chat keeps it. Files are grouped under their folder, headed with 📁 and the folder name, and files with no folder sit under Project root. Saving a file uses its own name. Download all (.zip) keeps the folders. A name that is not safe is refused with the reason, as in the zip part.

2. Ask for a whole project

   Set Purpose to Coding. A box called Build appears next to Language. Its choices are One file or snippet (as before), Website (HTML, CSS, JS), Web app (React + Vite), Web app with database (Express + SQLite), Python API with database (Flask + SQLite), Android app (Kotlin + Compose), Flutter app and React Native app (Expo). Pick Website (HTML, CSS, JS), type a request such as a todo website with add, complete, delete and a filter, and press Send.

   Instead of text, the answer area shows "Planning the project..." and then which file it is writing now, with a count such as 3 of 9. If files are missing, it says it is getting the missing files and counts the round, at most 4 follow-ups. A small or free model often cannot write a whole project in one go, so expect this.

3. The Project card

   The answer ends with a card headed Project with the project's name and the number of files, a folder tree, and a folder-by-folder list under Files. If some files are still missing it says, for example, 7 of 9 files - 2 are missing, lists their names, and offers Ask for the missing files. If the model did not list its files first, the card says "The model did not list its files first, so PetAi can only check the links between them." Buttons: Download .zip (with the folders), Open in AI Code, Build it properly in AI Code, and Preview, which appears only when the project has an index.html.

4. Preview in the chat

   Click Preview on the card. The page opens under the card in a sealed frame: its own scripts run, but it cannot reach PetAi or your other tabs. Buttons Phone, Tablet and Desktop change its width. The button now reads Hide preview. A page that asks its own server for data will not show that data, because the server is not running. A React + Vite project, a mobile app or a project with no index.html gets a short note in place of a page.

5. Hand it to AI Code

   Open in AI Code makes a new AI Code project with these files, folders kept, and takes you to it. Undo this run there takes them out again. Build it properly in AI Code sends only your request, with the same model, so AI Code plans the project, writes one file at a time, builds it, runs the project's own test suite when it has one, and fixes whichever one breaks before calling itself done.

6. Link a local folder

   Next to Open from GitHub in AI Code's right-hand panel is Local folder: paste the full path to a real folder already on this PC and press Link folder. The agent then reads, edits, runs and commits that folder in place - like opening Claude Code on a project folder - instead of a sandboxed copy. It refuses a drive root, your home folder, or a path already inside AI Code's own projects folder, since that would let it touch everything on the PC rather than one project.

7. Starters in AI Code

   On AI Code, in the row of boxes above the task, a list begins Starter: from your words, then Starter: none, then the same stacks as Build. A starter is the handful of files a stack always needs. They are written first, shown in the file list with a small badge, starter, and then the planner writes the rest. The list disappears once the project has files; the big button then reads Apply change instead of Build.

   Android, Flutter and React Native projects are written but never built or previewed here: the list's tooltip says "Android, Flutter and React Native are not built or previewed here." Each one's README names the first command to run (for Flutter, flutter create . ; for Android, gradle wrapper, then open the folder in Android Studio). The planner writes up to 120 files in one plan and says in its log when it had to cut a longer one.

8. Clone a website

   Next to the starter list is a checkbox, 🔗 Clone a website. Turn it on and send any prompt for a new project, and before building anything PetAi asks you, right there in the chat, four things in turn: the URL you want to start from; what you want changed versus the original (content, branding, colors, sections) - or "no changes, just clone it" for a straight copy; whether you want the result scaffolded as Next.js with ISR or left as organized static HTML; and a page design style - minimal, maximal/bold, classic/corporate, playful, or auto to let the agent pick one from the site's content/niche and say which one it picked and roughly why. Once all four answers are in it crawls the site - same-origin pages only, following internal links up to a cap of 30 pages - and organizes what it finds into a pages/ folder (one HTML file per route, named by its path) and assets/css/, assets/js/, assets/img/, assets/fonts/. The chat log shows progress while this runs ("fetched 12/30 pages…"), and if the 30-page cap is hit before the crawl runs out of links it says so rather than going quiet. This is a plain web fetch, not a browser: it captures a page's structure and look, not a script's behavior after the page loads, so a simple site comes through well and a JavaScript-heavy app's interactions will not; images and fonts are listed by URL in an assets/img/MANIFEST.txt / assets/fonts/MANIFEST.txt rather than downloaded as bytes. Internal page links and stylesheet/script references that point at something the crawl actually saved are rewired to relative local paths, so the result works correctly when opened straight off disk with a static file server (e.g. VS Code's Live Server) - clicking a nav button opens the right sibling page instead of 404ing or jumping back to the live site. It is meant for using an existing site as a design reference to build your own, so only reuse content and design you actually have the rights to use.

   Say nextjs and it also writes a nextjs/ scaffold: each crawled page becomes one file under nextjs/pages/. A page whose original HTML had a <form>, a fetch/XMLHttpRequest/axios call, or hit an /api/ path gets getStaticProps with a revalidate number wired up (Incremental Static Regeneration); every other page stays a plain export default function Page() with no revalidation. Be clear with yourself about what that actually buys you: ISR is not live server rendering - Next.js serves a pre-built static page immediately and regenerates it in the background at most once per revalidate window, so the very next visitor after a regeneration still gets a cached page, just a fresher one. The scaffold's own README says this plainly and lists which routes it judged dynamic. The dynamic routes' getStaticProps genuinely reads seeded content - a small file-based "database" (nextjs/data/db.json, seeded with each crawled page's title/text and any detected form fields, read through nextjs/lib/db.js) - instead of a TODO, and there's a plain, unstyled admin page at nextjs/pages/admin to view and edit that data (saved back via a Next.js API route, which only works while a Node server is actually running - next dev or next build + next start, not a static export opened off disk). The admin page has no authentication, so the README is explicit that it must be password-protected or removed before deploying for real. It's still a starting point, not a finished Next.js app. Say static instead and you get only the organized pages//assets/ folders, no Next.js scaffold.

9. Preview in AI Code

   Above the file list, click Preview ▶ (grey until the project has files). A Preview panel opens with Phone, Tablet and Desktop, Refresh, Open in new tab and Close. The page runs in the same sealed frame. Open in new tab is the older preview: it starts a small server on the PC and opens the project in a browser tab, which is what a page that calls its own server needs. Without an index.html the panel says "This project has no index.html to preview yet."

10. Android app (.bat), not just a zip

    For a Website or React project (and a Flask or Express project, once its own build has made a static dist/build/public/out folder), a second button sits next to Download .zip: Android app (.bat). It downloads a different zip: your built site, a small WebView-wrapper Android app template, a build script and a BUILD_APP.bat you double-click. It is honest about what that means - read README.txt inside the zip, which says the same thing: the first run needs an internet connection (it downloads Java, the Android SDK and Gradle - a few GB in all, the same tools PetAi.exe and the phone app's own build use) and takes 10 to 30 minutes; later runs are faster. Nothing needs administrator rights. The finished .apk lands in a build folder next to BUILD_APP.bat - copy it to an Android phone and tap it to install. The app shows your site in a WebView and works fully offline once installed; it is not part of the main PetAi phone app and has none of its features. Mobile starters (Android, Flutter, React Native) never get this button: they are not built here at all, by design.

11. Build and Auto detect

    Under Auto detect, a message such as build me a todo website is read as coding, but PetAi does not choose the Build for you: the Build box shows only when Purpose is Coding. As far as the code shows, a Build you chose earlier still applies to a coding message sent under Auto detect even while the box is hidden, so set it back to One file or snippet when you are done.

The phone has no Build box. Code blocks and a project zip on the phone are in the phone group.

• Whether a message is a snippet or a project: the Build box.
• Whether a project goes on to AI Code, and which starter it begins from.
• Nothing leaves your PC except the calls to the model you picked.

1. Live steps and Stop

   While an answer is on its way, a bar under the chat shows "Working on it" and the steps so far. Press Stop to end it before its next step; steps that already ran stay done. If you reload the page, the panel picks the answer up again.

2. The Thinking box

   A model that thinks out loud now shows that part in a closed box called Thinking above the answer. Click it to open it. It is never read aloud, copied or sent back to the model. The main chat on the Chats page does the same.

3. The context meter

   Beside Shorten, a small "ctx" figure with a percentage shows how full the model's context window was. Over about 75%, PetAi summarises the older messages by itself, and a quiet note under the answer says "Older messages were summarised to fit."

4. Memory that learns

   Press 🧰 and scroll to 🧠 Memory. The box Learn from chats automatically is ticked unless you untick it. After every 4th message a model picks out lasting facts about you and things you dislike, from your own words only: never from web pages or files, and never passwords, PINs, codes, keys or card numbers. Such notes say "learned from your chats".

5. Tidy memories

   In the Memory heading, Tidy memories merges notes that say the same thing. Pinned notes are never changed.

6. An incognito chat

   Press the dark-glasses button (🕶) in the panel's top row. Its tooltip is "Incognito: nothing from this chat is saved, logged or learned". A bar under the model list says so too. Press it again to go back to your saved chat. The work you ask for is still done; only nothing is kept. Research cannot run in an incognito chat.

1. Steps, Stop and Thinking

   On the Assistant tab, the answer shows its steps while it works, with Stop. It keeps going while the phone is out of reach of the PC. A thinking model's text sits in a closed Thinking row above the answer. "Older messages were summarised to fit." shows under an answer when that happened.

2. Incognito

   Tap the dark-glasses chip at the top of the Assistant tab. A note under it says that nothing from this chat is saved, logged or learned on the PC. Tap the chip again to go back to your chat.

3. Learning and tidying

   In Settings, the card What the Agent remembers has the switch Learn from chats automatically and a Tidy memories button.

• Whether the Agent learns from your chats: Learn from chats automatically. Turning it off stops it at once.
• When to use incognito. Nothing from that chat is saved, logged or learned.

All of this is in the Agent panel. Its round 🤖 button is on the Chats, Design, AI Code, Customize and Screener pages. It is not in the left menu, and not on Schedule or this Guide.

1. Open the Agent

   Click the round 🤖 button at the bottom-right of the page. You can drag it anywhere. Its tooltip starts "PetAi Agent — ask me anything". The PetAi Agent panel opens.

2. Ask something

   Type in the box ("Ask me… or attach a file/screenshot with 📎 or Ctrl+V") and press Enter or ➤.

3. Read the word under the answer

   Under the answer, on the right, a small grey line says "answered by" and the model's name. Right after it is a tiny outlined word. local is drawn in teal; cloud and gateway in grey. Hover the word to see what it means; the table under these steps has all three. Older answers have no word.

4. Open the Toolbox

   Press 🧰 in the panel's top row. It sits between 👥 and 🤖, and its tooltip starts "Toolbox: choose which tools the Agent can use". A panel titled Toolbox covers the chat. It reads fresh numbers each time it opens.

5. How many answers of each kind

   Scroll to 📈 Usage and insights. The first line reads like "12 turns in the last 7 days, 3 today, …". The next line splits them, for example "Local: 2 turns - Cloud: 8 turns - Gateway: 2 turns". It shows once at least one answer has a word. Refresh beside the heading reads the numbers again. With nothing yet it says "Nothing yet - ask the Agent something."

6. What the Agent did

   The last section of the Toolbox is 📋 Activity, the last 30 tools the Agent ran, newest first. Each row shows the tool's name in bold and a short preview of what it was given. Under that: the time, ok or a red failed, how long it took (like 412 ms or 3.2 s), and "turn" with the first 6 letters of that answer's id. Keys, tokens and passwords are left out, and file contents are never logged. With nothing yet it says "Nothing yet - the Agent has not run a tool."

7. Only one answer's tools

   Press Show turn on a row. The list keeps only the tools of that one answer, and a bar reads like "Only turn 3f9a1c - 3 tool calls". Press Show all to see every answer again. Refresh keeps the view you are in.

8. Back to the chat

   Press ✕ at the top-right of the Toolbox, or 🧰 again.

Only the Agent panel is counted. The chat box on the Chats page and AI Code show no word, and are not in Usage or Activity.

1. The word on replies

   On the Assistant tab, ask something. Under the reply, on the right, a small line reads "Answered by" the model's name, then " - local", " - cloud" or " - gateway". It does not show on a reply from the phone's own model.

2. What the Agent did

   Open the PC tab. Below PC screen (and Waiting for your OK, when something waits) and above Tell the PC what to do is a card called What the Agent did. It lists up to 10 of the latest tools, one per line, like "read_file - ok - 412 ms". A failed one is red and says "failed". The card shows no time and no turn id, and it is hidden while the PC has no tool calls.

3. When the list updates

   The card is read again when you open the PC tab, after Run on the PC, and after Run it or Cancel. It has no Refresh button of its own. The Refresh next to PC screen only refreshes the picture.

4. How many answers of each kind

   Open Settings and scroll to How it talks. At its bottom, under Usage, the first line reads like "12 turns in the last 7 days, 3 today". The next reads like "Local: 2 turns - Cloud: 8 turns - Gateway: 2 turns". With nothing yet it says "Nothing recorded yet - ask the Agent something."

• Nothing to switch on. It works once the PC runs the new code and the phone has the new app.
• To ever see local, set up a model on this PC (G2) and pick it in the Agent's Model list.

Open the Agent with the round 🤖 button, then press 🧰 to open the Toolbox.

1. Find the permissions

   Scroll past the tool groups and the line "N of M tools on." to 🔒 Permissions. It starts with "Ask me before:".

2. Choose what it asks about

   There are six boxes, all ticked on a new install. A ticked box means the Agent waits for your OK first. The table under these steps says what each one covers.

   A tick or untick is saved at once. If it cannot be saved, the box flips back and "That change was not saved." appears.

3. The files it can never read

   Under the boxes, a line starts "The Agent can never read: ". It lists config.json, secrets.json, omniroute-data/, .env files, .git-credentials, .ssh/, the Agent's memory file agent-data/agent-memory.db and a few more files in agent-data. There is no button to change this list.

4. Approve or cancel a waiting action

   When the Agent waits, its answer tells you to press Confirm. Right under the answer is a card headed PROPOSED ACTION and the kind, with "Run" and the tool's name, and a short preview of what it will do. Press Confirm & run: it shows "Running…", then "✓ done" with the output, or "✗" and the reason. Or press Cancel, and the card says "cancelled". Each action runs once only.

   A command that could erase files is marked ⚠ destructive. Its button reads Confirm anyway, and the browser asks you once more.

5. Waiting actions do not wait for ever

   A waiting action is dropped after one hour. Pressing Confirm & run after that shows "✗ no pending action … (it may have expired)". The PC has no separate list of waiting actions: one that the phone or a schedule left waiting shows up under the Agent's next answer.

6. What is always refused

   Whatever the boxes say, and even after a Confirm, the Agent cannot read, copy or change anything on the never-read list, however it asks (a command such as type config.json is refused too). It cannot send a web request to PetAi's own console (port 8123) or to the gateway (port 20128). It cannot forget notes by a phrase, only one note by its number. In 📋 Activity a refused call shows as failed.

7. JARVIS lets some things through

   The 🤖 button inside the panel's top row turns JARVIS on. Its tooltip is "Turn on JARVIS mode (voice replies + mic + admin, Gemini-first)". It is not the round 🤖 that opens the panel. With JARVIS on, file changes, web requests and clicks, and commands run without asking. Paid credit, messages and trades, and deleting memories still ask when ticked. The section then says "JARVIS is on, so changing files, web requests and commands run without asking. Paid credit, messages and deleting memories still ask when ticked." Refusals stay refused.

8. Schedules cannot wait for you

   The last note says "Scheduled routines cannot wait for you: an action that needs your OK is left in the Confirm list." On the PC, that action shows up as a card under the Agent's next answer. On the phone it is in Waiting for your OK. In ⏰ Schedule, such a run is marked Needs you.

1. The same six switches

   Open Settings and scroll past JARVIS mode to Permissions. It shows while the phone is paired. The card starts with "Ask me before:" and has six switches with the same labels as the PC's boxes. A switch saves on the PC. If it cannot, it flips back and says "The PC did not save that." or "Couldn't reach the PC to save that." The phone does not show the never-read list.

2. JARVIS

   The JARVIS switch is in the JARVIS mode card. With JARVIS on, the Permissions card adds "With JARVIS on, file changes, web requests and commands run without asking; the rest still ask when switched on."

3. Approve from the chat

   On the Assistant tab, when the PC is waiting, a card called Waiting for your OK appears just above the message box. Each row shows the tool and a short preview, with Run and Cancel. While it runs you see "Running...". The result is added to the chat, or "Done." if there was no output. The card shows up to three rows, then "and N more on the PC tab".

4. Approve from the PC tab

   On the PC tab, under the heading Waiting for your OK, each waiting action has its own card. It says "Confirm this action:", or "⚠ This looks destructive:" for one that could erase files, then what it will do. Press Run it or Cancel.

• Which of the six boxes to keep ticked. All six are ticked at first.
• What your schedules may do alone. A schedule that should change files, run commands or send web requests on its own needs the matching box unticked, or JARVIS on. Otherwise its actions wait for you, and are dropped after an hour.
• More files to protect. Add them by hand in config.json, under "agent_policy" then "protected_paths". There is no button, and the built-in entries cannot be removed.
• Keep the LOCAL_AI folder with PetAi. Without it, the Agent's tools are switched off.

1. Open AI Code

   Click AI Code in the left menu and run a task as usual. Before the run starts, PetAi saves a copy of the project. A project that uses git keeps its history untouched. A folder without git gets its copies in a hidden .aicode-undo folder.

2. Press Undo this run

   When the run ends, the bar at the top shows Undo this run with a note saying what can be undone. The same button is on the run's done card in the log.

3. Confirm

   The browser lists the files and asks "Undo this run?". After OK, a card in the log lists what was put back, what was removed and anything left alone because it changed after the run.

A Qwen Code run in a folder without git shows only the note, with no button.

1. Open the Activity list

   Open the Agent with the round 🤖 button, press 🧰 to open the Toolbox, and scroll to Activity. Its note says "Undo puts back the files one answer changed."

2. Press Undo

   The newest row that changed files in each answer has an Undo button. The browser asks "Undo the file changes of this turn?". The button reads "Undoing..." and then the row reads "Undone".

3. Or ask the Agent
   PetAi Agent

   Undo your last change.

   Copy

   The Agent asks you first, as for any file change, unless JARVIS is on.

1. Find the card

   Open the PC tab in the bottom bar and scroll to What the Agent did.

2. Tap Undo

   The newest row that changed files in each answer has an Undo button. The dialog asks "Undo the file changes of this turn?". Tap Undo or Keep them. The row then shows "Undone", with a note under the list such as "Undone: put back 2 files, removed 1 file it had made."

The Agent's shell commands are checked more strictly. PetAi does not count a command it cannot fully follow as a simple look, for example a command with a variable, a cd first, a loop, or one that runs python or node. Such a command is refused if it names one of your key files. If it runs in a folder that holds them, such as the PetAi folder with config.json, it waits for your Confirm, even with JARVIS on. Plain looks such as listing a folder, reading an ordinary file or git status run as before.

