Self-host pocketpaw
v0.4.18Multi-channel AI agent you run on your own machine
curl -fsSL https://pocketpaw.xyz/install.sh | sh- Stars
- 894
- RAM
- at least 4GB of RAM for a 7B parameter model
- Port
- 8888 (Web dashboard / API server)
What is pocketpaw?
PocketPaw puts an AI assistant on hardware you own instead of in someone else's cloud. You talk to it through a built-in browser dashboard or through adapters for the messaging services you already use, and every surface reaches the same agent with the same conversation memory. Under the hood an event-driven message bus feeds a configurable agent loop, and six interchangeable engines all implement one protocol, so the LLM stack can be switched without touching anything else. Ollama support makes fully local operation possible with no API cost. Capability comes from 50-plus built-in tools covering Playwright browsing, web search, image generation, speech in and out, OCR, research and delegation, plus Google Workspace apps, Spotify, Reddit and MCP servers. Because such an agent can reach your shell, files and API keys, protection is built into the pipeline: a second Guardian AI model vets risky commands, prompt-injection scanning runs on input, tool access is governed by three-tier profiles where deny lists always win, plan mode demands human sign-off, credentials are encrypted at rest, and each tool execution lands in an append-only log. Installation is a pip command, the repo's compose stack, a one-line shell script, or native bundles for Windows, macOS and Linux; the dashboard listens on port 8888. It is MIT licensed with no telemetry or paid tier, but the project is still marked beta with breaking changes between versions, and shell and browser tooling on Windows works best under WSL2.
Should you self-host pocketpaw?
pocketpaw scores 48/100 on selfhost.directory (how scoring works) and ranks #52 of 120 in AI Assistant. It has 894 GitHub stars.
It shipped 10 stable releases in the past year. The latest, v0.4.18, came out on 29 April 2026. New versions land roughly every 3 days.
Setup is easy: most people get it running in one sitting. It is MIT-licensed and written mostly in Python.
Its closest alternative on selfhost.directory is hermes-agent (89/100, 257 957 stars). It scores higher overall, so compare the two before you pick. See pocketpaw vs hermes-agent side by side.
Features
- Append-only audit log
- Browser AutomationBrowser automation and shell tools work best under WSL2 on native Windows.
- Built-in agent tools
- Guardian AI tool-call review
- Local model / offline operation
- MCP server support
- Multi-agent Command Center
- Multi-channel chat (Telegram, Discord, Slack, WhatsApp and more)
- Native desktop app
- Persistent memory
- Plan mode with human approval
- Prompt-injection scanning
- Swappable LLM agent backends
- Tool policy profiles
- Web Dashboard
pocketpaw README
FAQ
How do I install PocketPaw?
Run `pip install pocketpaw` or use the interactive installer, then launch `pocketpaw` to start the web dashboard; it needs Python 3.11+.
What's the difference between an AI agent and a chatbot?
A chatbot only generates text, while an agent takes actions such as browsing, writing code, managing files or sending email; PocketPaw ships 50+ built-in tools for that.
Is PocketPaw free?
It is free and MIT licensed; only the cloud provider's own API usage costs apply, and Ollama with local models gives zero-cost operation.
What messaging platforms does PocketPaw support?
Nine-plus channels: Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Microsoft Teams, Google Chat and a built-in web dashboard, all sharing one agent and memory.
Can I run PocketPaw without an API key?
Yes — pair it with Ollama to run open-source models such as Llama, Qwen or Mistral on your own hardware, with no API keys or usage fees.
What is a self-hosted AI agent?
A self-hosted AI agent runs on your own computer or server rather than in someone else's cloud, and can answer questions, run code, browse the web, manage files and act on your behalf.
More in Instant Messaging
Prosody IMProsody is a modern, lightweight, and extensible XMPP/Jabber server written in Lua, designed for easy setup, efficient resource usage, and federation with other XMPP services.
RevoltA self-hosted, open-source group chat app for communities offering servers, channels, calls, file uploads, and themes — a Discord alternative you fully control.
Rocket.ChatSelf-hosted, open-source chat platform for teams and citizens: channels, threads, voice and video, end-to-end encryption, LDAP/SSO, and Matrix federation.
sharkordOpen-source, self-hosted chat with voice, video, and screen sharing for small groups. Runs as a single binary or Docker container, no external database needed.
Signal-APISelf-hosted, open-source Docker container that wraps signal-cli as a REST API, so scripts can register a Signal number and send, receive and manage messages.
SimpleX ChatOpen-source messenger without user IDs: E2E encrypted chats, groups, calls and files, with mobile, desktop and terminal clients and self-hosted relay servers.
SoketiOpen-source WebSocket server that speaks Pusher Protocol v7, self-hosted via Docker, npm or Helm, with app credentials in arrays, MySQL, Postgres or DynamoDB.
SpacebarSpacebar Server is a self-hosted, open-source Discord backend reimplementation (AGPL-3.0) with API, gateway, CDN, and experimental WebRTC voice.
Spectrum 2Spectrum 2 is a self-hosted, open-source XMPP transport gateway letting XMPP users chat with contacts on IRC, Slack, Facebook, MSN, Yahoo, and other legacy IM networks.
StoatSelf-hosted, open-source deployment of the Stoat group chat platform: back-end, web client, file server, voice and image proxies run as one Docker Compose stack.
TailchatSelf-hosted open-source IM platform with plugin system, microservices, panels, RBAC, bots, OAuth, voice/video calls.
The LoungeA self-hosted web IRC client that keeps you connected to chat networks even when offline, featuring push alerts, link previews, and multi-user accounts.
0 comments