Report ready
⚙️ Preferences
Agent Pi server v…
An agent is a saved chat: its own system prompt, model, tools and @handle. Open one in its own tab, or type its @handle in any chat to call it.
No custom agents yet
Prebuilt roles. Adding one gives you an editable copy.
Configure default orchestration when multi-agent @mentions or channel conversations run.
This chat runs on …
Checking…
Which engine runs what
Little Coder: models on this computer (LM Studio, Ollama).
Pi CLI: beta testing only.
oh-my-pi: everything else.
In use - v?
Everyday Essentials
Choose your AI models, voice reading, typing helpers, and shortcuts
🧠 Default Models & Planning Which AI answers you by default and takes over when quota runs out
🔊 Read Replies Aloud & Voice
Read replies aloud: when a reply in the chat on screen finishes, Agent Pi reads it to you. Code and app lines are left out. Press Esc, or Stop reading, to stop.
✨ Prompt Autocomplete
Instant inline predictions and suggestions as you type in the message input. Press Tab to accept, Esc to dismiss.
Finishes the word you are typing (from words you used before), slash commands and @names. No whole sentences.
Uses local models (Ollama or LM Studio) on loopback (127.0.0.1 / localhost / ::1) when running. Cloud models remain disabled by default to prevent leaking draft text.
⇄ Tabs & Connection Retry
⇄ Auto-sort tabs
When a reply stops with an error, try again by itself. A lost connection waits until it is back.
Safety, Protection & Approvals
Control which folders the AI can touch, when it must ask you, and when autonomous goals stop
Undo changes: remember how your files were before each message, so they can be put back. You switch it on per chat, here or in the Addons menu.
🛡 Protection
How far an agent may go. Each project folder has its own level, on the lock beside the folder.
Starting level for new projects ?
Held by rules. ?
Pictures of your screen ?
Protected list
Agentic Secrets
A locked store for keys. Agents use a key without seeing it.
🛡️ Safety & Command Approval
This machine
🎯 Goal limits (all chats)
Goal off - each message is its own request. Goal on - the chat carries an objective and Agent Pi sends the next turn itself after each reply, until the agent says the goal is complete, says it needs you, you press Stop, or one of these limits is reached. Stop a goal after:
Messages
Minutes
Usage budget
Spend ($)
No goal is running in the chat you have open.
Alerts, Teamwork & Tool Display
Desktop popups, sound chimes, parallel helper agents, and tool step display
🔔 Notifications & Sounds
50%
Fine-tune individual alert types (7 events: response, question, error, stall, timers)
Per-Event Notification & Sound Configuration:
✅ Response Complete
Agent finished responding
❓ Decision Needed
Interactive question or escalation
⚠️ Errors & Failures
Tool error or process failure
⏳ Still Working / Stalls
Silent step running > 45s
🐢 Tool taking too long
One command/tool still running after
min
⏱️ Long-running turn
Agent has been working on one turn for
min
🤝 Long AI-to-AI conversation
Agent-to-agent channel run lasting over
min
🐝 Parallel helpers Multi-Agent & Swarm
🧰 Tool Calls
AI Engine, Memory & System
Turn-by-turn reasoning, token compaction, chat fonts, long-term memory, and server hardware
🧭 Reasoning & Quality
1 Start of every turn
Active for chats on the providers ticked below.
Active Providers:
2 Every file edit
3 End of every turn
💾 Keep long chats small Context & token optimization
1 As each tool result comes back
Ready (bundled)
2 As each request is sent
Off (API-key chats only)
Engine chats (omp) never go through this proxy; they rely on Compaction (step 3).
3 When the chat fills up
Compact at: 75%
and never later than thousand tokens (~750 words per 1k)
🔤 Chat text
Line height
Line height
🧠 Memory & Auto-Actions
at most tokens
Automated Triggers (after each assistant turn):
🧠 Agent memory & folder search
Gives every saved agent a private long-term memory and its own searchable index of the folder it works in. Nothing leaves your computer - it runs in three local Docker containers. Agents work fine without it; this only adds recall.