• Whether to undo. Nothing is put back until you press Undo and confirm.
• Whether a command that waits on its Confirm card runs.

Open the Agent with the round 🤖 button at the bottom-right of the Chats page. JARVIS is the same panel with voice.

1. Make a PowerPoint, Excel, Word or PDF file

   Ask, for example:

   PetAi Agent

   Make a PowerPoint about my shop's sales this month, 6 slides.

   Copy

   Under the answer a small chip shows the file's name and the word Download. Click it to save the file. A new file never replaces one you have. This uses the Files (read & write) group in the Toolbox, which is on by default.

2. Ask about your screen
   PetAi Agent

   What is on my PC screen?

   Copy

   It works only while the sidebar switch Floating desktop bot is on. If it is off, the Agent says "PetAi reads this PC's screen only while its floating desktop bot is on." It also needs one of your models that can see pictures.

3. Show it something with the camera

   Press the camera button (📷) next to 📎, at the left of the chat box. PetAi takes one picture with this PC's camera and shows it small above the box. Press Retake to take it again or Remove to drop it. If the PC has more than one camera, a list lets you pick one, and PetAi remembers it. Then type your question about the picture and send. The camera is only on for that one picture.

4. Tidy a folder

   Ask, for example:

   PetAi Agent

   Tidy my Downloads folder by type.

   Copy

   Nothing moves yet. A card shows a table with File and Moves to, and says "Nothing moves until you press Apply. One Undo in Toolbox > Activity puts everything back." Press Apply to move the files, or Cancel. After Apply the card has an Undo button too; it asks "Put every file back where it was?". It can sort by type, date or name, never writes over a file, and refuses PetAi's own folder and Windows' system folders.

5. Turn on Messaging

   Messaging is off until you turn it on. Press 🧰 to open the Toolbox and tick Messaging (WhatsApp Web, no key). Under it, press Link WhatsApp. A browser window opens on the PC: scan its QR code with WhatsApp on your phone (Settings, then Linked devices). When it works the card says "WhatsApp Web is linked. Messages go from your own WhatsApp after you press Send on the card."

6. Add the people it may message

   In the same card, fill in the name (for example Ravi Kumar), the number with its country code, and other names for the same person if you like. Press Add contact. Each contact has Edit and Remove. The Agent finds people by the names in this list. It can also use a full number with its country code, and that message still waits for your Send.

7. Send a message

   Ask, for example:

   PetAi Agent

   Tell Ravi I'm running late.

   Copy

   A card shows who gets it, the number and the exact words. Press Send to send it from your own WhatsApp, or Cancel. Every message waits for Send, even with JARVIS on. A name that is not in your contacts, or that fits two people, is asked about first and nothing is sent.

1. The same requests

   On the Assistant tab, with Actions run on set to PC, ask for a file, the screen, a tidy-up or a message the same way. The PC does the work.

2. Download a file

   A file the PC made shows under the answer as Download and its name. Tap it. The phone keeps it in Downloads/PetAi (on Android 10 and newer) and opens it in a viewer app. On the PC tab, a file from a confirmed step has the same button.

3. Take a picture for the PC

   Tap the camera icon beside the paperclip. The phone's own camera app opens. Take the picture, and "Picture ready for the PC" shows above the box with Remove. Ask your question and send. The PC answers about that picture.

4. Send a WhatsApp message

   With Actions run on set to PC, the card Waiting for your OK shows "Send a WhatsApp message" with the person and the words. Tap Send. On the PC tab the same card has Send and Cancel.

   With Actions run on set to This phone, the phone opens WhatsApp with the message already typed, and you tap Send in WhatsApp yourself. The names still come from the contacts on the PC.

• Whether to turn on Messaging (WhatsApp Web, no key), and who goes in the contacts. WhatsApp Web needs Playwright on the PC; the card says so if it is missing.
• Whether a tidy-up runs. Nothing moves before Apply, and one Undo puts it all back.
• Whether to turn on Floating desktop bot so the Agent can read your screen.

Open the Agent with the round 🤖 button, then press 🧰 to open the Toolbox. In the tick list, Files (read & write) and the new group Git and GitHub are on by default. Untick a group and the Agent cannot use its tools. Everything below is in words you say to the Agent.

1. Search, read and edit files

   Give it a folder of your own and try:

   PetAi Agent

   Find the word TODO in the files under C:\Work\MyProject and tell me which files have it.

   Copy
   PetAi Agent

   Show me lines 1 to 40 of C:\Work\MyProject\notes.txt.

   Copy
   PetAi Agent

   In C:\Work\MyProject\notes.txt change colour to color and grey to gray, in one go.

   Copy

   Searching skips .git, node_modules, binary files and the files the Agent may never read. When several edits go in together, either all are made or none, and a failed one leaves the file as it was. Changing a file asks for your OK while the box Changing or overwriting files in 🔒 Permissions is ticked, as in G4.

2. Delete with Undo

   Ask it to delete a file:

   PetAi Agent

   Delete the file C:\Work\MyProject\old-notes.txt.

   Copy

   Nothing is erased for good. The file is moved into a folder named PetAi Trash in your home folder, and one Undo in Toolbox › 📋 Activity puts it back. A folder is different: it asks on every request, JARVIS or not, and the card says how many files and folders go with it. The Agent refuses a whole drive, your home folder itself, PetAi's own folder, and a folder that holds files it must never touch.

3. Work folders

   In the Toolbox, under 🔒 Permissions, there is a heading 📁 Work folders. It says "Empty = the Agent may write anywhere except protected files (it still asks first). Add a folder and its file tools change, create and delete files only inside the folders listed here; reading is not limited. Commands it runs are not limited by this list." Type the full path of a folder in the box and press Add folder. Each folder in the list has Remove; with none, the list says "No work folders set."

   Try it: add one test folder, then ask the Agent to create a file in your Documents folder. It is refused with a reason. Ask it to create one in the work folder and it works, asking first as usual. The check follows shortcuts and junctions to the real place, ignores capital letters, and refuses network paths. Only you can change the list, here; no Agent tool can. One honest limit: the Agent's commands are not held to the list, only its file tools are.

4. Look at git, make a branch, commit

   In a work folder, or in a folder named PetAi Repos in your home folder, the Agent can show a repository's status, changes and history (reading works anywhere), start a branch, switch branch and commit. Try, in a repository inside your work folder:

   PetAi Agent

   In C:\Work\MyProject show me the git status. Then make a branch called petai/try-1 and commit README.md with the message Fix a typo.

   Copy

   A commit holds exactly the files you named, never everything and never the files the Agent must not touch. It carries your own git name and a line, Co-authored-by: PetAi Agent. Switching branch is refused while there are changes not yet committed. A repository whose own settings name a program to run is not used, and hooks never run. The changing steps ask like any file change; the looking steps do not.

5. Sign in to GitHub without saving the token

   Open Customize, the Connectors tab and the GitHub card. Its text now says your token stays in Windows Credential Manager (or Git Credential Manager, or the GitHub CLI) and that PetAi does not save it in its own files. Click Sign in to GitHub; it uses the sign-in this PC's Git already has. Or paste a token and click Connect. When connected, the card says where the token rests, for example "The token is kept in Windows Credential Manager, not in PetAi's files." Disconnect removes it. A token an older PetAi saved in its settings file is moved out and wiped from the file the next time PetAi needs it. AI Code's push, the Schedule routines that watch GitHub and the Agent all take it from the same place.

6. Copy a repository, or read one without copying

   A private repository, such as your own PetAi one below, needs GitHub connected first; a public one does not.

   PetAi Agent

   Clone https://github.com/octocat/Hello-World and tell me what is in its top folder.

   Copy

   The copy goes into PetAi Repos in your home folder (a name that is taken gets -2, -3). It is a shallow copy of the latest 50 commits, with no submodules and no hooks. Only https://github.com addresses are used; ssh, file and other hosts are refused. The Agent can also read one file or list one folder of a repository without copying it. What it reads there is treated as outside content, not as orders: if a README tells it to push, the push still waits for your OK like any other, and any step that would change something after it has read a repository asks first, even with JARVIS on.

7. Push a branch, open a draft pull request

   These two always ask, JARVIS or not, and nothing turns that off. Ask for them in a repository whose origin is a github.com address, with GitHub connected:

   PetAi Agent

   Push my branch petai/try-1 to GitHub, then open a draft pull request titled Fix a typo.

   Copy

   The card for a push names the branch and the repository it goes to, says how many commits and files go, and lists each file with its added and removed lines (the first 8 files). A line in red appears when the push changes GitHub Actions workflows, with their names. The buttons are Push and Cancel. The card for a pull request names the repository, the branch and the title, with Open draft PR and Cancel.

   The rules behind it: it never pushes main, master or the repository's main branch, and never forces a push. If the commits to send seem to hold a key or a password, nothing is sent and the Agent says which file. If GitHub already has a different branch of that name, it says so and asks for another name. The pull request is always a draft, and there is no merge tool: you merge it on GitHub yourself.

8. Build a whole project

   Ask for an app or a site and for the files:

   PetAi Agent

   Build me a small todo website with a README and give me the files as a zip.

   Copy

   Under the answer a chip shows the zip's name, "Project (zip)", its size and the word Download. A zip is a new file, so it does not ask. If you name a new or empty folder inside a work folder, it writes the files there instead, one at a time so Undo takes each back; that asks like a file change, and it never writes into a folder that already has files. For a very large project it can hand only your request to AI Code, which then builds it itself (not from a helper or a team bot).

1. A push or pull request waiting for your OK

   With your phone paired, a push or pull request the PC Agent holds shows in the Waiting for your OK card on the Assistant tab in words: "Push a branch to GitHub" or "Open a draft pull request on GitHub", then up to 8 lines of what goes (the branch, the repository, the files and line counts) and, in red, the workflow warning. The buttons read Push or Open PR, and Cancel. This is built, and needs your phone to confirm.

• Which folders the Agent may change files in (📁 Work folders), and whether the Git and GitHub group is on.
• Every push and every pull request, one by one, on the PC or the phone.
• Whether GitHub is connected at all, and which sign-in it uses.

There is nothing to switch on. You see it the next time the Agent wants to send code out.

1. What always waits

   These ask every time:

   • A push: git push, git subtree push, and git remote set-url, which changes where the next push goes.
   • GitHub's command line: gh pr create, gh pr merge, gh repo delete, gh release create, and a gh api call that changes something.
   • The same commands run inside another shell, such as bash -c or powershell -Command.
   • GitHub tools connected to the Agent that push files, open, update or merge a pull request, delete a branch or a repository, or write a file straight to a branch.

   Looking is not affected: git status, git log and reading a pull request still run at once. Saving work on your own PC with git commit is not part of this rule, only sending it out is. Making a new branch, repository or fork follows the Permissions boxes as before.

2. The waiting card

   It is the card from G4: headed PROPOSED ACTION and the kind, with the command on it, and the buttons Confirm & run and Cancel. Read the command before you press Confirm & run. A force push, or a push straight to main or master, says so in the first words on the card. The list under these steps shows the words.

   A force push is also marked ⚠ destructive. Its button then reads Confirm anyway, and the browser asks once more.

3. Nothing turns it off

   These asks are not among the six boxes in 🔒 Permissions, so there is no box to untick. JARVIS does not let them through either: with JARVIS on, a push still waits.

4. If you are not sure, cancel

   Press Cancel. The card says "cancelled" and nothing is sent. A waiting action is dropped after one hour, as in G4.

5. See it without sending anything

   Ask the Agent for a dry run. It shows what a push would send and sends nothing, even after you press Confirm & run:

   PetAi Agent

   Run git push --dry-run origin main in my PetAi folder.

   Copy

   The card still carries the line about main, because PetAi reads the command, not what git does with it.

What the first words on the card say:

• A push to any other branch, such as git push origin feature: nothing extra. The command is on the card.
• A force push, such as git push --force origin feature: "Force push: it can erase commits that are already on GitHub."
• A push to main, or to master: Push to main: it changes that branch on GitHub directly. For master, the same words with master.
• A force push to main, such as git push -f origin main: Force push to main: it can erase commits that are already on GitHub.

The same always-ask rule is in place for deleting a folder. The Agent has no tool for that yet, so it has nothing to ask about today. Commands that erase files are still marked ⚠ destructive, as in G4.

1. The same card, under Waiting for your OK

   When the Agent on the PC holds a push, a pull request or a merge, the phone lists it in the Waiting for your OK card above the message box on the Assistant tab, with Run and Cancel, like any other waiting action in G4. The row shows what the Agent says it will do, with the force-push or main line first, and "Looks destructive:" in front of a force push. This part changed nothing in the phone app. A push or pull request made by the Agent's own Git tools (see the Agent group) reads in words on that card, with Push or Open PR in place of Run.

• Whether each push, pull request or merge goes out. Nothing sends code to GitHub unless you press Confirm & run on the PC or Run on the phone.
• There is nothing to set up. There is no switch for these asks.

Open the Agent with the round 🤖 button, then press 👥 in its top row. Its tooltip starts "Workers: build a team of bots a Manager drives". Pick a team.

1. The Manager asks first

   Type a goal and press Run team. You see "The Manager is checking what it needs before it starts..." If it needs nothing, the run starts at once. Otherwise a form says "Before we start, I need:" with the Manager's questions (a text box, a list to pick from, or a folder on this PC) and checks PetAi makes itself, each with a tick or a cross. Start the run stays greyed out until every question not marked "(optional)" is answered and every check passes. Fix what a check names, then press Check again. Cancel drops the form and nothing runs.

   Keys, tokens and passwords are never asked in the form. Such a question becomes a box to tick, and its text starts "PetAi never asks for keys, tokens or passwords here." Add the key in Settings first. A form left open too long says "That form is too old. Press Run team again."

2. Follow the run

   The button reads "Running..." and a box under the goal lists the steps as they happen, with "The team is working on it" and how many steps so far. You can close the panel or reload the page: the run carries on, and opening the team again shows its steps or its result. If PetAi restarts meanwhile, the box says "Can't reach PetAi right now. Still trying..."

3. Stop a run

   Press ■ Stop in that box. A step already under way finishes, and nothing is undone. The result says "Stopped, as you asked. What the team did before that is below."

4. Steps that wait for your OK

   With Let workers use the Agent's tools on, a worker's file change, and every change after a worker read a web page, waits for you. Its hint now says "Off: they reason and write. On: they can also search the web and read files." and "Trading stays off. Writes and other risky steps wait for your Confirm." The result says how many "actions are waiting for your OK", then shows a card per step with Confirm & run and Cancel. A step that waited too long while the team ran says it "waited too long for your OK while the team ran" and was not done.

5. Team chat and each bot's chat

   Each team has Team chat beside Delete team. It talks to the Manager, who remembers this chat and every run of the team, so a goal like the same as last week, but for March, works. Each bot has Chat beside Profile: it talks to that bot alone, with its role and what you taught it.

   Type in the box and press Send. While the answer is written you see "Writing the answer..." and ■ Stop. Answers are formatted, with Copy. Your own messages have Copy and Edit; Edit says "Editing your message: the answers after it are replaced when you send." and Cancel edit backs out. Clear chat deletes what was said, and Close hides the chat. When the bot's model changed, the answer has the "Switched from" line too.

6. Schedules ask first too

   In ⏰ Schedule, press Create (or Save on a changed schedule). If the task needs something, the same form shows with Save with these answers, and nothing is saved until it is complete. A schedule that still misses something is marked Needs your answer and does not run on a guess. Open it: it says "It does not run until you answer, so it never guesses.", or, when a run stopped to ask, "The last run stopped here:" and what it asked. Press Answer, fill in the form and press Save my answers. It runs again on its next turn. The answers a schedule runs with show under Your answers.

When the Agent hands a goal to a team in a chat, it asks you the Manager's questions in one message first.

Open the Workers tab while the phone is paired.

1. The same form

   Type a goal under Give the Manager a goal and tap Run the team. The phone shows "The Manager is checking what it needs before it starts..." and then, if needed, the same form as the PC: "Before we start, I need:", Start the run, Check again and Cancel.

2. Follow and stop the run

   A card called "Team run on the PC" lists the newest steps and has Stop. The run works on the PC, so it carries on when the phone loses Wi-Fi. Meanwhile the card says "Can't reach the PC right now. Still trying..." When the run ends, the card becomes the result and stays until you tap Clear. Each team has its own run, so two teams can run at once.

3. Steps that wait for your OK

   With Let workers use the Agent's tools on, the result shows the waiting steps as cards with Confirm and Cancel. Confirm runs the step on the PC. Cancel says "Cancelled. It was not done."

4. Team chat and bot chats

   Each team has Team chat beside Delete team, and each bot has Chat beside Edit. They are the same chats as on the PC. Each message has Copy, your own messages Edit, and the chat has Send, Clear chat and Close. If the PC runs an older PetAi, the chat says "This PC's PetAi has no team chats yet. Update PetAi on the PC."

5. Schedules with the PC's fields

   Scroll to Schedule. Under New schedule you now find what the PC's Schedule page has: Model:; Connectors with Every connected one or Only these; Behavior with Each run starts fresh or Each run carries on from the last one; and Notifications with After every run, Only when a run fails or Notifications off. A template says which connectors it works best with ("This template works best with:"). Tap Save schedule. A time more often than once an hour is refused with "A schedule runs at most once an hour."

6. Answer a schedule that waits

   A schedule that needs something says "Needs your answer." and has Answer. Fill in the form and tap Save my answers. The answers show under Your answers. Saving a schedule that needs answers says "It needs your answer before it runs: press Answer on it in Workers > Schedule."

7. Run now and the whole answer

   Run now is now followed until that run ends, however long it takes. Each of the last runs has Show the whole answer when the answer says more than its summary.

Schedules still run on the PC, so the PC must be on with PetAi running. Editing a saved schedule stays on the PC's Schedule page.

• The answers the Manager asks for. Nothing runs until you give them.
• Whether to switch on Let workers use the Agent's tools for a team. Its writes still wait for your Confirm.
• Each waiting step: Confirm & run on the PC, Confirm on the phone, or Cancel.

1. The switch

   In 👥 Workers, each team has a second switch under Let workers use the Agent's tools: Let workers act on my phone. It is off until you turn it on. Its hint says "Only for runs you start from the phone's Workers tab. Every phone action waits for your Confirm on the phone." It is the same setting as the switch on the phone.

2. Follow the phone steps

   In a run started from the phone, a phone step is a line with 📱 and no button. It says "Waiting for your OK on the phone", then "Done on the phone" or "Skipped on the phone". If nobody answers in three minutes it says "The phone did not answer, so this was not done." If PetAi is closed on the phone, it says "The phone is not following this run, so this was not done." The PC's own steps carry on.

3. Settings are only opened

   Android lets no app switch Wi-Fi, Bluetooth and the like for you. A goal such as turning on Bluetooth only opens that screen, and the result says "Not marked done: that switch is your tap on the phone".

A run started on the PC, a schedule, or a team the Agent starts from a chat never acts on the phone. A goal that needs the phone and is started on the PC is stopped by a check: "This goal needs your phone, and a run started on the PC never acts on the phone."

