PocketPaw

PocketPaw is a self-hosted AI agent for people who want an assistant to write code, manage files and act on websites from their own computer or server.

• Local or cloud models: Use Ollama with Llama, Qwen or Mistral for fully offline operation without an account or API key, or connect to Anthropic and OpenAI through supported backends.
• Shared conversations: Chat through the web dashboard, Telegram, Discord, Slack or WhatsApp with the same agent, tools and memory; Signal, Matrix, Microsoft Teams and Google Chat are also supported.
• Multi-agent tasks: Command Center divides a goal into tasks, assigns agents and displays their progress, with an approval pause before work ships.
• Browser and skills: The agent can read and interact with websites through Playwright, while skills add capabilities through plugins.
• Memory and controls: Session history and long-term facts retain context, with optional Mem0 semantic search; encrypted credentials, tool deny-lists, Guardian AI checks on dangerous commands and an audit log provide controls over agent activity.

PocketPaw is free and open source under the MIT license and runs on macOS, Windows and Linux. With Ollama, inference happens on your hardware. Cloud model requests go to the selected provider, and messaging services and web tasks need connectivity. PocketPaw collects no telemetry. The stated minimum for a 7B local model is 4 GB of RAM.