🧠● Optional add-on
Memory stack
Optional local Docker recall & folder index
🔌 Connections, Linked Peers & Diagnostics
Start a connection only when it is used. A chat is shown each connection's tools from memory, and the program behind one is started the first time a tool of it is called, then closed after 10 minutes without a call.
Linked Agent Pis (preview): link your Agent Pi with another person's to plan together. You read every message before it goes, and nothing runs on your computer at their request.
Requests (for admins): see what was sent to the model, part by part and by size. It is the third tab of the window behind the status in the corner. Needs the wiretap add-on.
Process limits: how long a process an agent started may sit idle, and how much memory it may use, before it is stopped for you.
🚀 Server auto-start
When enabled, the Agent Pi server will start automatically when you log into your computer. The server registers the login entry itself - no browser extension required.
💾 Conversation cache
Cached conversations open instantly. Old entries are evicted automatically once the cache passes the size cap.
When the cache exceeds this, the least-recently-used conversations are dropped. Default: 1 GB.
1 GB cap (evicts oldest automatically)
⚡ Performance Hardware & memory tuning (restart required)
Allocated to Node.js V8 heap. Higher = faster but uses more RAM. Default: 2048 MB. Set to ~50% of system RAM.
UV thread pool size for async I/O. Default: 4. Match your CPU core count for best performance.
Text-based app doesn't need GPU. Saves ~100MB RAM. Recommended ON.
Removes CSS animations and smooth scrolling. Reduces CPU usage.
Click Change, then press the keys you want. Alt answers either Alt key; pick Right Alt (AltGr) to bind only the right-hand one. Shortcuts work while you type, except combinations that would break text editing - those are refused. Changes apply immediately (no Save needed).
Always available
Part of how the message box and its menus work, so these cannot be changed.
🔐 Subscriptions
Sign in with an account you already have. No API key.⚡ Agent Pi API One balance, every model Coming soon
No provider sign-ups and no keys to manage. Not available yet.
Claude Pro / MaxChecking…
Codex / OpenAI ChatGPT Plus / Pro / Business Checking…
Antigravity / Gemini Plus / Pro / Ultra Checking…
Meta / Muse Muse Code plan Checking…
Qoder Qoder plan Checking…
Kimi Code Coding Plan Checking…
✍️ TheCopywriter.ai Writing
Checking…
🔐 More subscriptions
🔑 Bring your own keys
Enter your personal or organization API keys for direct, unmetered access to providers. Keys are stored encrypted on this machine only.No provider by that name.
🔐 Key storage security
How your API keys are encrypted on this machine. Keys never leave your computer.
Checking…
🖥️ Local Model Runtimes
Everything that runs models on your machine: chat apps through their OpenAI-compatible/v1 address, and the apps that make pictures, video and voice. No account needed.Model apps on this PC. LM Studio and Ollama are found by themselves; llama.cpp, Jan and Unsloth Studio are one click below; any other goes in by its port. Jan and Unsloth Studio ask for their own key. ComfyUI (pictures and video) and VoiceStudio (voice) are listed with the address their menu in the chat uses. Start starts an app that is here, Install opens its download page.
Checking…
🧪 Free Proxy
Free web accounts, used as models.○ Stopped
⚠️ Unofficial. Accounts can be limited or banned, so use throwaway accounts only.
Sign in with
👤 Accounts
Loading accounts…
Gateway · runs in the background
Local Gateway Service /api/free-proxy/gw/v1 Open dashboard
Checking gateway status…
Provider fixes are checked automatically every day.
🖥️ System Tools
Apps, command-line runtimes and system tools available to Agent Pi. Missing a tool? Install it below or click Re-check.
🐚 Shell Tools Shells, compilers and interpreters
Loading detected tools…
⭐ Recommended Tools Local tools and engines for agents
Checking recommended tools…
Storage
Measuring
📊 Usage & Efficiency
Measured across your recent conversations - tokens, prompt-cache hit rate, how often context is compacted, what tool output costs, and flags for anything that looks wrong.
Open this panel to measure.
Agentic Secrets Free
One locked place for your keys and passwords. Your agents use a key without ever seeing it.
From the makers of Agent Pi
On this computer
Checking…
Set it up ?
Downloads ?