1. Turn it on for a team

   In the Workers tab, open a team. Under Let workers use the Agent's tools, turn on Let workers act on this phone. Its hint starts "Every phone action waits for your Confirm here, at most 4 per run." On Android 13 and later, the phone then asks whether PetAi may show notifications. If you say no, the hint adds "You will not get a note when a Worker needs the phone." Runs still work while PetAi is open.

2. Run the team

   Type a goal and tap Run the team. Keep PetAi open on the phone while it runs.

3. Confirm or Skip each action

   Each action a worker wants shows a card under the team. It names the worker and says what it wants to do on this phone, with Confirm and Skip. An app card adds "It brings the app to the front; it cannot tap or type inside it." A settings card says the setting is not switched, and you do that there. A WhatsApp card shows the person, the number and the whole message, and says "Nothing is sent: you tap Send in WhatsApp yourself." After the run read a web page, the card names where.

4. A late tap does nothing

   A card waits at most three minutes. After that, or after Stop, Confirm runs nothing and the card says "Too late: the team moved on, so this was not done." If the phone could not reach the PC, the card says "Couldn't reach the PC, so nothing was done. Press Confirm or Skip again."

5. When PetAi is behind another app

   A notification says "A Worker wants to use this phone. Open PetAi to see it." If PetAi is open on another tab, the Workers tab's icon turns into a red bell. On the Workers tab, a line says which team "wants to use this phone." with Show to open it.

Listing the apps and reading the battery need no card, and run only while PetAi is on the screen. A Skip counts towards the four actions, and a fifth is refused with no card. With PetAi swiped away or the phone asleep, no card can show, and phone steps end as not done. A phone paired the old way, with the shared pairing code, gets "This phone uses the old shared pairing code, so Workers cannot act on it." Pair it again with the QR code.

• Which teams may act on your phone. The switch is off for every team until you turn it on.
• Each action, with Confirm or Skip. Nothing on the phone happens without your tap, and WhatsApp messages are sent only when you tap Send in WhatsApp.
• Whether to allow PetAi's notifications, so you hear when a Worker needs the phone.

1. Find the tab

   With no PC paired, the bottom bar shows Assistant, Workers, My phone and Settings. Open Workers. It says "Teams kept on this phone, run on the phone's own model."

2. Build a team

   Type a name and tap Add team. Then tap Add a bot, give it a name, a role, a purpose and what to teach it, and tap Save bot. There is no model box: every bot uses the phone's one model.

3. Choose what the workers may use

   Let workers act on this phone works as with a PC. Its hint says "Closing PetAi stops the run." Let workers use GitHub shows only while GitHub is connected on this phone, and is off at first. With it on, they can read your GitHub repository and "propose a change as a draft pull request, which waits for your Confirm."

4. Run the team

   Type a goal under Give the Manager a goal and tap Run the team. The same "Before we start, I need:" form comes first, with Start the run, Check again and Cancel. With no model set up, it says "PetAi has no model of its own to run this team. Set one up, then press Run the team again." and opens the model setup in Settings instead.

5. Follow, confirm and stop

   A card called "Team run on this phone" shows the newest steps, "Keep PetAi open while the team runs." and Stop. Each change shows a card with Confirm and Skip, the same as with a PC. A card waits at most three minutes; then it says "Too late: the team moved on, so this was not done." When the run ends, the result stays until you tap Clear.

6. If PetAi was closed

   A run lives inside PetAi. If PetAi was closed while it ran, the run shows "Stopped: PetAi was closed." when you come back.

7. Team chat and bot chats

   Each team has Team chat, and each bot Chat. They are kept on this phone; the team chat says "Kept on this phone. The Manager remembers this chat and every run of the team." They have Copy, Stop, Clear chat and the "Switched from" line after a model change.

8. No schedules

   The Schedule section says "Schedules need the PC." Pair a PC to run a team on a timetable.

When you later pair a PC, the phone's own teams stay on the phone and come back when you use PetAi without the PC. They are not copied to the PC. Pairing a PC, or setting up another model, stops a run that is going on the phone. The teams are left out of Android's cloud backup.

• Which model the phone uses. Each run spends that model's credit: one plan, up to twelve steps and a check after each round.
• Whether a team may act on this phone or use GitHub. Both switches are off at first.
• Each change, with Confirm or Skip.

Nothing is on by default. Until you pick a device, PetAi behaves as before and uses the default one.

1. Open the devices list

   Click ⚙ Settings & appearance at the bottom of the Chats sidebar. Open the group Camera, microphone & speaker. You see three lists: Camera, Microphone and Speaker. Each starts at Default camera, Default microphone or Default speaker.

   If the lists show only generic names, click Show device names. Your browser asks once for permission to use the camera and microphone. Click Allow, and the real names appear.

2. Pick and test each one

   Choose a device in a list. The line under the buttons says Saved. Then test it:

   • Test camera shows the picture from the chosen camera. The button turns into Stop camera. Click it when you are done.
   • Test microphone shows a level bar. Say something. The bar moves, and after about eight seconds PetAi says The microphone hears you. or says it heard nothing and suggests another microphone.
   • Test speaker plays a short beep from the chosen speaker.

   Every test turns itself off again. Nothing keeps recording.

3. Talk with the mic button

   Open the PetAi Agent panel and click the mic button. With a microphone picked, PetAi records that microphone itself and turns your words into text on this PC. The browser's own speech listening always hears the system default, so PetAi does not use it when you have picked one. With no pick, nothing changes from before.

   Jarvis works the same way. If you change the microphone while Jarvis is listening, Jarvis restarts and listens to the new one from your next sentence. With a pick, the Jarvis line reads Listening on the microphone you picked in Settings.

4. Hear replies on your speaker

   With a speaker picked, the PC turns each part of a reply into a sound file using the voices Windows already has, and plays it on that speaker. No key and no download. Pause, Resume and Stop work as before.

   If that cannot work, the reply is read in the browser's voice on the default speaker, and PetAi tells you why in one plain line. For example: Windows has no voice for that language to send to the picked speaker, so the reply is read on the default speaker.

5. If a device is unplugged

   Say your headset is picked and you unplug it. The line under the lists says your saved device is not plugged in, and PetAi uses the default until it is back. Plug it in again and the pick works again. You do not have to choose twice.

These parts changed nothing in the phone app. They pick the devices of the PC, not of the phone.

• Which camera, microphone and speaker PetAi uses. Leave a list on its Default entry to keep the old behaviour.
• Whether to allow device names. Show device names is only a button; the browser asks you first.
• The choice is saved on this PC. It is not sent anywhere.

1. Switch it on

   Click ⚙ Settings & appearance at the bottom of the Chats sidebar and open Camera, microphone & speaker. Under Phone camera on this PC, tick Let my phone show its camera and microphone here. The line below says On. Waiting for the phone: press Show my camera on the PC in the PetAi app.

2. Open the live window

   Click Open the live window. This also switches the receiver on if it was off. A small window titled PetAi Live opens. It says Waiting for the phone… until the phone sends. PetAi opens it as a plain app window in Edge, Chrome or Brave, whichever the PC has, or in your normal browser if none is found.

3. Start it on the phone

   Do the phone steps below. When the picture arrives, the badge turns to LIVE and the line under the window says which phone it is from.

4. Sound

   The phone's microphone plays on the PC. Click Mute to silence it. The button then says Unmute. If your browser asks for a click before it plays sound, the window says Click once in this window to hear the phone.

5. Stop it

   Click Stop in the live window. It switches the receiver off and forgets the last picture and sound, then the window closes. The window then reads Stopped. Nothing is kept. It also switches itself off after about a quarter of an hour with no picture, and whenever PetAi restarts. You can also untick the box in Settings.

This needs the phone paired with the PC and the newest phone app.

1. Tap the button

   On the Home screen tap Show my camera on the PC. The phone asks for the camera and the microphone. Allow the camera. If you refuse the microphone, only the picture is shown, with the line The microphone is not allowed, so only the picture is shown. Allow it in the phone's Settings to send sound too.

2. It asks the PC first

   The phone shows Asking the PC… and then Starting the camera…. If the PC has not switched it on, nothing is sent and you see: The PC has not switched this on. On the PC, open Settings > Camera, microphone & speaker and tick "Let my phone show its camera and microphone here".

3. While it runs

   You see LIVE on the PC and a Stop button. The phone sends about four small pictures a second from the back camera. Keep PetAi open on the screen. If you leave it, Android may switch the camera off, and the phone says so.

4. Stop it

   Tap Stop. It also stops by itself when you click Stop on the PC, when the PC stops answering, or when Android takes the camera away.

• Whether it is on at all. It is off until you tick Let my phone show its camera and microphone here on the PC. A phone cannot switch it on itself.
• When it ends: Stop on either side, at any time.
• Whether the sound is heard: Mute on the PC, or refuse the microphone on the phone.

Everything starts from the Customize › Connectors › Local AI card.

1. Read the launcher line

   After the library check, the launcher window shows one of these lines:

   Launcher window

   Local AI: Ollama running (N models)
   
   Local AI: Ollama is installed but not running - start it from the Start menu. Not fatal.
   
   Local AI: Ollama not found - Customize > Local AI has an Install button. Not fatal.

   The launcher never installs Ollama by itself.

2. Open the Local AI card

   In the left menu, click Customize, then the Connectors tab. Scroll past Phone app, GitHub and Tailscale to the Local AI card. It says "Runs models on this PC. Nothing leaves the machine. No key." The small tag by its title first reads checking..., then not installed, installed, not running or running - N models.

3. Install Ollama

   If the tag says not installed, press Install Ollama. The note beside it says "with winget; Windows may ask for permission". Say yes if Windows asks. The line under the card shows "installing Ollama with winget (Windows may ask for permission)" and updates by itself. When it is done you see "✓ Ollama is installed. It starts with Windows; if it is not running yet, start it once from the Start menu." The Install button then goes away.

   No winget on this PC? Then the button is greyed out and the note reads "winget is not on this PC — download Ollama, install it, then click Refresh". Click download Ollama, install it, then come back to the card.

4. Start Ollama if it is not running

   If the tag says installed, not running, start Ollama once from the Windows Start menu. Then press Refresh on the card.

5. Download a model

   When the tag says running - 0 models, the list reads "No models yet — pull one above." The box ("model name, e.g. qwen2.5:3b") is already filled with a model that suits your graphics card: qwen2.5:3b with no graphics card or under 8 GB of graphics memory, qwen2.5:7b from 8 GB, qwen2.5:14b from 16 GB. The card calls this "a suggestion, not a measurement". You can type another name.

   Press Pull model. The line under the card follows the download, with a percentage and how many GB are done. At the end it says "✓ qwen2.5:3b is ready — pick ollama/qwen2.5:3b in any model menu", and the model is listed as ollama/qwen2.5:3b with its size. Only one download runs at a time.

6. Use it in Chats

   Go back with ‹ Chats at the top of Customize. Set Purpose to "Chat / general questions" or "Coding (any language)". Open the Model list and pick ollama/qwen2.5:3b. Typing ollama in the filter… box narrows the list. Press reload if it is not there yet. An embedding model such as ollama/nomic-embed-text shows only under the Purpose "Embeddings".

7. Use it in the Agent

   Click the round 🤖 button at the bottom-right of the Chats page (it is also on Design, AI Code, Customize and Screener) to open the PetAi Agent panel. In its Model list (the first entry is "Auto (best available, with fallback)"), local models sit in a group named ollama. Pick one. It is saved at once. After the Agent has tried a model, the list shows ✓ tools or ✗ no tools beside it.

8. Check everything at once

   Open Command Prompt in the PetAi folder and run:

   Command Prompt, in the PetAi folder

   python -m LOCAL_AI

   Copy

   It prints "PetAi Local AI doctor", then one line each for Python, the console, the gateway, Ollama (with every ollama/<name> and its size) and Git. If python is not found, try py -3 -m LOCAL_AI. Type nothing after LOCAL_AI: an extra word such as doctor only prints the help. python -m LOCAL_AI chat "hello" gets one answer from the first chat model.

Not the same thing: the ⚙ Settings & appearance menu also has a line "Local AI engine (free images & video on your PC)". That line is about pictures and video, not Ollama.

1. See whether it runs

   Open the PC tab. The top card shows the PC's name, then "Local AI: Ollama running, N models" or "Local AI: not running - start or install it in Customize > Local AI on the PC". The phone reads this when it connects. Reconnect the phone to see a change you made on the PC.

2. Install on the PC only

   The phone has no button to install Ollama or download a model. Use the Local AI card on the PC.

3. Run the Agent on a local model

   Open Settings and find Agent model. Tap the button that shows "Auto (best available)" and pick an ollama/<name> entry. The card says "Applies to the chat, the floating bot and this phone." The model still runs on the PC.

4. Check the PC can do it

   In Settings, under Connection, the Can do row lists local_ai when the PC supports it.

• Whether to install Ollama at all. It adds a tray app that starts with Windows, and each model is several gigabytes. PetAi never installs it by itself.
• Which model to pull. The suggested one fits your graphics card, and you can type another. Each needs gigabytes of disk space and time to download.
• Where to use it. Pick ollama/<name> yourself in the Chats Model list and in the Agent's Model list. The Agent's Auto uses a local model only when its turn comes.

1. Tools as text

   In the Agent's Model list, a model that writes its tools as text is marked tools (text mode). Nothing to set: PetAi switches by itself.

2. Fewer tools for small models

   In the Toolbox, under the tool groups, Pick tools per question (for small models) offers Auto (on for models with a context window of 16k or less), Always on and Off (send every tool that is on). Auto is the default. Groups you turned off are never sent.

3. Find other local servers

   Open Customize › Connectors › Local AI. Start LM Studio, llama.cpp or another server first, then press Find local servers. PetAi looks only on this PC, never on your network. Press Use beside a server to use its models. If none answers, it says "No local AI server answered on this PC."

4. Best models for this PC

   Under Best models for this PC, models are ranked by what fits your memory and graphics memory. It is a suggestion, not a measurement. Put in Pull box copies a name into the box above; nothing downloads until you press Pull model.

5. Smarter search

   In the Toolbox, under 📚 Knowledge, tick Smarter search (uses a local embed model). It says "Needs Ollama; no key." It then finds your memories and your knowledge base passages by what they mean, not only their words. If no small embed model is there yet, a button offers to download nomic-embed-text, about 270 MB. Without one, search matches your words as before.

6. Teach your local model

   When a model on this PC fails an answer and another model answers it, a chip under the answer reads "Teach my local model this". Click it to see a draft skill with Name, When to use it and Steps. "Nothing is saved until you press Save skill." Press Save skill to keep it, or Cancel.

1. See what the PC found

   In Settings, the card Local AI on the PC lists the servers and the best models, read only. Tap Check the PC to read them again. It says "To use a server or download a model, open Customize > Local AI on the PC."

2. Teach from the phone

   The same "Teach my local model this" chip shows under such an answer on the Assistant tab. Tap it, check the draft, and tap Save skill or Cancel.

3. Smarter search

   The card What the Agent remembers in Settings has the same Smarter search (uses a local embed model) switch.

• Whether to turn on smarter search, and whether to download the small embed model for it.
• Which lessons to keep. A draft skill is kept only when you press Save skill.

1. Install it once

   Open AI Code from the left menu. Set the engine list to Claude Code CLI (runs on your PetAi models). If it is not on the PC, press Install Claude Code (npm). Claude Code is Anthropic's software under Anthropic's terms, and you accept them yourself. It needs Node.js on the PC; if Node.js is missing or too old, the page says so, and starting PetAi again with launch_web_ai.bat installs it.

2. Run a task

   Pick a model and run the task as usual. The note under it says "It is not free Claude. Real Claude needs your own Anthropic account." It edits files and runs commands without asking, the same as running it yourself. In a project that uses git, G5's Undo this run still works. In a folder without git, a Claude Code run cannot be undone, so use git or keep a copy.

3. Use it in your own terminal or VS Code

   Open the Connect Claude Code card on the same page. Choose a Big model, a Normal model and a Small model, then press Connect. PetAi keeps a backup of Claude Code's settings first. Test sends one tiny request. Disconnect takes PetAi's settings out again.

• Free Claude. No tool gives it. Claude Code on PetAi runs your other models, not Claude.
• Claude Code on the phone. It runs on the PC only.
• Smart home control. PetAi cannot switch lights or other home devices.
• The phone reading or tapping inside other apps on the phone. That needs an Android accessibility service you would switch on yourself, and it is not built.
• Live camera video. The camera takes one picture at a time.

Browser tasks, voice and text chat with JARVIS, phone pairing and step-by-step automation were already in PetAi before G5.1.

• Whether to install Claude Code at all, and whether to Connect it to your own terminal.

1. Get the notebook

   Open Customize from the left menu and click the Connectors tab. Find the card Remote GPU node, just below Local AI. Click How to set up a free Kaggle node to open the steps, then click Download the notebook. You get a file named petai_kaggle_node.ipynb.

2. Put it on Kaggle

   On kaggle.com choose Create › Notebook › File › Import and pick the file. Kaggle wants a phone-verified account before it gives out GPUs and the Internet switch. In the notebook's settings choose Accelerator: GPU T4 x2 and turn Internet on. Then click Run All.

   The notebook installs Ollama, the image and video tools, starts a small guard that only answers requests with the right token, and opens a Cloudflare tunnel so your PC can reach it. This takes a long time the first time: it downloads the model qwen2.5:14b (change the line MODEL to pick another), Stable Diffusion XL for pictures and Wan 2.1 for video (about 29 GB). Set ENABLE_IMAGE or ENABLE_VIDEO to False in the notebook to skip one.

3. Make the node reconnect by itself (recommended for Kaggle)

   A Kaggle session is temporary and the next one has a new link. To stop pasting it, do this once. The full steps with examples are in Set up a GPU from scratch › Route 1. In short: (1) create a secret gist on gist.github.com and copy its id; (2) at github.com/settings/tokens choose Tokens (classic), not Fine-grained tokens (those cannot use gists), and tick only gist; (3) think up a token text of 16 or more characters; (4) in the notebook choose Add-ons › Secrets, add PETAI_TOKEN, PETAI_SYNC_GIST and PETAI_SYNC_GH_TOKEN, and tick Attached for each. Then click Run All. The notebook now keeps the same token every session and writes each new link into your gist.

4. Copy the link and the token

   The last cell prints a box like this:

   Example of what it prints

   Paste these into PetAi: Customize > Connectors > Local AI > Remote GPU node
   
     Link : https://something.trycloudflare.com
   
     Token: a-long-random-text

   Copy

   That link is a free Cloudflare quick tunnel (no account needed) the notebook opens for you. Leave that cell running. If you stop it, the node goes away. Without the secrets above, a new session prints a new link and a new token. With them, the token stays the same and the notebook says Link saved to your gist.

5. Paste them into PetAi

   Back on the Remote GPU node card, paste the link into the first box and the token into the second box. Open Reconnect by itself each session and paste the gist id and the GitHub token, then click Save. PetAi saves and checks the node at once. The line next to the buttons says saved - the node answers, or saved - but the node is not answering yet. The tag in the card's title changes from not set up to something like answering - 240 ms. After you save, the token boxes stay empty and say saved - paste a new one to replace it. That is on purpose.

