> ## Documentation Index
> Fetch the complete documentation index at: https://docs.openpawz.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> The open-source AI agent OS — 25,000+ integrations via MCP bridge, multi-agent, multi-channel, fully local, infinitely extensible.

<Note>
  **OpenPawz is not a chatbot.** It's an AI agent operating system that runs on your desktop — with 25,000+ integrations via MCP bridge, 11 chat channels, multi-agent orchestration, and zero cloud dependency.
</Note>

## The AI agent OS you actually own

Built on [Tauri v2](https://v2.tauri.app/) with a Rust backend. While other tools give you a text box and an API bill, OpenPawz gives you a complete agent runtime:

<Columns cols={3}>
  <Card title="25,000+ integrations" icon="puzzle">
    400+ native + 25,000+ community integrations via embedded n8n and MCP Bridge. Agents discover, install, and execute workflows at runtime.
  </Card>

  <Card title="11 channels" icon="message-lines">
    Telegram, Discord, Slack, WhatsApp, Matrix, IRC, Twitch, Mattermost, Nextcloud, Nostr, Webchat — all running simultaneously.
  </Card>

  <Card title="Multi-agent" icon="users">
    Boss/worker delegation, persistent squads, inter-agent messaging, parallel workflows. Your AI army.
  </Card>
</Columns>

<Columns cols={3}>
  <Card title="3-chain DeFi" icon="bitcoin-sign">
    Coinbase + Solana (Jupiter/Pump.fun) + Ethereum (Uniswap). Trading policy engine with guardrails.
  </Card>

  <Card title="Semantic memory" icon="brain">
    BM25 + vector + MMR re-ranking + temporal decay. The most advanced retrieval in any agent platform.
  </Card>

  <Card title="Voice" icon="microphone">
    Three TTS providers + speech-to-text. Agents speak and listen. Auto-speak mode reads responses aloud.
  </Card>
</Columns>

<Columns cols={3}>
  <Card title="Visual editor" icon="diagram-project">
    Drag-and-drop workflow builder. Wire triggers → agents → tools → outputs. No code required.
  </Card>

  <Card title="10 AI providers" icon="server">
    Ollama (local), OpenAI, Anthropic, Google, DeepSeek, Grok, Mistral, OpenRouter, Moonshot, plus custom endpoints.
  </Card>

  <Card title="Self-improving" icon="wand-magic-sparkles">
    Agents search PawzHub and install new skills for themselves — at runtime, conversationally.
  </Card>
</Columns>

***

## Who is it for?

<Columns cols={3}>
  <Card title="Developers" icon="code">
    Full source access, MIT licensed. Build custom tools, integrations, and extensions. Rust + TypeScript stack with Tauri IPC.
  </Card>

  <Card title="Power users" icon="rocket">
    Run a personal AI fleet across every platform you use. Set up once, message your agents from anywhere. No coding after initial setup.
  </Card>

  <Card title="Teams" icon="building">
    Private, auditable AI that never phones home. Multi-agent workflows for research, content, support, trading, DevOps — all on-premise.
  </Card>
</Columns>

***

## How it stacks up

|                            | **OpenPawz**                      | **ChatGPT / Claude** | **AutoGPT / CrewAI** |
| -------------------------- | --------------------------------- | -------------------- | -------------------- |
| **Runs locally**           | 100% — your hardware              | No — cloud only      | CLI / Python lib     |
| **Desktop native**         | Tauri/Rust                        | No                   | No                   |
| **Chat channels**          | 11 simultaneous                   | Web/mobile only      | 0                    |
| **Multi-agent**            | Boss/worker + squads              | Single assistant     | Yes (framework)      |
| **Memory**                 | Hybrid BM25+vector+MMR+decay      | Per-conversation     | None / basic         |
| **Built-in integrations**  | 25,000+ (400 native + MCP bridge) | Limited plugins      | \~15 / code-defined  |
| **DeFi trading**           | 3 chains                          | None                 | None                 |
| **Voice (TTS/STT)**        | 3 providers                       | Voice mode only      | None                 |
| **Extension marketplace**  | 3 tiers + PawzHub                 | None                 | Plugin store / none  |
| **Self-installing skills** | Agents install their own          | No                   | No                   |
| **Session compaction**     | Auto, 40–60% savings              | No                   | No                   |
| **Model routing**          | Auto complexity-based             | N/A                  | No                   |
| **Open source**            | MIT                               | No                   | MIT                  |

***

## The three pillars

<Tabs>
  <Tab title="Memory">
    Your agents **never forget**. Every important fact is automatically captured with vector embeddings and recalled when relevant.

    * **Auto-capture** — agents extract key info from every conversation
    * **Auto-recall** — relevant memories injected into context before the agent responds
    * **Memory Palace** — browse, search, edit everything your agents know
    * **Hybrid retrieval** — BM25 + vector + MMR + temporal decay
  </Tab>

  <Tab title="Token savings">
    **Automatic session compaction** — when conversations near the context limit, older parts are summarized. Key facts preserved in memory. Full session archived. 40–60% token savings on long sessions.
  </Tab>

  <Tab title="Security">
    Zero-trust, local-first — **10 security layers**:

    * **100% local** — no data leaves your machine
    * **OS keychain** — credentials in macOS Keychain / Windows Credential Manager / Linux Secret Service
    * **Risk classification** — every tool rated 1–5. Dangerous ops require confirmation
    * **Container sandbox** — Docker isolation with resource limits
    * **Prompt injection defense** — dual implementation, 30+ detection patterns
    * **TLS certificate pinning** — Mozilla root CAs only, OS trust store excluded
    * **Request signing** — SHA-256 hash of every outbound AI request for tamper detection
    * **Memory zeroing** — API keys auto-zeroed from RAM on provider drop
  </Tab>
</Tabs>

***

## Core philosophy

<Columns cols={2}>
  <Card title="Privacy first" icon="shield-halved">
    All data stays on your machine. Credentials in the OS keychain. No telemetry, no analytics, no phone-home. The only network traffic is to AI providers you explicitly configure.
  </Card>

  <Card title="Agent-native" icon="robot">
    Not a chatbot wrapper — a full agent runtime with tool use, memory, tasks, compaction, trading, voice, and multi-agent delegation built in from day one.
  </Card>

  <Card title="Open and extensible" icon="code-branch">
    MIT licensed. Three extensibility tiers — Skills, Integrations, Extensions — plus MCP and PawzHub. Agents install their own skills at runtime.
  </Card>

  <Card title="Desktop-grade" icon="desktop">
    Tauri v2 for native performance. Rust backend, vanilla TypeScript frontend, SQLite storage. 530 tests. Zero CVEs. macOS, Linux, Windows.
  </Card>
</Columns>

***

## Next steps

<Columns cols={2}>
  <Card title="See it in action" href="/start/showcase" icon="eye">
    Feature highlights and use cases.
  </Card>

  <Card title="Install it" href="/start/installation" icon="download">
    Get OpenPawz running in 5 minutes.
  </Card>
</Columns>