Why not just paste the key?
A pasted key can be readA key typed into a chat or saved in a settings file can be read by any agent, end up in a log, or be sent to a model.
Here the agent never sees itThe key stays in a locked store. The agent asks to use it, Agentic Secrets makes the call itself, and only the answer comes back.
You decide who may askEach program has to be allowed by you once, and every use is written down with who asked.
Made for Windows today. The source code is private for now.
TheCopywriter.ai
The cure for generic copy. 58 specialist writing roles, an AI detector and a humanizer, inside your chats.
From the makers of Agent Pi. It is what pays for Agent Pi.
Checking…
WriteSales pages, emails, ads, product pages, SEO and more, each written by a role that does only that.
CheckIs this text AI? The detector is free and needs no account.
FixHumanize a text, or proofread it without changing what it says.
How you get it
1Choose a plan on thecopywriter.ai
2Check outYou are signed in there with your email.
3Press Connect Agent Pi on the thank-you pageAgent Pi opens, asks you once, and you are signed in. Nothing to copy.Soon
Then say it in a chat
Write a product description for a steel water bottle with TheCopywriter. Is this text AI? Check it with TheCopywriter. Humanize this with TheCopywriter.
📡 Integrations
Agent Pi keeps the credentials on this computer, so they never appear in prompts.
🧬
n8n ● checking…
Self-hosted workflow automation
Loading…
Loading…
🔄 Backup & Sync
Back up Agent Pi
One file with your chats, your settings, or both. ?
Locks the file with AES-256. No password: this backup opens only on this computer. Add a password to restore it elsewhere. ?
Automatic keep the newest
Sync between computers
Your chats on each of your computers, kept the same. ?
Chats from your other AI tools, in your History. ?
Only from one folder
Shows only the chats that were made in this folder. Empty shows them all.
🌐 Language
Agent Pi takes your browser's language the first time, and stays in English when it does not know it. Machine-translated: some parts may still be in English.
Now in English. The page reloads when you switch.
👤 What you do
Not chosen yet
🎨 Style
Chat background
Shows at once and is kept on this computer. Every background is drawn by Agent Pi: nothing is downloaded.
🔗 Linked Agent Pis
Loading…
Telegram
Talk to Agent Pi from your phone while you are away: ask what is running and how it is going, have it write a prompt and send it to a chat, start BrowserOS, and be told when a chat finishes. Everything is kept in Agent Pi's own chat.
Agent Pi answers with…
Your phone
Checking…
Commands
/statusWhat is working now: each chat's name, its team or agent, its model, your last message and the last reply./alerts all · long · offWhen a chat finishes, tell me every time, after a minute or more, or never./helpWhat Agent Pi is for, and these commands.anything elseA message to Agent Pi, for example "how is everything going?" or "start BrowserOS".🔌 Agent Pi MCP
Agent Pi MCP lets another AI on this computer, like Claude Code, Codex or Cursor, work with your Agent Pi: see your chats and how they are going, send a chat a message, open or stop a chat, start BrowserOS, Docker or n8n, and message your phone.
Connect it
Claude Code: run this once in a terminal.
…Codex, Cursor and other AIs: add this to their list of MCP servers.
…What it can do
list_chatsYour chats: name, folder, team or agent, model, working or not, last active.progressEverything active lately: what was asked last and the latest answer.read_chatThe last messages of one chat.send_to_chatA message to a chat, as if you typed it. That chat's own agent answers, with its folder, model and permissions.new_chatA new chat, on a folder if you like, with a first message.stop_chatStops a chat that is working.list_servicesWhether BrowserOS, Docker, n8n and Agentic Secrets are installed and running.start_serviceStarts one of them.notify_phoneA message to your phone, through Telegram.Good to know
- It works on this computer only: Agent Pi answers nothing from the internet.
- Agent Pi has to be running.
- A message it sends to a chat is answered by that chat's own agent, with that chat's folder, model and permissions.
📝 file
🤖 Agents
Saved Agents
Loading…
Agent Templates read-only · adding one gives you an editable copy
Loading…
Loading conversation…
Restoring team activity…
🧑🤝🧑 Helpers - 0 of 0 done