   From now on, when the node stops answering, PetAi reads the gist about once a minute (and when a model menu opens). When a new session has written its link there, PetAi switches to it by itself. You only press Run All on Kaggle. PetAi cannot start a Kaggle session for you.

6. Look at the health

   The card shows Answering. and how many models it lists, or Not answering. with the reason and a note that models on this PC answer meanwhile. Click Check now to look again. PetAi also looks about once a minute, and the card refreshes itself every 15 seconds while it is open.

7. Run the benchmark

   Click Run benchmark. The box beside it, model to test (blank = the first), lets you name a model. It takes a few minutes. You get the speed in tokens a second, the longest prompt the node reads in full, the graphics memory it uses and whether eight requests in a row all worked. The last ten results are kept on your PC.

8. Watch the hours

   A free notebook has a weekly allowance. Type it in the box Hours a week your notebook allows (0 means no limit). The card then shows how many hours the node has answered in the last 7 days and warns when you are Close to your weekly allowance. This is PetAi's own count, not the provider's.

9. Use it

   Keep the box Let Auto use the node ticked. Auto then tries up to two node models first and up to two models on your PC right behind them. In the Agent's answer, the badge says Answered by your GPU node (a notebook or GPU box you connected), and the usage line shows GPU node next to Local, Cloud and Gateway. You can also pick a node/… model yourself from any model menu.

10. Make pictures and video on the node's GPU

    Open the Image or Video menu and pick node/sdxl-base (pictures) or node/wan2.1-t2v-1.3b (video). The GPU works on them only when you ask for one. With two T4s the language model keeps GPU 0 and pictures and video use GPU 1. With one GPU the language model steps aside while a picture or clip is made and loads again on your next chat. Start with a clip of 1 or 2 seconds: Wan on a T4 is slow, and a clip that needs too much memory fails with a CUDA out-of-memory error (make it smaller or shorter).

11. When the node goes away

    If the node stops, the next model answers, and the reply says The GPU node is not answering, so this was answered on this PC. (It says "by a provider's own API" or "through the gateway" if that is where it was answered instead.) Run the notebook again when you want it back. With the gist set up, PetAi finds the new session by itself within about a minute. Without it, paste the new link and token and click Save.

12. Remove it

    Click Forget node and confirm. The link and the token are removed. Untick Let Auto use the node instead if you want to keep the node saved but keep your chats on your PC.

Saving the node and the benchmark work from the PC only. The phone cannot do them. The phone's chat answers can still come from the node, because the PC does the asking, and the answer says where it came from.

• Whether to set up a node at all. Nothing is contacted until you save an address.
• Whether Auto may use it: the Let Auto use the node box. Models you pick yourself always work.
• When to forget it: Forget node.

1. Pin

   Hover a chat in the sidebar and click the pin icon (📍). A pinned chat shows 📌 next to its title and sorts to the very top of the list, above every chat that isn't pinned - including ones started more recently. Click the pin icon again to unpin it. This is saved with the project, so it is still pinned after you close and reopen PetAi.

2. Rename

   Click the pencil icon (✎) and type the new name. It replaces the auto-generated title everywhere the chat is shown, and is saved the same way pinning is.

3. Delete

   Click the trash icon (🗑). PetAi asks you to confirm before anything is removed - a single click never deletes a chat. Confirming removes the project's folder and files from disk. A chat that is still running can't be deleted until it finishes or you stop it, and AI Code's own developer-mode session (the one bound to PetAi's own source folder) can't be deleted from here at all.

1. See the chips

   Start a task in AI Code. If the agent needs a choice from you — for example, which backend framework, whether to add authentication, or which colour palette — its question appears in the log and a row of chips appears below it, one per option. The chips are styled to match PetAi's own theme and swap between light and dark automatically.

2. Pick one

   Click the chip that matches what you want. The choice is sent as your reply, the chips disappear, and the agent carries on building. You can still type a free-form answer in the task box at the bottom if none of the chips fit.

3. What triggers chips

   Chips appear when the agent's reply ends with a question followed by a list of short options (lines that start with a number, a letter, or a bullet). Questions without a list, or long free-form questions, show as plain text with no chips.

1. Make the account

   Sign up at kaggle.com. Then open Kaggle Settings and verify your phone number under Phone Verification. Without it Kaggle gives no GPU and no Internet switch.

2. Download PetAi's notebook

   In PetAi open Customize › Connectors › Remote GPU node › How to set up a free Kaggle node and click Download the notebook.

3. Import it

   On Kaggle choose Create › New Notebook › File › Import Notebook and pick petai_kaggle_node.ipynb.

4. Switch on the GPU and the Internet

   In the notebook's right-hand Session options set Accelerator to GPU T4 x2 and turn Internet on.

5. Create a secret gist (the notice board where the notebook leaves its link)

   Sign in at gist.github.com (a free GitHub account is enough). Fill the form like this:

   Example

   Gist description:   PetAi node link
   
   Filename:           petai_node.json
   
   Content:            {}

   Copy

   Click the small arrow beside the green button and choose Create secret gist. Do not choose Create public gist: a public gist can be seen by anyone. After it is created, look at the address in the browser. It looks like https://gist.github.com/your-name/8f3a9c0d1b2e4f5a6b7c8d9e0f1a2b3c. The long text at the end is the gist id. Copy only that part (PetAi also accepts the whole address).

6. Create a GitHub token (a password that can only touch gists)

   Open github.com/settings/tokens. You can also get there with your profile picture › Settings › Developer settings (last item on the left) › Personal access tokens. GitHub offers two kinds:

   Choice	Use it?	Why
   Tokens (classic)	Yes, choose this one	It has a gist permission and it works for gists.
   Fine-grained tokens	No	GitHub does not let fine-grained tokens read or write gists, so the notebook would get an error 404 or 403.

   Click Tokens (classic), then Generate new token › Generate new token (classic). GitHub may ask for your password. Fill the form like this:

   Example

   Note:         PetAi Kaggle node
   
   Expiration:   90 days   (or "No expiration" if you do not want to renew it)
   
   Select scopes: tick ONLY  [x] gist   - leave everything else empty

   Click Generate token at the bottom. The next page shows the token once, text like ghp_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX. Copy it now: GitHub never shows it again. If you lose it, delete it and generate a new one. If you set an expiry, the reconnect stops on that day and you make a new token then.

7. Think up the node token

   This is the password your PetAi and the notebook share. Any text of 16 or more characters works, for example blue-tiger-4821-river-lamp-93. Make up your own and do not use this example. Keep it somewhere safe, you paste it into PetAi once.

8. Add the three secrets in the Kaggle notebook

   With your notebook open choose Add-ons (top menu) › Secrets. For each row below click Add secret, type the Label exactly as shown (capital letters and underscores matter), paste the Value and click Save. Then tick the Attached box beside each of the three, so the notebook is allowed to read them.

   Label (exactly)	Value (example)
   PETAI_TOKEN	blue-tiger-4821-river-lamp-93 (your own text from the step above)
   PETAI_SYNC_GIST	8f3a9c0d1b2e4f5a6b7c8d9e0f1a2b3c (your gist id)
   PETAI_SYNC_GH_TOKEN	ghp_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX (your classic token)

   If a secret is missing the notebook does not stop: it makes a new random token each session and does not write the gist, so you paste the link and token by hand.

9. Run it

   Click Run All. The first run takes a long time: it installs Ollama and the image and video tools, downloads the language model (qwen2.5:14b), Stable Diffusion XL and Wan 2.1 (about 29 GB for video), starts a token guard and opens a Cloudflare tunnel. Models are downloaded again in each new session.

10. Copy the link and the token

    The last cell prints Link and Token, and Link saved to your gist. Paste the link and token into PetAi's Remote GPU node card, open Reconnect by itself each session, paste the gist id and the GitHub token, and press Save. Leave the notebook running. In every later session just press Run All: PetAi finds the new link in your gist within about a minute.

11. Try a picture and a clip

    Pick node/sdxl-base in the Image menu, then node/wan2.1-t2v-1.3b in the Video menu (start with 1 or 2 seconds). If the notebook shows a CUDA out-of-memory error, make the size smaller or set ENABLE_VIDEO = False.

1. Open Colab

   Go to colab.research.google.com and sign in with your Google account.

2. Upload the notebook

   Download petai_kaggle_node.ipynb from the PetAi card as in Route 1, then in Colab choose File › Upload notebook.

3. Pick a GPU

   Choose Runtime › Change runtime type and select a GPU (T4 on the free tier).

4. Run all

   Choose Runtime › Run all, wait for the last cell, and paste the printed Link and Token into PetAi as in Route 1. Keep the tab open: Colab stops an idle or closed session.

1. Rent a machine

   Make an account on RunPod or Vast.ai (or any GPU host you already use). Rent one NVIDIA GPU. About 16 GB of graphics memory runs a 7B model; 24 GB or more suits bigger ones. Choose a template with CUDA and PyTorch, and open one HTTP port (8000) if the site offers it.

2. A token for gated models (only if needed)

   Open-licence models such as Qwen need no token. Models that ask you to accept a licence need a Hugging Face access token (a free Read token) and a click on Agree on that model's page. Then on the machine run export HF_TOKEN=your-token.

3. Install and start vLLM

   Open the machine's terminal and run:

   On the rented machine

   pip install vllm
   
   python -m vllm.entrypoints.openai.api_server --model Qwen/Qwen2.5-7B-Instruct --host 0.0.0.0 --port 8000 --api-key CHOOSE-A-LONG-SECRET

   Copy

   Replace CHOOSE-A-LONG-SECRET with a long text only you know. That text is your token. The first start downloads the model and takes a few minutes. Details are in the vLLM documentation.

4. Get an address

   Either use the host's own link for port 8000 (RunPod shows a proxy link such as https://…-8000.proxy.runpod.net), or run cloudflared on the machine: cloudflared tunnel --url http://localhost:8000 prints a https://….trycloudflare.com link.

5. Paste into PetAi

   In the Remote GPU node card paste the address (ending it in …/v1 is fine but not needed) and your secret as the token, then Save.

1. Install Ollama

   Download it from ollama.com/download (Windows, Mac, Linux). On a Linux server: curl -fsSL https://ollama.com/install.sh | sh.

2. Get a model

   Run ollama pull qwen2.5:7b. The model list is at ollama.com/library.

3. Let it be reached

   Ollama answers only on its own machine by default. For a link that works from your PC, run cloudflared: cloudflared tunnel --url http://localhost:11434. Or join the machine and your PC to one Tailscale network and use its 100.x address on port 11434.

   Ollama has no password. Anyone who finds a plain tunnel link can use your GPU. Put a token guard in front of it, or use Tailscale, which only your own devices can reach.
4. Paste into PetAi

   Paste the address into the Remote GPU node card (leave the token empty if Ollama has none) and press Save.

1. Check it answers

   After Save, the tag in the card's title says something like answering - 240 ms. If it says not answering, the line below says why: wrong link, the notebook stopped, or the token is wrong.

2. Run the benchmark

   Click Run benchmark to see the speed, the longest prompt it reads in full and the memory it uses.

3. Use it

   Keep Let Auto use the node ticked, or pick a node/… model yourself in any model menu. Under each answer the badge says it came from your GPU node.

4. Keep it healthy

   Free notebooks stop after hours: when that happens PetAi answers on this PC. When you run the notebook again, PetAi finds the new link in your gist by itself (Kaggle route with the secrets set); otherwise you paste the new link and token. For a rented machine, stop or delete it on the host's site when you are done, or it keeps billing.

1. See the search engine

   Open the Agent, press 🧰, and scroll to Search engine. It reads "Search engine: DuckDuckGo (no key)". Type something in the box ("Try a search, e.g. weather in Delhi") and press Try it to see results. Brave and Tavily are optional: each is used only after you save a free key of your own for it.

2. Start a research

   In the Agent panel's top row, press 🔎. Its tooltip starts "Research: PetAi searches the web in rounds". Type the question in "What should PetAi research?", pick Quick: 2 rounds, Standard: 4 rounds or Deep: 6 rounds, and press Start research. You can also type /research and a topic in the chat box.

3. Watch it work

   A bar shows Researching with the round and the pages read so far. It takes a few minutes and keeps going if you reload the page. Stop ends it early; the report is then made from the notes gathered so far. The report comes with numbered sources and a link, "Open this report in Research".

4. The Research page

   Click Research in the left menu, under Schedule. Every report is listed, with Search reports at the top. You can start a new research here too. Open one to read it with its table of contents and sources. All reports goes back to the list.

5. Discuss, Continue or Delete

   Discuss opens the Agent with this report, so you can ask about it. The Agent asks before any change. Continue searches further from the report's notes and writes a new report, with 2, 4 or 6 more rounds. Delete removes it.

6. Research on a schedule

   In ⏰ Schedule, open the Templates tab and pick Research. Fill in Topic and pick a Depth. Each run saves a report, and its result ends with "Saved in Research."

1. Read the reports

   On the Assistant tab, tap Research above the message box. The reports saved on the PC are listed with Search reports. Tap one to read it with its Sources.

2. Discuss a report

   Tap Discuss. A chat about that report starts on the Assistant tab, answered by the PC.

3. Start one

   Type /research and a topic in the message box. The phone shows how far the research has got.

• Nothing to switch on. Search and research need no key.
• Whether to add a Brave or Tavily key for more results. Both are optional.
• How deep a research goes. More rounds read more pages and take longer.

1. Say where reminders go

   Open the Agent, press 🧰, and find Notes, todos & reminders (on by default). Press Where reminders go under it. The Alerts window opens on the console's main page. The easiest choice with no key is ntfy topic (no key): tick enabled, press New topic, then "Install the free ntfy app on your phone, tap +, and subscribe to this same topic." Or use Discord webhook (no key), or the Telegram and WhatsApp rows you may already use. Keep note reminders ticked under What to send, then press Save channels. Send test message checks it.

2. Ask for a reminder
   PetAi Agent

   Remind me to call mum at 5pm.

   Copy

   It understands words like "tomorrow 9am", "in 20 minutes" and "every weekday at 9". A time that has passed, or words it cannot read, give a plain error, never a guess. The answer has a link, Open Notes.

3. The Notes page

   Click Notes in the left menu, under Research. Across the top are Notes, Reminders, Done and Archive, with Search notes. Pinned notes come first. Type in "Take a note..." and press Enter for a quick note, or press + Note or + Checklist.

4. Give a note a time

   In the editor, under When, type words such as "tomorrow 9am", pick a date and time, or press In 1 hour, Tonight, Tomorrow 9 AM, Monday 9 AM or No time. PetAi says the time back before you save. Repeat offers Every day, Every weekday, Every week, Every month and Every year. Keep Send me an alert at that time ticked, tick Pin to the top if you like, then press Save.

5. Look after your notes

   Each note has a 📌 pin, a colour (🎨), Archive (or Restore) and Delete. A checklist also has tick boxes and ✓ Done. Drag a note to change the order.

6. A quick note with no model

   In the Agent's chat box, type /notes and a few words, such as /notes buy milk. It is saved at once. /notes alone says how many notes you have and when the next reminder is.

1. The Notes tab

   While the phone is paired, the bottom bar has a Notes tab. It shows the PC's notes with Notes, Reminders, Done and Archive, and Search notes.

2. Add or change a note

   Tap + Note or + Checklist. Under When, type a time in words and tap Check. Keep Send me an alert at that time on and tap Save. Each note has Pin, Archive (or Restore) and Delete. A checklist also has Done. Move up and Move down show when there is another note to move past.

3. By voice or chat

   On the Assistant tab, with Actions run on set to PC, say "remind me to call mum at 5pm", or type /notes and a few words.

Reminders are sent by the PC, so PetAi must be running on the PC at that time. A reminder missed while it was off is sent once, late, when PetAi starts again.

• Where reminders go. ntfy and Discord need no key. Anyone who knows an ntfy topic name can read it, so keep the long random one.
• Whether a note gets a time, and whether it repeats.

1. Open Documents

   Click Documents in the left menu, under Notes. Type a title in "New document title" and press + New, or ask the Agent, for example "draft a plan for the Goa trip as a document". The answer then has a link, Open Documents.

2. Four views

   A document has Review, Edit, Changes and History. In Edit, write and press Save (or Ctrl+S); each save is a new version. Discard changes goes back to the saved text.

3. Accept or dismiss a suggestion

   In Review, a suggestion shows in the text: words taken out are struck through, words put in are underlined. Press Accept to take it or Dismiss to drop it.

4. Go back to an older version

   In History, press View to read one, or Restore to make it the newest again.

5. Export

   Pick Word (.docx), PDF (.pdf), PowerPoint (.pptx), Excel (.xlsx), Markdown (.md) or Text (.txt) and press Export.

1. Attach PowerPoint, Excel and EPUB

   The 📎 button now also takes .pptx, .xlsx and .epub files, as well as PDF and Word. A file too long for one chat is kept whole in a knowledge base named Attachments, and the answer says so. A book locked against copying cannot be read.

2. Chat folders

   In the Chats sidebar, under Search chats…, choose Chats, Starred or Archive. + Folder makes a folder; drag a chat onto it. Each chat's ⋯ menu has Star, Move to folder… and Archive. A folder's ⋯ menu has Rename folder and Delete folder (keeps its chats).

3. Several chats at once

   Press Select, tap the chats, then Move, Star, Archive or Delete. Done ends it.

4. Tidy and Sort with AI

   Tidy lists empty and throwaway chats and removes only those, after you press its remove button. It never touches a pinned, starred, archived or foldered chat, or one used in the last hour. Sort with AI asks the chat's model to suggest folders. You can untick or rename one, and nothing moves until you press Apply.

5. Blind Compare

   Click Compare in the left menu. Type a question in "What should the models answer?", choose 2 to 4 models (+ Add a model), and press Compare. The answers show side by side with no names. Vote with the "is best" button under the best one. With two answers you can also press Tie or Both bad; with three or four, All equally good or All bad. Then the names show and the Scoreboard counts it. Reveal without voting shows the names without a vote. ↻ Reroll asks one model again. Past comparisons keeps the rounds.

1. Review documents

   On the Notes tab, tap Docs. The documents are listed with Search titles. Open one to see its suggestions marked in the text, each with Accept and Dismiss.

2. Attach a file

   On the Assistant tab, tap the paperclip. It takes the same kinds of file, up to 20 MB, and sends it with your question to the PC.

3. Compare

   On the Assistant tab, tap Compare above the message box. The steps are the same, with the answers one under the other.

The phone has no list of the PC's main chats, so chat folders are on the PC only.

• Which suggestions go into your documents. The Agent's edits to a document you wrote are only suggestions.
• Which chats Tidy removes, and which folders Sort with AI makes. Nothing happens before you press the button.

1. Scroll the links

   Point at the links and use the mouse wheel. Only the links move; the chats and the page stay where they are.

2. Change the height

   Under the links there is a thin line with a small bar. Drag it up to show fewer links and more chats, or down to show more links. With the keyboard, click the bar and press the up or down arrow. Double-click the bar (or press Escape on it) to go back to the usual height. PetAi remembers your height in this browser.

