Your AI agent. Modular. Secure. Everywhere.
One-click install
$ curl -fsSL
https://pocketpaw.xyz/install.sh | sh
Most AI agents take thirty minutes to set up. Docker, OAuth, tunneling, fifty config options, all before you send your first message.
Your API keys sit in plain‑text config files. Dashboards get exposed on Shodan. Safety controls can be toggled via API while the agent runs. You install hundreds of packages whether you need them or not. One bad dependency update breaks everything, even features you never asked for.
One install gives you everything you need. Add channels when your life demands them. Never carry what you don't use.
$ curl -fsSL https://pocketpaw.xyz/install.sh | sh # Interactive installer
$ pip install pocketpaw[memory] # Add Mem0 semantic search
$ pip install pocketpaw[all] # Everything, all channels
The full includes:
Telegram, Discord, Slack, WhatsApp, and web dashboard. Same agent, same context, every channel.
Claude Agent SDK, OpenAI, or Ollama for fully local models. $0 API cost with local inference.
Fernet AES with machine-derived PBKDF2 key. Real encryption, not base64 obfuscation.
Secondary LLM reviews dangerous commands before execution. 3-tier tool policy with deny-lists.
Session history and long-term facts. Upgrade to Mem0 semantic vector search when ready.
Playwright-powered with accessibility tree snapshots. Navigate, extract, interact.
pip extras for genuine dependency isolation. Toggle channels from the dashboard. Auto-install, no restart.
Self-installing plugin system. Add capabilities from a registry or build your own. The agent installs what it needs.
Append-only JSONL. Every tool execution logged with severity levels. Tamper-evident by design.
Your AI agent has terminal access, file access, browser control, and your API keys. The security architecture should match that power.
Describe your goal. Command Center breaks it into small meaningful tasks, assigns agents, streams progress live, and pauses for your approval before anything ships.
$ curl -fsSL
https://pocketpaw.xyz/install.sh | sh
A self-hosted AI agent runs on your own computer or server instead of in the cloud. It can answer questions, execute code, browse the web, manage files, and take actions on your behalf. Your data stays on your machine. PocketPaw is an open-source self-hosted AI agent that connects to Telegram, Discord, Slack, WhatsApp, and more.
Yes. PocketPaw works with Ollama, which runs open-source models (Llama, Qwen, Mistral) directly on your hardware. No API keys, no accounts, no usage fees. You need at least 4GB of RAM for a 7B parameter model.
PocketPaw connects to 9+ platforms: Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Microsoft Teams, Google Chat, and a built-in web dashboard. All channels share the same AI agent, tools, and conversation memory.
Completely free and open-source under the MIT license. If you use cloud providers like Anthropic or OpenAI, their standard API pricing applies. For zero-cost operation, use Ollama with local models.
A chatbot generates text responses. An AI agent can take real actions: browsing websites, writing code, managing files, sending emails, and completing multi-step tasks. PocketPaw ships with 50+ built-in tools that let it act on your behalf. Read the full comparison.
Run pip install pocketpaw or use the interactive installer. Then run pocketpaw to start the web dashboard. Takes under 5 minutes. Requires Python 3.11+. See the full installation guide.