System Prompts of Personal AI Agents
Agent Prompts collects the full, verbatim system prompts of 17 personal AI agents: 6 commercial products (Muse, dots, Claude Cowork (+ Dispatch), Poke, Limitless Pendant, Rabbit r1) and 11 open-source projects (OpenMinis (Open Minis), OpenMuse (CopilotKit), nanoMuse, OpenMuse (OpenMuseAgent, archived), OpenClaw, Hermes Agent, Omi, Noa (Brilliant Labs Frame), 01 (Open Interpreter), Leon, Khoj). Every prompt links to its source at a pinned commit and is labeled official/open-source or community-leaked (unverified). Last updated: .
Commercial agents
Muse
Meta's personal AI agent that runs on the Muse Secure VM and chats with you in the Muse app and WhatsApp.
dots
The always-on personal agent in ChatGPT, with its own cloud computer. The leaked collection only contains the sub-agent, tool and voice parts; the main dot prompt has not been seen.
Claude Cowork (+ Dispatch)
Anthropic's desktop/cloud agent for non-coding work; Dispatch is its orchestrator.
Poke
A personal assistant you talk to over iMessage, WhatsApp or SMS; it handles email, calendar and reminders for you and dispatches background execution agents to finish tasks.
Limitless Pendant
A wearable pendant that continuously records your life as a lifelog; AI helps you recall and search past conversations.
Rabbit r1
A pocket AI device (Pocket Companion) that can hail rides, order food, play music, take notes and more by voice.
Open-source agents
OpenMinis (Open Minis)
An open-source, on-device AI agent for iOS/Android with a built-in Alpine Linux shell. The prompt is assembled at runtime in the source code.
OpenMuse (CopilotKit)
CopilotKit's open-source personal agent template (browser, terminal, files, long-running tasks).
nanoMuse
An open-source personal agent modeled on Meta Muse, spanning phone, desktop and browser; the successor of OpenMuse (OpenMuseAgent).
OpenMuse (OpenMuseAgent, archived)
A self-hosted personal agent, archived at versions 0.1.0–0.6.0 and migrated to nanoMuse.
OpenClaw
A self-hosted personal assistant that connects to WhatsApp, Telegram, iMessage and many other channels.
Hermes Agent
Nous Research's open-source, self-improving personal agent.
Omi
An open-source AI wearable (pendant / omiGlass glasses, formerly Friend) that continuously records conversations and acts as a personal assistant and mentor that answers questions and gives advice.
Noa (Brilliant Labs Frame)
The personal AI assistant in Brilliant Labs Frame AR glasses; it sees what you see, answers questions and translates.
01 (Open Interpreter)
An open-source voice AI device (01 Light) and platform that acts as a voice executive assistant, completing tasks on your computer.
Leon
An open-source, self-hostable personal AI assistant with its own personality, moods and memory.
Khoj
An open-source 'second brain' personal assistant that answers, researches and runs scheduled tasks based on your notes and documents.
FAQ
What is Agent Prompts?
Agent Prompts is a collection of the system prompts of personal AI agents. It currently covers 17 agents (6 commercial products and 11 open-source projects), each with source links.
Where do the prompts come from? Are they reliable?
Open-source prompts are taken directly from the projects' GitHub repositories at a pinned commit, with file and line numbers. Commercial prompts come from public community leak repositories (such as asgeirtj/system_prompts_leaks); they are labeled 'community-leaked (unverified)' and are not confirmed by the vendors.
What does 'extracted from source code' mean?
Some open-source projects build their prompts in code. Only the text actually sent to the model is kept and code syntax is removed; values filled in at runtime are written as {placeholders} and conditionally added parts are marked with [bracketed notes]. The wording is kept as in the original, not rewritten or translated. Section headings and bracketed notes in extracted prompts are editorial annotations written in Chinese.
What is included?
Only personal agents (assistants that handle your everyday tasks, memory and errands, including wearable/hardware devices) whose prompt text is publicly available. Browser agents and products without a public prompt are not included. Secrets and personal data are replaced with [REDACTED].
How can I download or cite a prompt?
Every agent has a static page at /en/agents/<id>/ and a Markdown version at /agents/<id>.md; all prompts are in /llms-full.txt and structured data is in /data.json. When citing, also credit the original source repository; copyright remains with the original authors.
How often is it updated?
The data was last updated on 2026-10-26.