The phone app has its own tabs, not this menu.

• How tall the links box is. Nothing else changes.

1. Card Workflow Character

   Fill in a name, country, level, goal (now called your task) and a set of look options; the right card shows a live member card (with your own brand name, defaulting to Profit Hunter, and the label MEMBER) and a live image prompt below it. Copy the prompt, upload a clear front-facing photo of yourself to ChatGPT or Gemini, and paste the generated prompt.

2. Character Creator

   Build a full character - niche, archetype, personality, backstory, voice and visual identity - branded PROFIT HUNTER GURU. The right card assembles a live "Character Bible". Download .txt saves it; Open in Full Studio carries it straight into step 3.

3. Full Studio

   Paste or import a Character Bible (automatically carried over from Character Creator) and generate a live, tabbed production pack: Character Sheet, Angle Pack, Scene Prompts, Video Prompts, and Hooks & Script. Use the still prompts with ChatGPT or Gemini, and the Video Prompts with Google Flow or Higgsfield.

1. Open the Image editor

   In the console, press Editors in the left menu (below Compare). In the pop-up press Image. You can also open /editors and press the Image tab.

2. Choose a picture

   Press Choose a picture, or drop a picture on the dashed box. PNG, JPG, WebP, GIF, BMP and TIFF work, up to 30 MB. A GIF uses its first picture. PetAi keeps a copy in its own folder for two days so you can keep editing, and your own file is not touched.

3. Crop

   Open Crop on the right. Drag on the picture to choose the part to keep, or press Square, 4:3, 3:4, 16:9 or 9:16 for a centred shape. Then press Crop to the selection.

4. Turn it, make it smaller, change the colours

   Rotate and flip has Left, Right, Flip across and Flip down, and any angle. Resize takes a width and a height (keep the shape fills in the other one), or Half size. In Light and colour move the sliders: you see the result at once, and it only counts when you press Keep these changes (or Set back to drop it).

5. Try a look, add words or a border

   Looks has Black and white, Old photo, Opposite colours, Soft blur, Sharpen, Edges, Relief, Smooth and Outline. In Text type the words, then move Size, Across and Down and pick a colour; press Put the text on. Border adds a frame of the width and colour you pick.

6. Go back a step

   Under the picture, Steps lists what you did in order. Undo removes the last step, and Start over returns to your original.

7. Save the new picture

   Open Save, pick PNG (sharp, any size), JPG (smaller photos) or WebP, and press Save the new picture. A link Download ...-edited.jpg appears; the file is also kept in PetAi's media folder. The new file has no camera or location data. A see-through picture saved as JPG gets a white background.

Open Editors > Image, choose holiday.jpg, press Square, press Black and white, type "Goa 2026" in Text, press Put the text on, then Save the new picture as JPG.

1. Open Editors

   Connect to your PC. On the Notes tab press ✂ Edit at the top.

2. Choose a file

   Press Choose a file and pick a picture, PDF, sound or video from the phone (up to 60 MB). It is sent to your PC, which does the editing; your original stays on the phone as it is.

3. Add steps and save

   Press the buttons for what you want (the Steps line lists them, Undo removes the last), then press a save button. The new file is kept in Downloads/PetAi on the phone and opens in the app that shows it.

The phone gives the main jobs for each kind. The full tools (crop, text on a picture, watermark, password, joining files, drag-selecting on the wave) are on the PC's Editors page. This screen was written but has not been compiled or tried on a phone yet.

• Which picture to open, and whether to keep each change: nothing is saved until you press Save the new picture.
• There is nothing to switch on and no key to enter. It needs Pillow, which launch_web_ai.bat already installs.

1. Make the canvas any size

   Open Canvas size and press Drag the sides. Small squares appear on the corners and sides of the picture. Drag one out to add room, in to trim; the line under the buttons shows the new size. Press Apply this canvas. The new room is see-through, or a colour if you set New room to a colour. You can also type a width and height and say where the picture stays (middle, a corner, a side), then press Set this size. Resize still stretches the picture itself.

2. Make a collage

   Open Collage of pictures, press Choose pictures and pick two or more (press it again to add more, or + the one I am editing). Pick the columns, whether each picture fills its box or shows all of it, the box size, gap and background colour, then press Make the collage. The collage opens in the editor so you can keep working on it and save it.

3. Remove or change a background

   Open Background: remove or change. Choose what the background becomes (see-through, a colour, or another picture) and press Show it. How alike the colours decides how much counts as background; Soft edge smooths the cut. If a part is missed, tick I will click the background and click each missed part on the picture. keep the background instead cuts the subject out. Press Keep this. It works best on an even background such as a wall, a sky or a studio sheet; it does not understand hair or fur on a busy background.

4. Layers

   Open Add a picture on top (layer), press Choose a picture, then drag on the picture to move it, and use Size, Turn, Strength and Round corners. Press Keep this layer. There is no small limit: add as many layers as you need (up to 400 steps in one go). Under the picture, the Steps list shows every step as a small tag: press ✕ to take any step out (even an old one) or ◀ ▶ to change the order.

5. Levels, warmth, hue and more

   Levels, warmth and hue has darkest and brightest points, middle tones, warm/cool, green/pink, colour shift and dark edges (a vignette), plus Auto levels. Looks now also has Auto contrast, Even out light, Poster, Solarize, Remove noise and More detail.

6. Effects, shapes, shadow, corners, and only one part

   Open Effects, shapes and corners: Blocks (pixelate, good for hiding a face or a number plate), Round corners, Drop shadow and Shape (rectangle, oval or line, filled or not). Tick only inside the box I drag on the picture, drag a box on the picture, and a look, light and colour change, or blocks apply to that part only.

Editors > Image > choose a product photo, open Background, pick "fill it with a colour", press Show it and Keep this. Then Collage: choose three photos, 3 columns, Make the collage.

Not on the phone screen yet: it keeps the simple main jobs. Use the PC's Editors page, or ask the Agent in the phone app's Assistant tab for a simple change. Not compiled or tried on a phone.

• Which steps to keep, and in what order: nothing is saved until you press Save the new picture.
• Nothing leaves this PC; there is no key and no model download. These tools need only Pillow and numpy, which launch_web_ai.bat installs.

1. Add your files

   Press Add files (or drop videos, pictures and songs on the page). They appear under Your files. Drag one onto a lane of the timeline, or press Add to put it at the line. Sound only uses just the sound of a video.

2. Layers

   Layer 1 is the bottom picture lane; a higher layer is drawn over a lower one. Press + Layer on the bar for another layer, and + Sound lane for another sound lane: there is no small limit (up to 60 of each, and 200 clips). Dropping or adding a clip where its lane is busy opens a new lane for it. Tidy lanes removes empty lanes at the end. The timeline scrolls up and down by itself when you have many.

3. Transitions

   Click a clip, open Transitions on the right and pick how it comes in and goes out: Fade, Slide from any side, Zoom in or out, or Wipe from any side, with how many seconds it takes. Put the red line near the start or end of the clip to see it. Use these on every clip copies the same pair to all video, picture and text clips. Saving gives the same effect.

4. Mute the video's sound and put in a new one

   Click a video, open Sound on the right. Mute this clip silences it. Replace its sound... mutes the video and puts the song or voice you choose in its place, at the start of the clip. Detach the sound turns the video's own sound into a separate clip on a Sound lane (so you can move, fade or lower it) and mutes the video. Add sound on the bar puts any song or voice on a Sound lane at the red line.

5. Move, trim, split

   Drag a clip left or right to move it, or up and down to change layer. Drag its ends to trim. Click the ruler to place the red line, then press Split at the line (key S). Copy, Delete, Undo and Redo are on the bar. Space plays; arrow keys step.

6. Place it on the picture

   Click a clip on the picture: drag it to move, drag a corner to resize, drag the round dot to turn it. The right side has exact numbers, strength, mirror, fade in and out, speed, volume, colours, blur, black and white and old photo.

7. Text

   Press Add text, then type the words, pick the size, colour, a box behind them and how they line up.

8. Green screen

   Under Green screen pick the colour to remove from a video or picture. It is not shown while you edit: press Check the real result to see it.

9. Save

   Pick a name and a size, then Save the video. A bar shows the progress and Stop cancels it. Check the real result makes a small quick copy first.

Editors > Video studio > Add files: a clip and a song. Drag the song to Sound 1. Add text "Welcome", move it to the bottom of the picture, fade it in for 1 second. Save the video at HD 720.

Not on the phone yet. The page opens in a phone browser and works by touch, but the Android app has no studio screen.

• Nothing is saved as a video until you press Save the video. Your project is remembered in this browser, so you can come back to it.
• Nothing leaves this PC, and there is no key. It uses the ffmpeg that launch_web_ai.bat installs.

1. Get the phone listening

   The same as for My phone: the PetAi app is installed and paired, and in the app Settings the switch Let my PC control this phone is on. Then open Left menu > My phone; the badge says Listening now.

2. Start a tile

   Under Watch live press Start on Front camera, Back camera, Screen or Microphone. PetAi switches its live receiver on and asks the phone. The badge goes from Starting to LIVE when pictures arrive. For the screen, tap Start now on the phone when it asks. You can run several tiles at once.

3. Look, listen, keep a picture

   The picture updates by itself. Save a picture downloads the one on screen as a JPEG. mute here silences the microphone on the PC without stopping it; the bar shows the sound level.

4. Stop

   Press Stop on a tile. The phone stops sending and the PC forgets the last picture. When no tile is running the receiver switches off; closing the page does the same, and so does 15 minutes without a picture.

A notification shows the whole time something is sent. Android asks you for the camera and microphone the first time, and for Start now every time for the screen. Needs the new phone app (it is built from this version of PetAi); an older app is told apart and the tile says so.

• Only you, on this PC, start it, and only for a phone you paired and switched on yourself. Nothing is recorded on the PC or the phone, and it is not covert: the phone always shows its notification.
• Many phones cannot open both cameras at the same moment. If one tile stays on Waiting for pictures, stop the other camera first.

1. Pair the phone first

   Customize > Connectors has the QR for the phone app. A phone that is paired shows in the Phone list on the My phone page.

2. Open My phone

   Left menu > My phone. The badge says Listening now only while that phone has the switch on (see the phone part). If it says Not listening, the page shows what to do.

3. Press what you want

   Each card asks once. The answer appears under the cards: a picture, a recording you can play, or the numbers. Everything asked is listed under What was asked.

4. What it cannot do

   Unlock is not possible: Android lets no app unlock a phone from outside. Use Light up the screen, then unlock it yourself. Saved Wi-Fi passwords are not readable either (Android gives them to no app without root); the page shows the Wi-Fi name, and on the phone Settings > Wi-Fi > the network > Share gives the password as a QR.

Left menu > My phone > pick "Pixel 8 (listening)" > Where is it. You get the coordinates and "Open on a map".

1. Switch it on

   In the PetAi app open Settings > PC & extras and switch on Let my PC control this phone. A notification stays while it is on, and the phone shows each request. Switch it off and the phone stops listening at once.

2. Allow what you want

   The camera, microphone and location use Android's own permission prompts. Lock needs PetAi switched on once as a device administrator. See the screen makes you tap Start now on the phone each time: Android requires it.

• Nothing happens unless the phone's switch is on. A phone you revoke in Customize stops answering at once.
• Only the PC's own console page can ask; a phone cannot send requests, and a chat-only phone is never controlled.
• The phone reaches the PC over your Wi-Fi or Tailscale. Nothing goes through any other server.

1. The tracks

   You start with three: Voice-over, Music and Effects. Pick a kind and press Add track for more. Rename a track by typing in its name. M mutes it, S plays only that track (solo), the cross removes it. The Sound preset list gives a track a sound: clear voice, clean up noise, warm, bright, bass boost, telephone, old radio, muffled, echo, big hall.

2. The sound library

   On the left: loops in Western pop, rock, hip hop, electronic, lo-fi, cinematic, acoustic; Punjabi (bhangra, Punjabi pop, giddha); dhol beats (slow, fast, roll); Indian folk (garba and dandiya); Indian classical (tabla teental and dadra, tanpura drone, sitar); devotional (bhajan) and Bollywood dance. Effects: drum hits, dhol hit, tabla tihai, risers, whooshes, impact, record scratch, rain, wind, vinyl crackle, heartbeat, applause, chimes, notification ding, coin, camera click, laser, siren. Use the search box or the style buttons. Press ▶ to hear one, then Add (it goes on the chosen track at the red line) or drag it onto a track. Loops repeat to fill their length: drag the end of the clip to make it longer.

3. Your own sounds

   Add my sound file puts a song or a recording on the chosen track. Record voice records from the microphone onto the Voice-over track starting at the red line; keep play the mix while I record ticked (use headphones) to sing or speak along with the music. Press it again to stop; the recording appears as a clip.

4. Move, trim, layer

   Drag a clip left or right to move it, or onto another track. Drag its ends to trim. Click the ruler to place the red line, then Split at the line (key S). Copy, Delete, Undo and Redo are on the bar. Space plays and pauses.

5. The chosen clip

   On the right: its volume, fade in and out, speed and pitch (faster is higher, like a tape), Repeat to fill the length and Play backwards.

6. Hear and save

   While you edit you hear the volume, pan, mute, solo, fades and speed at once. The track presets and bass and treble are heard in the saved file: press Hear it with the effects for a quick check copy, then pick a name and a format (MP3, WAV, M4A, FLAC or OGG) and press Save the mix. Tick even out the loudness when saving for a steady volume.

Editors > Music studio > Sound library > Punjabi > Bhangra classic > Add (Music track). Press Record voice and sing along. Drag Whoosh from Effects onto the Effects track at the start. Voice-over track: preset Clear voice. Hear it with the effects > Save the mix as MP3.

Not on the phone yet. The page opens in a phone browser and works by touch, but the Android app has no music studio screen.

• Nothing is saved as a file until you press Save the mix. The project is remembered in this browser; your sound files stay on this PC for a couple of days, so add them again if you come back much later.
• The library sounds are made by PetAi's own code (no download, no key, no licence): you can use them anywhere. They are simple, clean loops and effects, not studio recordings, so use them as a start for your own music.

1. Open the PDF editor

   Press Editors in the left menu, then PDF in the pop-up. Or open /editors and press the PDF tab.

2. Choose a PDF

   Press Choose a PDF, or drop a file on the dashed box. If the PDF has a password, type it in the box Its password, if it has one first, then choose the file. PetAi keeps an unlocked copy in its own folder for two days; your file is not touched.

3. Pick pages, then change them

   Click a page to choose it (click again to un-choose; Select all and Select none help). With pages chosen: Turn left or Turn right turns them, Earlier and Later move them, Delete removes them. You can also drag a page with the mouse and drop it anywhere in the order. Undo takes back the last change and Start over goes back to the beginning.

4. Join PDFs

   Press + Add a PDF and choose another file. Its pages are added at the end; move them where you want.

5. Add page numbers or a watermark

   Open Page numbers, tick add page numbers, and change the wording if you like ({n} is the page, {total} the number of pages) and the place. Open Watermark and type the words, for example DRAFT; Strength makes it lighter or darker. Use ordinary letters and numbers: some other alphabets, such as Hindi, do not draw in this part.

6. Title, password, smaller file

   About this file sets the title and author. In Password and size type a password (at least 4 characters) to lock the new file, and tick make the file smaller where it can to squeeze it without losing quality.

7. Save the new PDF

   Press Save the new PDF in Save. A link Download ...-edited.pdf appears; the file is also kept in PetAi's media folder.

8. Split, read the text, or make a PDF from pictures

   Split the first file: say how many pages each part has, or groups such as 1-3, 4-6, and press Split into a .zip. Get the text: press Read the text of the first file; a scan has no text to read, and PetAi tells you so. Make a PDF from pictures (under the drop box): press Choose pictures; each picture becomes a page. Extract pictures: choose which pages first (none chosen means every page of the first file), then press Get every picture on the selected pages for a .zip of them.

Open Editors > PDF, choose invoice.pdf, click page 3, press Delete, tick add page numbers, type DRAFT in Watermark, then Save the new PDF.

On the Notes tab press ✂ Edit, choose a PDF, then Get the text (shown on the screen), Add page numbers and Save a copy, or type the pages in each part and press Split into parts (a zip). The same screen takes pictures, sounds and videos. The other PDF tools are on the PC's Editors page. Not compiled or tried on a phone yet.

• Nothing is saved until you press Save the new PDF, Split into a .zip or the picture button.
• No key and nothing to switch on. The page pictures need a small library, pypdfium2, that launch_web_ai.bat installs on the next start; without it the pages show as numbered boxes and everything else still works.

1. Open it

   Open the Agent (the round bot) and press the red 🔴 button in its top bar. A small, separate Record screen window opens next to the console - it is its own window on purpose: clicking Customize, Phone, WhatsApp, Editors or any other button in the console now navigates the console's page only, and the recording keeps going in its own window regardless. Before this, the recorder lived inside the console page itself, so any one of those clicks (a normal link, not a same-page tab) unloaded that page and silently killed the recording with it; that no longer happens.

2. Choose what to record

   Press Choose a screen or window.... Your browser then lists every screen you have (Screen 1, Screen 2...), plus single windows and tabs, each with a picture. Pick one and a live preview shows under the button with its size. Start recording stays off until you have chosen. To record the PC's own sound, tick Share system audio in that list (Chrome or Edge on Windows).

3. Mix the sound

   Under Sound there is a row for PC sound and one for Microphone, each with a level meter, a volume slider and a tick to switch it off. Pick the microphone you want from the list under them. Reduce background noise on the microphone is on by default; untick it if your microphone already does its own noise cancelling and the two fight each other.

4. Add your webcam

   Under Webcam tick Show my webcam over the recording and pick a camera. A round box appears over the preview - drag it to move, drag its edge or use the Size slider to resize. Wherever you leave it, and however big you make it, is exactly where it ends up in the saved video, whichever screen or window you share, because PetAi draws it into the recording itself rather than just floating it on the page. Filter gives the webcam a look (black & white, warm, cool, soft); Blur what is behind me asks the camera to blur your background - it works when this PC's camera and browser support it, and says plainly when they do not (the filter still works either way).

5. Quality and options

   Open Quality and options: small, standard or best quality; 15, 30 or 60 frames a second; a 3, 5 or 10 second countdown; Also make an MP4 copy; Hide this panel while recording; and a small red timer.

6. Record

   Press Start recording. The panel hides and a small red timer opens in its own little window (drag it anywhere on your desktop): ⏸ pauses, ⏹ stops and saves. Being its own window, not part of the page, it is never in the saved video even if you share your whole screen - tick off Show a small red timer only if you would rather not see it at all. You can also stop with the browser's own Stop sharing bar.

7. Your recordings

   The panel's Your recordings list has Play, Save, MP4 (makes the MP4 copy) and Delete. Files are in PetAi's media folder, named recording-date-time. Open one in Editors > Video studio by adding it as a file to cut it, add text or music.

8. Keep working while it records

   While a recording is running, every page of the console - whether the floating desktop is on or off - shows a small dark badge in the corner with the running time and its own Stop button, so you can switch to Customize, Phone, WhatsApp, Editors or any other feature and the recording just keeps going; Stop there ends it exactly like the Stop in the recorder window itself. The one thing that still ends it: closing the small recorder window directly (its X, not the console) stops the recording, the same way closing the browser tab always has - no page in a browser can keep a screen-share stream alive once the window holding it is gone. If that happens, whatever was captured up to that point is still saved; it is simply treated as a stop rather than silently carrying on.

Agent > red button > Choose a screen or window... > "Screen 2" > keep PC sound and Microphone ticked > Start recording. Switch to Customize or Editors - the corner badge keeps counting. Talk, then press Stop from the badge or the recorder window. Open the list and press MP4.

Not on the phone yet. The phone app has no screen recorder; it would need Android's own screen-capture permission and a new app build.

• Nothing is recorded until you choose a screen and press Start recording; the browser shows its own "sharing" bar the whole time, and stops the moment you press its Stop sharing.
• The recording goes only to this PC, in pieces every few seconds, so a long one does not fill the browser and a crash still leaves a playable file. A recording stops at 8 GB.
• No key and no install: the browser records, and the MP4 copy uses the ffmpeg that launch_web_ai.bat installs.

1. Open the pop-up

   Settings & appearance > AI providers & keys > Add keys.

2. Press the provider's name

   The blue underlined names with an arrow (for example OpenRouter ↗) open the key page in a new window. Providers with no site to visit (your own GPU node or render worker) have a plain name.

3. Paste and save

   Copy the key, come back and paste it in the box, then press Save all. A blank box keeps the key you already saved.

The phone app takes its models from the PC, so keys are added on the PC only.

• The links only open a website in your browser. PetAi sends nothing to it and never sees what you do there.

1. Open it

   Left menu > 🔍 Screener, or open /screener.

2. Write the query

   Type rules in the Query box, joined with AND, using > < >= <= or =. Press one of the three example lines under the box to start from it. The list Available metrics on the right shows every name you can use (market cap, P/E, ROCE, ROE, debt to equity, revenue growth, promoter holding, current price and more). Market cap is in crores.

3. Choose the stocks to look at

   In Watchlist type NSE symbols separated by commas or spaces (for example RELIANCE, TCS, IREDA). Leave it blank to use PetAi's default list of about 50 liquid names. The screener looks at your list, not the whole market, because the data is fetched one stock at a time.

4. Run it

   Press ▶ Run screen. A progress line shows the stock being read; the stocks that pass appear in a table below.

5. Look at one stock's data

   In See all data yfinance returns for a symbol type a symbol and press Show fields to see every field the data source gives for it.

Query: Debt to equity < 0.5 AND ROE > 20 AND Revenue growth > 15

Watchlist: TCS, INFY, HDFCBANK, IREDA

Press Run screen.

Not in the phone app. The page opens in a phone browser when the phone is paired with the PC.

• Which stocks are checked: only the ones on your watchlist (or the default list).
• No key and no account: the numbers come from Yahoo Finance through yfinance, which launch_web_ai.bat installs. A symbol it cannot find is skipped.

1. Open the Toolbox

   Open the Agent (the round bot) and press 🧰 in its top bar. Scroll down to 🔌 MCP servers.

2. Connect an MCP server

   Type a Name, pick A command to run (for example npx with the arguments -y @modelcontextprotocol/server-filesystem .) or A URL (for a server on the web, with an access token only if the service gave you one), and press Connect. The server then shows how many tools it brought. Sign in appears for servers that use your account. Refresh asks every server what tools it has now.

3. Switch off or remove

   The ◉ button on a server's row switches it off and on; ✕ removes it. Turning the MCP group off in the Toolbox stops the Agent using all of them.

4. Web APIs: tools from an openapi.json

   Under 🌐 Web APIs give a short name and the address of the service's openapi.json, add a key header and the key if the service needs one, and press Add. Every operation in it becomes a tool. The key is kept here and sent only to that API. Clear all removes every API and its key.

5. Let other apps use PetAi

   Tick Let other apps on this PC use PetAi, pick the app, press Copy config and paste it where that app keeps its MCP settings. The last calls other apps made are listed under it (kept in memory only).

Toolbox > MCP servers > Name: files, A command to run, command: npx, arguments: -y @modelcontextprotocol/server-filesystem . > Connect. Then ask the Agent: "list the files in the current folder".

Not on the phone: servers and APIs are set up on the PC.

• Which servers and APIs exist, whether each is on, and whether other apps may use PetAi. Nothing is shared until you tick the box.
• The Agent still asks before an action that changes things, whatever tool it comes from.

1. Start

   Under What should we create? type what you want. Optionally pick a Template (or Blank) and a Design system (the look: colours, fonts, spacing). Paste a reference screenshot with Ctrl+V or attach one with +; paste existing HTML, CSS or JS with </> to improve it. Press the arrow to build; Build full app makes a whole app instead of one page.

2. Answer a few questions or skip

   PetAi may ask A few quick questions. Answer what you like and press Build, or Skip & build and the model decides.

3. Look and change

   The design appears under Preview; </> Code shows the code. Switch Desktop, Tablet or Mobile to see each size. Ask for changes in plain words ("make the header darker", "add a pricing table") and press Update: the design is regenerated each time. Stop cancels a build.

4. Edit it by pointing

   Press Edit. In Select & style click any element to change it (the Element tree lists them); Use / navigate makes the design's own buttons work so you can move between screens, then switch back to style them. Tweaks has quick settings. Fix navigation and Check for issues ask the model to repair links or spot problems.

5. Use or keep it

   Open in tab shows it full size; Download .html saves it; Build as app turns it into an app. Designs are kept under Projects (search, star, list or grid view) and in the design-history folder. The Design systems tab saves a look you can reuse; Templates lists starting points.

6. Record a video

   For a design with animation, Record video captures it playing (pause, restart and half, normal or double speed are beside it).

Design > Template: Blank > type "A landing page for a bakery in Jalandhar, warm colours, menu, opening hours and a WhatsApp button" > build > Mobile > "make the buttons bigger" > Update > Download .html

Not in the phone app. The page opens in a phone browser when the phone is paired with the PC.

• Which model builds it (use a strong code model), which look it uses, and whether to keep a design: nothing leaves this PC except the request to the model provider you picked.
• Providers marked ⚠ failed recently, and "-web" bridge providers need the headless browser (the console offers to install it).

1. Open the Audio editor

   Press Editors in the left menu, then Audio in the pop-up. Or open /editors and press the Audio tab.

2. Choose a sound

   Press Choose a sound file, or drop a file on the dashed box. MP3, WAV, M4A, OGG, FLAC and the sound of a video file work, up to 300 MB and 4 hours. The wave appears with a player under it. Click the wave to play from there.

3. Keep or remove a part

   Drag across the wave to select a part: its start and end show under the wave. Press Keep only the selection to throw away everything else, or Remove the selection to cut that part out. The wave is redrawn, and Steps lists what you did. Times always mean the sound as it is now, after the earlier steps.

4. Fade, volume, speed

   In Fade and volume type seconds for Fade in and Fade out and press Add the fade; type decibels (3 is a little louder, -3 a little quieter) and press Change volume. In Speed pick Half, Slower, A little faster, Faster or Double and press Change speed. The voice keeps its pitch.

5. Make it better

   Clean up a voice lowers hiss and rumble and evens the loudness. Even out the loudness does only the loudness. Cut long silences takes out pauses longer than about a second. Make it mono mixes both sides into one. Reverse plays it backwards (up to 30 minutes).

6. Join another sound

   Open Join another sound, press Add a sound to the end and choose a file.

7. Layers: sound on top of sound

   Open Layers (sound on top of sound). Set Starts at (or press Use the play position), the Volume and fades, then press Add a sound layer and choose the music, voice-over or effect. Do it again as often as you like: each layer is a step in the Steps line and Undo takes the last one off. Untick Keep the sound underneath to mute the sound you opened so the new one takes its place; tick Let the layer run past the end to make the result longer.

8. Go back

   Undo removes the last step. Start over returns to the sound you opened.

9. Write what is said

   Under the wave, in Transcript, leave Spoken language on Find it by itself or pick the language, then press Write what is said. The first time, PetAi downloads the speech model once (a few minutes); after that it works offline. A bar shows the progress, the lines appear with their times, and the language it found is named. Click a line to play from there. The transcript is of the sound as you have edited it. Press Save as text, Save as subtitles (.srt) or Save as web subtitles (.vtt) to get a file.

10. Save the new sound

    In Save pick the kind of file and press Save the new sound. A link Download ...-edited.mp3 appears; the file is also kept in PetAi's media folder.

Open Editors > Audio, choose interview.mp3, drag across the first minute and press Remove the selection, press Clean up a voice, press Write what is said with the language on Find it by itself, then Save as subtitles (.srt) and Save the new sound as MP3.

On the Notes tab press ✂ Edit, choose a sound, type From and To in seconds and press Keep only this part or Remove this part, add Fade in and out, Louder, Faster or Clean up a voice, press Write what is said for a transcript in the spoken language, and Save as MP3. Not compiled or tried on a phone yet.

• Which sound to open, and nothing is saved until you press Save the new sound or one of the transcript save buttons.
• No key. The editing uses ffmpeg, which comes with PetAi's imageio-ffmpeg. The transcript uses faster-whisper, the same offline speech engine JARVIS uses; launch_web_ai.bat installs it, and if it is missing the transcript tells you so.

1. Open the Video editor

   Press Editors in the left menu, then Video in the pop-up. Or open /editors and press the Video tab.

2. Choose a video

   Press Choose a video, or drop a file on the dashed box. MP4, MOV, MKV, WebM and AVI work, up to 300 MB and 60 minutes. PetAi makes a quick, smaller copy to watch while you edit; the saved video uses the full size.

3. Keep or remove a part

   Drag across the wave under the video to select a part. Press Keep only the selection to throw away everything else, or Remove the selection to cut it out. Steps lists what you did; times always mean the video as it is now.

4. Fade, sound, speed, turn, look

   Open Fade, Sound, Speed, Turn and flip or Look and press the button next to what you set. Turn the sound off mutes the video. A video with no sound still works.

5. Join another video

   Open Join another video, press Add a video to the end and choose a file. It is fitted to the size of the first one.

6. Go back

   Undo removes the last step. Start over returns to the video you opened.

7. Write what is said, and enhance it

   In Transcript and subtitles leave Spoken language on Find it by itself or pick the language, tick Enhance: Hindi/Urdu script to Hinglish if your video is Hindi or Urdu, then press Write what is said. The first time PetAi downloads the speech model once; after that it works offline. Click the time stamp on a line to play from there. Click the text of any line to edit it — your changes are used when saving with captions. Save the words as text, .srt or .vtt.

8. Save, with a caption template

   In Save pick the Size, pick Captions (written on picture / track / both) and a Caption template — twelve to choose from including Hormozi (bold yellow), MrBeast (giant white), Karaoke (large top-bar), Reels-style, Neon Pop and Classic. Press Save the new video. A link Download ...-edited.mp4 appears; the file is also kept in PetAi's media folder.

Open Editors > Video, choose lecture.mp4, select the first minute and press Remove the selection, press Write what is said with the language on Find it by itself, set Captions to Written on the picture, Caption template to Reels-style big text, Size to 720 tall and press Save the new video.

On the Notes tab press ✂ Edit, choose a video and use the same buttons as for sound, plus Turn right and Sound off. After Write what is said tick Write the words on the picture, then Save the new video (720 tall) or Save only the sound (MP3). The Hinglish enhancement and caption templates are PC-only for now. Not compiled or tried on a phone yet.

• Which video to open. Nothing is saved until you press a save button.
• No key. Editing uses ffmpeg from PetAi's imageio-ffmpeg; the transcript uses faster-whisper, the offline engine JARVIS uses. The Hinglish rewrite is a fixed Devanagari-to-Latin table PetAi runs itself, not an AI call.
• Which caption template. Twelve styles burn in with ffmpeg's subtitle filter. Hormozi (bold yellow bar), MrBeast (giant centered) and Karaoke (large top) are social-media presets; the rest give classic, minimal, neon and cinematic looks. No real per-word karaoke animation — all lines are whole-segment.
• Caption text is editable: click any line in the Transcript section and type. The edited text goes into the saved video.

Say "Plan my content for the next 30 days from what sold best." with /small-business:content-strategy. It reads your PayPal and QuickBooks sales and returns a 30-day brief: what to push and which offers to run.

For a reels plan, use the prompt 30-Day Reels Calendar under 📋 Prompts.

"I need to post more" starts social-content-engine. It keeps a rolling posting calendar, learns your voice, makes on-brand Canva graphics and writes the captions. It stages posts for your OK rather than publishing on its own.

"Run this brief" starts canva-creator. It builds the designs and captions for a campaign and stages the sends in HubSpot.

To post on its own, press ⚙️ Templates, pick LinkedIn post Agent or Instagram Reels Automation, and choose Run on a schedule. Each social account has to be connected inside that automation.

"Are my ads working?" starts ad-manager. It reads TikTok Ads directly, and other ad platforms through a Zapier connection, and explains the return in plain words.

For new ads, use /ad-creative, or the prompts in the Meta Ads and Google Ads categories (for example Google Ads for New Websites). The Meta Ads Reporting template sends a daily report.

• "Find me more customers like my best ones" starts lead-finder, a ranked prospect list built with Apollo.io or Clay.
• "Leads are going cold" starts speed-to-lead. It watches your inbox, web forms and CRM, and drafts a reply to each new inquiry.
• "Who should I call first?" starts lead-triage, which scores the HubSpot leads worth a call today and gives talking points.
• "Write this cold email" starts outreach-composer. "Update my CRM" starts crm-autopilot.

"Is my marketing working?" starts growth-pulse: sales by channel, conversion and campaign return. "Catch me up on the business" starts business-pulse, a one-page snapshot of cash, sales and pipeline.

For a report that repeats, name the skill: /report-builder Sales by product this week versus last week, from Stripe. It gives a chat summary and a spreadsheet. The Daily sales report and Stripe notifications for new purchases templates cover the rest.

Put the routine parts on Left menu › Schedule › + New schedule. Three that work well:

On the Connectors tab, tick only what each schedule needs. A scheduled run cannot stop to ask you, so any action that needs your OK is skipped and listed in the result, unless Auto-run is on in the Agent's settings.

1. Check it is on

   Open the Agent panel and press Toolbox. Editors (pictures, PDFs, sound, video) is on by default; switch it off there if you do not want it.

2. Say where the file is and what you want

   Type the file's place and the job in your own words. The Agent picks the right tool and shows you the result as a download.

3. What it can do

   Pictures: crop, rotate, resize, light and colour, looks, words, border. PDFs: keep or reorder pages, rotate, join several, split, watermark, page numbers, set a password, read out the text. Sound and video: trim, cut, fade, speed, volume, clean up, turn, and the sound of a video as MP3. Transcript: what is said, in the language spoken, as text or subtitles.

4. Long jobs

   A transcript is made with the offline speech model. If a recording is long and it is still working after about five minutes, the Agent tells you and points you to Editors in the left menu, where it can keep running.

Make ~/Pictures/goa.jpg square and black and white, and put "Goa 2026" on it.

Join ~/Documents/part1.pdf and part2.pdf, add page numbers, and give me the new file.

Write what is said in ~/Downloads/lecture.mp4 as subtitles.

Ask the Agent in the phone app's Assistant tab in the same words. The work runs on your PC, so the file must be on the PC and the phone must be connected to it. The simpler Editors screen (Notes tab > Edit) is still there for doing it by hand. Not compiled or tried on a phone yet.

• Which files the Agent may touch: it refuses PetAi's own settings and key files, and follows your work-folder and permission settings for the files it reads.
• Nothing is overwritten. Each result is a new file in PetAi's media folder, shown in the chat.

1. Open the page and link your WhatsApp

   In the left pane of the Chats page click 💬 WhatsApp. At the top is a card Your WhatsApp with a tag that says not linked. Click Link WhatsApp (scan QR). A QR code appears on the card. On your phone open WhatsApp, then Linked devices, then Link a device, and scan it. When it works the tag says linked and the card says "Linked. Messages go out from your own WhatsApp number." Cancel stops linking and Refresh checks again. If the card says Playwright is missing or there is no browser yet, it tells you what to do: close PetAi and start it again with launch_web_ai.bat once with the internet on, or install Microsoft Edge or Chrome.

   You only link once. The Customize page has the same link button on its WhatsApp Web card.

2. Step 1: choose who gets it

   On the Send tab, under Who gets it, use any of these. You can mix them, and the line beside the heading counts how many are chosen.

   • Numbers: type one person per line, like +91 98765 43210 or Asha, +44 7700 900123. For numbers without a country code, fill in Country code for numbers without one, for example 91.
   • My contacts: tick the people you saved. All and None tick or clear the list, and there is a Search box. The link edit goes to Customize › Connectors › Contacts where you add them.
   • My WhatsApp chats: click Read my chats to list the people, groups and broadcast lists in your linked WhatsApp. It takes a few seconds - a group or broadcast list is sent to the same way as a person.
   • Google Sheet: paste the sheet's link in the box and click Read. Share the sheet as "Anyone with the link" (Viewer) first. Or open Or paste the rows and paste them. After reading, pick which column is the Name, the Phone number, the Message (optional) and the Picture or video (optional).

   Example sheet rows:

   Rows for Or paste the rows

   Name,Phone,Message,Image
   
   Asha,+91 98765 43210,Hi Asha,photo.jpg

   Copy
   Putting a picture or video into the Image column. A spreadsheet cell can only hold text, never the picture itself, so Insert > Image > Image in cell does not work here - PetAi cannot read a picture pasted that way, so that person's picture would not go out. Two ways that do work:

   1. A file name you already added above: click + Add pictures or videos under step 2 first, then in the sheet's Image column type that exact file name, like photo.jpg - PetAi sends that same file to everyone whose row names it.
   2. A Google Drive link, one picture per person: upload the picture or video to Google Drive, right-click it, Share > Share, set it to "Anyone with the link", Copy link, and paste that link straight into the Image column cell (it looks like https://drive.google.com/file/d/XXXXX/view?usp=sharing). PetAi turns it into a direct download by itself - no need to change the link. A plain https:// link to a picture or video hosted anywhere else works the same way.
3. Step 2: write the message

   Under The message type the words. Write {name} where the person's name goes. Use {first_name} for the first word of the name. Any column of a Google Sheet works too, like {city} or {amount}; small buttons under the box put them in for you. Example:

   Message

   Hello {name}, your class starts at 9 on Monday.

   Copy

   The most you can type is 4000 characters.

4. Add pictures, videos and files

   Under Pictures, videos and files click + Add pictures or videos, or drop files on the box, or paste a link that starts with https:// and click Add link. They go with every message. A sheet can also give one per person. At most 5 files go with one message, and a file can be up to 64 MB. Pictures, videos, audio of any common format, PDFs, Word, Excel, PowerPoint (old and new formats), text, CSV and zip files are allowed. A picture or video over about 16 MB still goes out fine - PetAi quietly sends it as a plain file instead, since WhatsApp Web itself won't preview that large a picture or video.

5. Templates

   PetAi starts with 19 ready-made templates: Welcome, plus 18 for webinars, Instagram, YouTube, course updates, PetAI releases, testimonials and support. Pick one from Template above the message box, or click Save as template to keep words you typed yourself. The Templates tab lists them all, with Use, Edit and Delete on each (on a built-in one, Delete reads Reset and brings back its first words). To make one, type a name and the words under New template and click Save template. If a template needs a link such as a webinar or course panel address, a box for it appears under the message once you pick that template - fill it in once and the same link goes to everyone in that send.

6. Step 3: check it, then send

   Click Preview each message. PetAi shows how many are ready and how many have a problem (those are skipped). Under Pause between people choose how long to wait between messages: 4 s, 8 s, 15 s, 30 s or 1 min. The pause is there so WhatsApp does not mistake it for spam. When it looks right click Send now, one by one. A progress bar and a table show each person as they go. Stop ends the run. You can leave the page open or come back; the Activity tab lists recent sends.

   Try it first with two numbers you own.

7. Send it later

   Under Or send it later type when, in your own words: tomorrow 9am, in 2 hours or every monday at 8. Or pick Once, Every day, Every weekday, Every week, Every month or Every year. You can type a name in Name it (optional). Click Schedule it. It then shows on the Scheduled tab with the next time, how many people and files, and the buttons Switch off (it then reads Switch on) and Delete. PetAi must be running at that time. A Google Sheet is read again when the time comes, so new rows are included.

8. Ask the Agent to do it

   Open the Agent with the round 🤖 button, press 🧰 to open the Toolbox, and tick Messaging (WhatsApp Web, no key). It is off until you tick it. Add your people under that group (the Add contact button there), then say:

   PetAi Agent

   Send this photo to Asha and Ravi tomorrow at 9: C:\Pictures\offer.jpg. Say: Hi {name}, our offer starts today.

   Copy
   PetAi Agent

   Wish all my contacts happy Diwali.

   Copy

   The Agent can name several people, say "all contacts", add up to 5 files, give a time and a repeat (daily, weekdays, weekly, monthly or yearly), and use {name}. It sends to at most 100 people in one go. Nothing is sent until you press Confirm. The card lists the people, the exact words, the files and the time. A file from your PC is copied into PetAi only after you press Confirm. A later send then shows on the Scheduled tab of the WhatsApp page, and a send to many shows on the Activity tab.

The WhatsApp page is a page of the PC console, and I found no WhatsApp page in the phone app. The phone app does show the Agent's waiting card.

1. Confirm an Agent send from the phone

   With your phone paired, a WhatsApp send the PC Agent is holding shows in the Waiting for your OK card on the Assistant tab under the words "Send a WhatsApp message", with up to 8 lines of detail. The button reads Send, with Cancel beside it. The message still goes out from WhatsApp Web on the PC.

• Whether to link WhatsApp at all, and when to unlink it. Nothing is sent until you press Send now, one by one or Schedule it.
• Who is on the list, what the words say and the pause between people.
• Whether the Agent may use WhatsApp: it is off until you tick Messaging (WhatsApp Web, no key), and every send it holds waits for your Confirm.
• Which scheduled sends stay on: switch one off or delete it on the Scheduled tab.

1. Pack the site into website.zip

   First pull the newest PetAi (see the folders). Then in the PetAi folder double-click bat\pack-website.bat. Its window says "Make website.zip (the pet-ai.si site) ready to upload". If it says Python ("py") was not found, install Python from python.org and run it again. When it finishes, it opens File Explorer with website.zip selected. You can instead run py scripts/pack_website.py. The zip holds the website folder's files at its top and nothing else: not the program, not your keys, not README.md. Before it writes the zip, it now also scans every file for anything that looks like a real API key or this PC's own Tailscale/home-network address or device name, and refuses to make the zip at all if it finds one - so a PC's private details can never end up on the public site by accident. pet-ai.si also now sends the same kind of security headers and scanner-blocking rules as the other site.

2. Upload it in Hostinger

   Sign in to Hostinger yourself. Open Websites › pet-ai.si › File manager › public_html. The window's own list says:

   1. Delete the old site files in public_html (not the folder itself).
   2. Upload the website.zip you made.
   3. Right-click website.zip and choose Extract, then delete website.zip.
   4. Open https://pet-ai.si and press Ctrl+F5.

   If the site shows no padlock, turn on the free SSL in hPanel under Security › SSL.

3. Check that only the site is online

   Open https://pet-ai.si/.git/config and https://pet-ai.si/server.py. Both must say "not found". Also open https://pet-ai.si/terms, /privacy, /refund and /pricing. Each must land on profithunterguru.com.

4. The optional automatic way (GitHub to Hostinger)

   The repository has a workflow named deploy-website. It uploads only the website folder to Hostinger whenever that folder changes on main. You can also start it by hand: on GitHub open Actions › deploy-website › Run workflow. It first runs a safety check that stops the upload if the folder holds a program file, a key file or a line that looks like a key. The only program file it allows is _render.php.

   It needs three secrets that you add yourself, in the repository on GitHub under Settings › Secrets and variables › Actions: HOSTINGER_FTP_SERVER, HOSTINGER_FTP_USER and HOSTINGER_FTP_PASSWORD. Make the FTP account in hPanel. If the run stops with a message that the Hostinger login is not set, one of them is missing. An optional variable HOSTINGER_FTP_DIR names the site's folder as the FTP user sees it; it is often ./public_html/. Each run uses a minute or two of the GitHub account's Actions minutes.

   Do not use hPanel's "Git" deploy on the PetAi repository: it would copy the whole private repository, program included, into public_html.

5. Changing the words from the admin panel

   On profithunterguru.com go to Admin › Editor PetAi and change the words of the home page or the tour page. Nothing is sent to GitHub. The one program file on the site, _render.php, draws those two pages with your changes written in. It reads them from profithunterguru.com, keeps them for a minute, and falls back to the last copy, or to the page as it is in the zip, if that site cannot be reached. It needs a host that runs PHP, which Hostinger's web hosting does. A word-change is found by its exact words, so if you later reword that sentence in the website folder, the old admin change stops matching and the new wording shows.

   The automatic deploy never touches the _cache folder where those changes are kept.

6. The preview and its pop-ups

   On the home page click See the console. You can look at Chats, Customize, Documents, Notes, Research, Schedule, Screener, AI Code, Design, Compare, the Guide and WhatsApp. Each message gets the same stand-in reply. Click a button that would save, run or send something and a pop-up opens with the title "“Name of the button” is a PetAi button". It says you are looking at a preview, nothing runs or is saved or sent, and no account is needed. It has two buttons: Get PetAi and Keep looking. Press Esc or click outside to close it. Nothing is read from the visitor or sent anywhere.

7. When the console changes later

   The preview and the tour pictures are made by two tools, scripts/build_site_console.py and scripts/render_site_tour.py, and the phone pictures by scripts/render_site_phone.py. Run them from a fresh copy of the code, then commit and publish the site again. The website README lists the commands.

These parts changed nothing in the phone app. Visitors see the site on their phone's browser: the home page has a band called "PetAi on your phone", and the site's icon is the face when someone saves it to the home screen.

• Whether and when to publish: nothing goes online until you upload the zip or run the deploy.
• Whether to use the automatic deploy at all. It stays off until you add the three secrets.
• What the home and tour pages say, from Admin › Editor PetAi.

Steps 1 to 3 are in the launcher window. Press S when it asks, or start it with launch_web_ai.bat --show.

1. The new library is added once

   At [3/4] Checking Python dependencies (kept on the pendrive)... the first start after the update shows Adding only what is new: waitress>=3.0, then Dependencies OK. With no internet it says Could not download them now - continuing with the libraries already here (PetAi tries again next launch). PetAi then starts on the old server, and the next start tries again.

2. See which server answered

   At [4/4] Starting PetAi..., right after PetAi is running at http://127.0.0.1:8123, a new line reads Server: waitress. If it says werkzeug, the old Flask server is answering. If a PetAi was already running and could not be closed, the launcher says A console is already running at ... and shows no Server line.

3. A quieter window

   After Ready in N s. the window shows the console's own output. One line there reads [server] waitress, 16 threads. Flask's WARNING: This is a development server banner is gone, and so is the line for every request. If waitress could not be installed, you see [server] waitress is not installed - using Flask's built-in server. Relaunch once so the launcher installs it. instead.

4. If you picked the background

   The launcher's lines, Server: waitress included, go to petai-launcher.log in the PetAi folder. The console's [server] line goes to petai-console.log. The launcher line Its output is also written to ... gives that file's full path.

5. Optional: check in the browser

   On the PC, open http://127.0.0.1:8123/api/ping (use the port from "PetAi is running at" if yours is different). The reply includes "server": "waitress". No page of the console shows this.

6. Your schedules now fire

   Open ⏰ Schedule in the left menu. The page says "They run on this PC while PetAi is running, even with its window closed." That is now true when you start with launch_web_ai.bat too. PetAi checks every minute, and schedules set to a GitHub event are picked up the same way. If a run was missed while PetAi was off, the most recent one runs once after PetAi starts again, and its history says late run. Runs started by the timer carry the badge Scheduled.

7. The floating bot follows the console

   The same start now looks after the sidebar switch 🤖 Floating desktop bot: with it on, the floating bot appears while a console page is open and goes when the page is closed. When the console stops, the floating bot and OmniRoute are closed too.

There is no new screen for the web server. The phone talks to the same console, and replies still arrive as they are written.

1. See your schedules

   Open the Workers tab and scroll to Schedule. It says "Tasks your PC runs on a timetable or when something happens on GitHub, while PetAi is running." These now really run when the PC was started with launch_web_ai.bat.

2. Stop or run one

   Under Yours, each schedule has a switch on its row. The switch has no label. Turn it off to stop that schedule. Run now runs it once.

• After the first new start, open Schedule › Yours. Every schedule that is Active will now run on its own and may use model credit. To stop one, open it and press Pause. Run now runs it once by hand.
• Only if the new server gives you trouble: open Command Prompt in the PetAi folder, type set PETAI_WSGI=dev, then run launch_web_ai.bat --show from that same window. That one start uses Flask's old server. The setting is not saved, so the next normal start uses waitress again.

1. Pack it

   Double-click bat\build_mac_linux_app.bat. It writes PetAi-mac-linux.zip into %LOCALAPPDATA%\PetAi\desktop-apps and says where. Your keys, chats and settings are never put in it.

2. Start it there

   Unzip on the Mac or Linux PC. Mac: double-click PetAi.command (first time right-click > Open). Linux: ./petai.sh in a terminal. If Python 3.10 or newer is missing, the window shows the exact install command for that system.

3. Make it an app

   Once, run ./make-app.sh. On a Mac it makes PetAi.app in ~/Applications with PetAi's icon; on Linux it adds PetAi to the application menu.

Windows: bat\build_mac_linux_app.bat

Mac:     unzip PetAi-mac-linux.zip, double-click PetAi.command, then ./make-app.sh

1. Prepare the project

   Double-click bat\build_ios_app.bat. It writes PetAi-iPhone.zip (same folder as above): a small Xcode project that pairs with your PC and shows the console.

2. Build it on a Mac

   Install Xcode, unzip, run ./build-ios.sh, then open PetAi.xcodeproj, pick your iPhone and your Apple ID, press Run. A free Apple ID works for 7 days at a time.

3. Pair

   In the app paste the pairing link from Customize > Connectors, or type the PC's address and code. The PC must be reachable (same Wi-Fi, or Tailscale on both).

• Whether to carry your keys over: they are left out on purpose. Sign in again on the new computer.
• Whether to use an Apple ID (free, 7 days) or a paid developer account for the iPhone.

1. Open the build window

   On the pendrive, double-click build_petai_exe.bat. Its black window says "Building PetAi.exe - a window will open to choose what goes in." and then the Build PetAi.exe window opens. The first build gets the build tool from the internet once, into the pendrive's own folders. Nothing is installed for Windows.

   When PetAi runs from the pendrive, Customize › Connectors says the same at the top, under "PetAi build running", in a line that starts Make PetAi.exe.

2. What always goes in

   At the top, "Always included" counts PetAi's own program files. They always go in, so they have no boxes. Show list lists them.

3. Tick the extras

   Under Extras, each group has a box for the whole group, its size, a short note and a box per file or folder. Readmes and guides starts ticked. Runtime (Python) starts ticked when the pendrive's own Python can show windows; it lets PetAi run on a PC with no Python, and it makes PetAi.exe bigger. Keys and sign-ins and Chats and history are marked Sensitive and start unticked every time. Ticking one shows "Anyone who has this .exe can unzip it and read these. Leave them off unless the .exe stays with you." Build then asks first, with Go back and Build anyway. The bottom line says "PetAi.exe will be about" and a size.

4. Choose the icon

   Under the list, the Icon for PetAi.exe card shows the icon PetAi.exe will get. It starts with "PetAi's own icon". Choose icon... opens "Choose an icon for PetAi.exe", which lists icons and pictures (.ico, .png, .jpg, .jpeg, .bmp, .gif and .webp). A picture is "fitted into a square, never stretched". Use PetAi's icon goes back. Only that one file is used. A file that is gone or will not open is named on the card, and PetAi.exe gets PetAi's own icon. The build never fails over its icon.

5. Build and save

   Press Build PetAi.exe. The window shows four steps: "Getting the build tool", "Packing your choices", "Building PetAi.exe" and "Checking the result". Show details shows the end of the build log, and Cancel stops the build. It takes a few minutes. Then the Save PetAi.exe dialog opens in the folder you used last, or Documents, never the Desktop. The last view says "PetAi.exe is ready" with its size and place, and has Show in folder and Done.

6. Save it outside the PetAi folder

   A folder inside the PetAi folder asks first, because push-to-github.bat would commit the file and the launcher would copy it to every PC. Press Choose another. Save here anyway saves it there.

7. If Windows Security stops it

   The window says "Windows Security stopped PetAi.exe. Open Windows Security > Protection history > Allow, then build again." The build never adds an exclusion.

With a Python that cannot show windows, the black window shows a numbered menu instead. build_petai_exe.bat --console always uses that menu, and build_petai_exe.bat --icon <file> picks the icon there. Your ticks are remembered for the next build, except keys and chats.

• Whether to tick Keys and sign-ins or Chats and history. PetAi.exe is a plain zip behind a small starter, so anyone who gets a copy can unzip it and read them. With them left off, sign in on the new PC, or launch PetAi once from the pendrive on that PC to bring your settings and chats over.
• Where to save PetAi.exe, and whether it gets your own icon.

1. Double-click PetAi.exe

   It unpacks PetAi into %LOCALAPPDATA%\PetAi\live and starts it in the background, as launch_web_ai.bat does. A small "Setting up PetAi" window shows "Unpacking PetAi" while it works. If the PC has no Python that PetAi can use, and PetAi.exe carries none, it asks "PetAi needs Python 3.10 or newer. Install Python 3.12 for this user now? (no admin needed)" with Install and Cancel.

   A copy that came through a browser, mail or a cloud drive can first show "Windows protected your PC". Press More info, then Run anyway. A PetAi.exe built on the same PC does not show it.

2. The first time: Set up PetAi on this PC

   A window called "Set up PetAi on this PC" says where PetAi lives and has three boxes: "Start PetAi when I sign in (no permission needed)", "Start PetAi when this PC powers on, before anyone signs in (for the phone)" and "Put a PetAi shortcut on the desktop". The first two start ticked. Press Continue. Windows asks for permission once, Yes or No. The window then shows "Start at power-on: done" and "Phone firewall rule: done", with Done.

3. If you say No, or Not now

   Not now changes nothing. No on Windows' prompt leaves the start at power-on and the phone firewall rule as they were, and the window says "Start at power-on and the phone firewall rule were not changed: Windows' permission prompt was declined." Your sign-in choice and the shortcut are still kept, because they need no permission. Either way the window does not come back on that PC. Use Set up on the Tailscale card instead (see the next part).

4. Later double-clicks

   No window asks anything. PetAi.exe puts in any newer program files and starts PetAi, or opens the PetAi already running when nothing changed. Your settings, chats and data on that PC are only filled in when missing, never overwritten. If the PC already runs a newer PetAi, it asks once per PetAi.exe, for example "This PC already has a newer PetAi (from the pendrive).", with Keep the newer one and Use this exe's version.

5. When Windows asks again

   Only for a few admin changes, each with one Yes or No: the "Also start PetAi when this PC powers on" switch, Let the phone in through Windows Firewall, Keep Tailscale on after restarts, and Update or Remove on the "One-time setup" line. Installing Node.js or Tailscale, when PetAi finds them missing, also asks once. Nothing keeps administrator rights afterwards. On a Windows account that is not an administrator, Windows asks for an administrator's name and password at those same moments instead of Yes or No.

6. The file itself

   Starting at sign-in and at power-on uses the copy on the PC, not PetAi.exe, so you can move or delete PetAi.exe afterwards. If something goes wrong, every step and error is written to petai-exe.log in %LOCALAPPDATA%\PetAi, and every error message names it. Run PetAi.exe with --show to see the start-up in a window.

Nothing to install. With the start at power-on and the phone firewall rule done, the phone can reach PetAi again after a restart or a power cut, even before anyone signs in. Away from home this also needs Keep Tailscale on after restarts on the Tailscale card, and the PC must turn itself back on after a power cut, which is a BIOS setting.

• The three boxes in "Set up PetAi on this PC". Untick the start at power-on if the phone does not need PetAi before you sign in.
• Yes or No on Windows' prompt. No still leaves PetAi working at sign-in and on home Wi-Fi.

Open Customize › Connectors and scroll to the Tailscale card.

1. The One-time setup line

   Under Keep Tailscale on after restarts, a line says "One-time setup:" and then "done", "needs an update" or "not set up". The button beside it reads Remove, Update or Set up. The line shows on Windows, in a copy of PetAi on this PC that has this version, whether PetAi.exe or the pendrive made it. It shows only on the PC itself, never in the phone's browser.

2. Set up

   Opens the same "Set up PetAi on this PC" window on the PC's screen. The card says "The Set up PetAi on this PC window is open on this PC." If PetAi is running from the pendrive just then, it says "PetAi is running from the pendrive right now. Let the launcher start it from the copy on this PC first, then press this."

3. Update

   Shows when the start at power-on points at a Python that has moved, or still runs PetAi as SYSTEM. Under the line it says why, for example "The power-on start still runs PetAi as the Windows SYSTEM account, which any program on this PC could misuse. Press Update so PetAi starts as you. Windows asks for permission once." Update replaces only the start at power-on. The phone firewall rule stays as it is.

4. Remove

   Asks first: "Remove the one-time setup? PetAi will no longer start at power-on, and the phone firewall rule is removed. Windows will ask for permission once." After Yes, the card says "Removed. PetAi will not start at power-on, and the phone firewall rule is gone. The phone still works on home Wi-Fi." If you say No to Windows, it says "Not changed. Windows' permission prompt was declined."

5. An older start at power-on

   Where the One-time setup line is not there to do it, an older task that runs as SYSTEM shows under the "Also start PetAi when this PC powers on" switch: "On, but as the Windows SYSTEM account, which any program on this PC could misuse. Press Fix so PetAi starts as you. Windows asks for permission once." Press Fix. It replaces the task in one prompt. Where the One-time setup line shows, it says "needs an update" instead, and Update does the same, so there is one button, not two.

6. If Windows could not start it

   The switch's line then starts "Windows could not start PetAi as you at the last power-on" and says "Your account may not be allowed to run tasks while signed out." Ask the PC's administrator to allow it, or keep PetAi starting at sign-in.

• Whether PetAi starts at power-on at all. Remove takes it away together with the phone firewall rule.
• When to press Update or Fix. Until you do, an older task keeps running PetAi as SYSTEM.

1. How to update

   The copy from PetAi.exe has no git, so git pull does not work there. At the top of Customize › Connectors, under "PetAi build running", a line starts "This PetAi came from PetAi.exe", gives the date it was built and the commit, and ends "To update: pull on the pendrive and run build_petai_exe.bat." Do that, then double-click the new PetAi.exe on the PC. Launching PetAi from the pendrive on that PC updates the copy too. If the start at power-on then points at a Python that has moved, the One-time setup line says "needs an update": press Update.

2. The pendrive meets the copy

   The first time you launch from the pendrive on a PC where PetAi.exe set PetAi up, the launcher says "This PC's copy came from PetAi.exe: the pendrive's chats and settings are used." The PC's own versions of those files are moved aside beside the copy, not deleted, and the launcher says how many. After that the newer file wins again, as before.

3. Remove PetAi from a PC

   Open Command Prompt in the folder that holds PetAi.exe and run:

   Command Prompt, in the folder with PetAi.exe

   PetAi.exe --uninstall

   Copy

   A window called "Remove PetAi from this PC" opens, with Cancel as the default. Remove stops PetAi, takes away its start when you sign in, and asks Windows once to take back the start at power-on and the phone firewall rule, when there is one. Your chats and settings on that PC stay, unless you tick "Also delete my chats and settings on this PC". It ends with "PetAi was removed from this PC." It only removes the copy on this PC, never a pendrive.

From the pendrive, py run.py --forget-pc also removes the copy on that PC, after carrying your data back to the pendrive.

To remove this whole PetAi folder itself (not just a PetAi.exe copy) - its saved keys and settings, everything it installed for itself under %LOCALAPPDATA%\PetAi, and the folder it runs from - double-click bat/uninstall.bat, type DELETE when it asks, and press Enter. This cannot be undone: anything not pushed to GitHub or saved elsewhere is gone for good.

• Whether to keep your chats and settings on a PC when you remove PetAi from it.
• When to build a new PetAi.exe. A copy from PetAi.exe does not update itself.

1. Where things are now

   Open the PetAi folder in File Explorer. These are the new places:

   • docs: the guides and notes (AGENT.md, ANDROID-APP.md, GIT-SETUP.md, TRADING.md, WORKER.md and the others).
   • bat: the helper .bat files: build_petai_exe.bat, build_phone_app.bat, pull-from-github.bat, push-to-github.bat, pack-website.bat and uninstall.bat.
   • requirements: the pinned libraries, requirements.txt.
   • data: what PetAi writes by itself and can rebuild: provider-health.json, direct-dead.json and phone-seen.json. They used to sit at the top; PetAi moves them here the first time it starts.
   • logs: every log PetAi writes, such as petai-console.log and petai-launcher.log. The folder is made when PetAi needs it, so you may not see it until PetAi has run once.
   • The program's code is in module folders: agent, providers, trading, messaging, workspace, phone, screen, exe_build and training. They are only places to look. You never need to open them to use PetAi.

   If you looked for petai-console.log at the top of the folder before, it is now inside logs.

2. What stays at the top

   launch_web_ai.bat, run.py, server.py and the small launcher helpers stay where they were, because your Windows start-up entry and the phone's start by those paths. The pull helper is bat\pull-from-github.bat; nothing for it sits at the top any more. Your data stays too: chats.db, chat-history, config.json and your other private files are not moved.

3. Pull the update (a PC that still has the old layout)

   On a PC that still has the old layout there is no bat folder yet. Do this once:

   1. Close PetAi and its console window.
   2. In the PetAi folder, double-click the pull-from-github.bat at the top, if your PC still has one: it is the old script on that PC. Once the bat folder exists, use bat\pull-from-github.bat from then on.
   3. It shows what will change. Answer y to "Apply GitHub's version to this folder now? (y/N)".
   4. Open the PetAi folder and check that a bat folder now exists. That is how you know the pull worked.
   5. Double-click launch_web_ai.bat as usual.

   The window may close by itself, or print a stray line, right after "Updating your local code to match GitHub...". That is the old script being replaced while it runs. The pull has still happened, so check for the bat folder. From then on use the one in bat; the old one at the top is removed by this pull. The window ends with "Done. PetAi's files are now organized in folders".

4. Your shortcuts to the other helpers

   If you made a desktop shortcut to build_petai_exe.bat, build_phone_app.bat or push-to-github.bat, those files are now in bat, so the shortcut will not find them. Make the shortcut again from the file in bat. They start correctly from there, because each one moves up one folder first.

5. The copy PetAi runs from

   PetAi runs from a copy of its folder in %LOCALAPPDATA%\PetAi\live. That copy used to keep old files forever, and an old agent.py or db.py at its top would have been used instead of the new one in a folder. So at start-up PetAi now removes from that copy any top-level .py file that has moved into a module folder. It prints a line saying how many old files it removed from this PC's copy. It removes nothing else. Files of your own are left alone.

6. PetAi.exe and the pendrive

   The lists of what goes into PetAi.exe and onto the pendrive follow the new places, so the build window still offers the same boxes. The logs folder is not copied to the PC's copy and is not committed to GitHub.

These parts changed nothing in the phone app. The phone starts PetAi by the same launch_web_ai.bat, run.py and server.py paths as before. The phone-app build script is now bat/build_phone_app.bat.

• When to pull the update. Nothing moves until you answer y.
• Whether to remake any shortcut you had to a helper .bat file.

1. Why the Agent asks

   A confirm card can now carry a line such as "Asked because this answer read outside content (a web page) before this step." Press Confirm & run or Cancel as before. That OK is for that one step and lapses after 10 minutes. Answers that read nothing from outside work as before.

2. Your paired phones

   Press 📱 in the Agent panel's top row. Under Paired phones each phone shows its name and when it was last seen. Tick Chat only to let a phone chat but not see or control the PC's screen. Press Revoke to shut out one phone: "It stops reaching this PC at once and has to scan the QR code again." The others keep working. The same list is on the Tailscale card in Customize › Connectors.

   A phone that paired before this update shows as Phones on an older app (shared pairing code). It gets its own key the next time it pairs with the new app.

3. Share PetAi with other apps

   In the Toolbox, scroll to Share PetAi with other apps. Tick Let other apps on this PC use PetAi. Pick Claude Code, Claude Desktop or Cursor, press Copy config, and follow the line under it, for example "Run this once in a terminal on this PC. Claude Code then has PetAi in every project." Those apps can then recall and save memories, search your knowledge bases, list and add notes, read your documents and make a picture (only if your 🔒 Permissions let the Agent do that without asking). They cannot send anything, run commands, or change or delete what is there. PetAi must be running. This is on the PC only.

4. Backup

   Click ⚙ Settings & appearance at the bottom of the Chats sidebar and find Backup. Export backup saves one file with your memories, skills, prompt profile, notes, documents, knowledge base list, chat folders and a few settings. "Keys, passwords and paired phones are never in it." Import backup first shows what it would add; nothing changes until you press Import. Tick "Also replace what is newer on this PC" only if you want that. Import only a backup you made yourself.

5. A check before you upload to GitHub

   If you upload with push-to-github.bat, it now looks for keys in what you are about to send. If it finds one it stops with "STOP: possible key found above. Nothing was uploaded."

1. Why it asks

   The cards Waiting for your OK on the Assistant tab and the PC tab show the same reason line.

2. A chat-only phone

   In Settings, the This phone row adds "(chat only)". The PC tab says "This phone is set to chat only on the PC, so it cannot see or control the screen." Untick Chat only for it on the PC, then reconnect.

3. Notes from another app

   A memory another app saved reads "from another app over MCP". Check it before you act on it.

• Which phones stay paired, and which are Chat only.
• Whether other apps may use PetAi. Let other apps on this PC use PetAi is off by default.
• When to make a backup, and what to import.

There is nothing to switch on. You see it the next time you download a zip.

1. Download a project from AI Code

   From the left menu open AI Code, open a project that has files (the button stays grey until it has at least one), and click Download .zip above the file list. Unzip it. You get the project's folders and files and nothing else.

2. What is left out

   PetAi's own notes about the project (.aicode.json and the .aicode-undo copies it keeps for Undo), the .git folder, and the folders that hold installed packages, caches and build output, including node_modules, __pycache__, .venv, venv, dist, build and .next. Capital letters make no difference, and the rule holds at any depth, so a node_modules folder inside a sub-folder stays out too. Run npm install again after you unzip, as you would after downloading any project.

3. Nothing piles up

   PetAi used to write the zip into the ai-code-projects folder and leave it there. Now it makes a temporary file that is deleted as soon as the download is sent, cut off or refused.

4. The chat's zip

   In Chat, when a reply has files, click Download all (.zip) in the files pane. It works as before. What changed is the safety check: a file name that is an absolute path, climbs out with .., starts with a drive letter, is a Windows reserved name such as CON or NUL, or touches .git or .aicode is refused, and the whole zip is not made. The chat shows a message that says why. A zip is also capped at 300 files, 2 MB a file and 20 MB in all. Two files with the same name (capital letters ignored) get _2, _3 and so on, so none overwrites another.

   Replies can now give files folder names, so this zip can hold folders: see whole projects in Chat and in AI Code.

These parts changed nothing in the phone app.

• Whether to download a zip at all. Nothing is sent anywhere; the file goes to your browser's downloads.
• There is nothing to set up. There is no switch for these parts.

There is nothing to switch on. You see it the next time you change the look, or open one of those pages.

1. Choose your look

   Click ⚙ Settings & appearance at the bottom of the Chats sidebar. Under Appearance pick Light, Dark or System (System follows your Windows setting). Under Text size pick Small, Normal or Large. Under Theme pick a colour; point at a swatch to see its name: Studio Teal, Indigo, Violet, Azure, Emerald, Rose, Copper, Graphite, Midnight or Paper.

2. Open the other pages

   From the left menu open Customize, Schedule, Research, Notes, Documents, Compare, AI Code and Screener. Each one now shows in the look you chose: light or dark, in your colour, and Text size changes their text too. The Agent panel follows it as well.

   Terminal and log blocks stay dark in both modes, so code stays easy to read. The pairing QR on Customize keeps its white square so a phone can still scan it.

3. Check the text on buttons

   Pick Dark, then go through the colours. Text on a button or tag filled with your colour is near-black in dark mode and white in light mode, which reads on all ten colours. In light mode, Emerald and the grey helper text are a little darker than before so they can be read too.

4. A change reaches everything that is open

   Open PetAi in two browser tabs and change Appearance in one. The other follows without a reload. The floating desktop bot and its panel pick up the change within about five seconds, or at once when you click them.

5. Ask the Agent to do it

   The Agent can change the look for you, one thing at a time: a mode, a colour name or a text size. Try:

   PetAi Agent

   Switch PetAi to dark mode.

   Copy

   It answers with a short line saying what it set, and the page changes at once. Graphite is the colour the Agent calls slate. Midnight and Paper came later: see Settings.

These parts changed nothing in the phone app.

• Which mode, colour and text size you use. Nothing changes until you pick one, or ask the Agent to.
• There is nothing to set up. There is no switch for these parts.

There is nothing to switch on.

1. Open the groups

   Click ⚙ Settings & appearance at the bottom of the left menu. At the top are a box, Search settings, and a link, Open full settings. Under them are seven groups you open and close with a click: Appearance, Startup & browser, AI providers & keys, Local AI, Auto detect, Backup & data and Danger zone. PetAi remembers which ones you left open. Appearance is open the first time, and AI providers & keys opens by itself once you have picked a provider. Reset sidebar width now sits inside Appearance.

2. Search

   Type backup in Search settings. The groups with no match disappear and the ones that match open. The search reads only the words on the page, never what you typed into a box, so a key or an address can never be matched or shown by it. If nothing matches, the menu says "No setting matches that. Clear the search to see them all." Press Esc in the box to clear it.

3. The full-size window

   Click Open full settings. The menu becomes a window in the middle of the page: the groups as a list on the left, the chosen group on the right, the search on top. The list ends with Agent > Toolbox and Customize > Connectors, for the settings that live there (permissions and work folders; GitHub, Ollama and the plugins), so you can find every setting from one place. Close it with Close or Esc. Esc closes only this window and never touches the message box behind it.

4. A link to one group

   Add a group name after your PetAi address and the full window opens on it. For example, #settings/backup opens Backup & data. The names are appearance, startup, providers, local-ai, auto-detect, backup and danger, each after #settings/.

5. The Danger zone

   It holds only one button, Reset all credentials & dashboard logins. It deletes every saved key and login, so it sits alone, at the bottom, in red. There is no need to press it to try this part.

6. An Appearance button on every other page

   Open Customize, AI Code, Compare, Design, Documents, the Guide, Notes, Research, Schedule or the Screener. In the top bar, next to the Website link, there is a button with a half-filled circle: Appearance. Click it for a small box with Mode (Light, Dark, System), Text size and Colour, and a Preview with a button, muted text and a link. A change shows at once on every page and window you have open.

   In Customize, the Connectors tab also has a card, Appearance, with the same three rows and the line "One look for every PetAi page, the panel, the avatar and a paired phone."

7. Two new colours

   In ⚙ Settings & appearance, under Theme, there are now ten swatches. Point at the last two: Midnight and Paper. Each is a full set of colours with a light and a dark version, not just a new button colour. The same ten show in the Appearance button's box. The Agent knows them too, so it can switch to either when you ask.

8. The Customize link has a new icon

   In the left menu, Customize now has a puzzle piece instead of a gear, so it is not mistaken for ⚙ Settings & appearance. On Customize, the start-with-Windows rows say "Also in Settings > Startup & browser."

The phone's own Settings was regrouped and made searchable in the same round: see the phone group.

• Which groups stay open, which colour, mode and text size you use, and whether you ever use the full window.
• There is nothing to set up. There is no switch for these parts.

1. Copy and Save on a code block

   In the Assistant tab, ask for a small piece of code, for example a Python script called hello.py. Above each code box, on the left, is the file's name if the answer gave one, otherwise the language. On the right are Copy and Save. Copy puts the code on the phone's clipboard and shows "Copied". Save writes it to Downloads/PetAi on the phone under that file name, and shows where it went. This needs no PC.

2. Save a project as a zip

   Ask for a small website with index.html, style.css and script.js, each in its own named code box. When an answer names two or more files, a button Save N files as a zip appears under it. It makes one zip with the folders kept, under the same safety rules as the PC's, and saves it in Downloads/PetAi.

3. A project zip from the PC Agent

   With the PC paired, ask the Agent for a project as a zip (see the Agent group). The answer shows a chip, Download and the file's name. Tap it: the phone fetches the zip from the PC and saves it. A zip has no viewer, so it is not opened; the line under the chip says where it is and to open it from the Files app.

4. The PC's look, or your own

   Open Settings, then the Appearance section, and its Appearance card. With a PC paired, a switch, Follow my PC's look, copies the PC's light or dark and its colour; the card says they come from "Customize > Appearance" on the PC. Below it, Light or dark offers Same as the phone, Light and Dark, and Colours lists the same ten as the PC: Teal, Indigo, Violet, Azure, Emerald, Rose, Copper, Graphite, Midnight and Paper. Choosing one here switches Follow my PC's look off; switching it on copies the PC's look again. As far as the code shows, the phone reads the PC's look when it connects, not while it is already connected, and it does not copy the text size. To try: choose Rose on the PC, then close and reopen the app.

5. Settings in six sections

   The Settings tab now begins with a box, Search settings, and then six sections you open and close by tapping their title: Account & connection (Connection, GitHub, App lock, Log out), Appearance (Appearance, Floating avatar), Agent (Agent model, JARVIS mode, Permissions, How it talks, Where commands run), Data & memory (What the Agent remembers, Training data), PC & extras (Plugins, Local AI on the PC) and About. Only the first is open at first. Some headings show only when a PC is paired. Type model or theme or auth into the search box: the sections that have a match open, the others go, and the headings that do not match are hidden. If nothing matches, it says so, and the small cross in the box clears it.

6. "Auto picked" under an answer

   With the PC paired and the Agent's model on Auto (best available) (Settings, Agent section, Agent model), ask the Assistant something and read under the answer. A small line says, for example, Auto picked qwen3-coder for coding. It is the PC's pick (see Auto detect). With a model chosen by hand there is no line.

7. The question before a pull request

   In Settings, Account & connection, GitHub, connect your repository. In the Assistant tab ask it to fix a typo in the README. Before it makes anything, a box titled Open a pull request? asks a question such as Create a draft pull request changing README.md on branch petai/fix? with Create and Cancel. Cancel makes no branch and no pull request, and the Assistant is told so and tells you nothing changed on GitHub. It is always a draft that you merge on GitHub yourself.

8. The GitHub token is locked

   There is nothing to see. The token the phone keeps is now sealed with a key that lives in the phone's key store and never leaves it. After the update you should still be signed in to GitHub: an older plain token is read as it is and sealed at the next start. If the key is ever lost (a restore onto another phone, a reset), the phone asks you to sign in again; it never sends the sealed text.

Nothing changed on the PC for these parts. The look it hands to the phone is the one from Settings.

• Whether the phone copies the PC's look or keeps its own.
• Whether each pull request is made (Create or Cancel), and each push the PC holds.

There is nothing to switch on. You see it the next time you build PetAi.exe.

1. Build PetAi.exe

   Double-click bat/build_petai_exe.bat in the PetAi folder. In the build window, under the list, find the card Icon for PetAi.exe. It shows the icon PetAi.exe will get. With no icon chosen it shows PetAi's face and the name PetAi's own icon. Click Build PetAi.exe, choose where to save it, and look at the new PetAi.exe in File Explorer. Its icon is the face. The icon file holds every size Windows asks for, from 16 to 256 pixels, so it stays sharp in small and large views.

2. Use your own icon instead

   On the same card click Choose icon... and pick an .ico or a picture. PetAi fits a picture into a square and never stretches it. Click Use PetAi's icon to go back to the face. PetAi remembers your choice.

3. Where else the face appears

   The face is the Agent's avatar (the round bot you open with 🤖). On the website it is the logo in the page corner, the small icon in the browser tab and the icon when someone saves the site to a phone's home screen. Those come from logo.svg, favicon.ico, icon-192.png, icon-512.png and apple-touch-icon.png in the website folder. See the website.

These parts changed nothing in the phone app.

• Whether PetAi.exe keeps the face or gets your own icon with Choose icon....