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

# agtOS

> Voice-native AI agent platform

# agtOS Documentation

agtOS is an open-source, voice-native AI agent platform. It provides the infrastructure for AI agents that can **hear**, **speak**, **think**, and **act** — connecting voice input/output with LLM reasoning, tool execution, and persistent memory.

<CardGroup cols={2}>
  <Card title="Download the App" icon="desktop" href="/quickstart">
    Install the desktop app and start chatting in minutes. No terminal required.
  </Card>

  <Card title="Developer Setup" icon="terminal" href="/installation#cli-for-developers">
    Clone, install, and run from source with full CLI and API access.
  </Card>

  <Card title="Use Cases" icon="lightbulb" href="/use-cases">
    Personal assistant, smart home hub, developer tool, IoT voice device.
  </Card>

  <Card title="API Reference" icon="code" href="/api-reference/introduction">
    70+ REST endpoints, WebSocket voice protocol, and 10 MCP tools.
  </Card>
</CardGroup>

***

## How it works

```
Voice In → VAD → STT → Model Router → Claude / Ollama → TTS → Voice Out
                              ↓
                     Agent Loop → Tool Registry → MCP Tools
                              ↓
                     Memory (Working + Episodic + Semantic)
```

agtOS processes voice through a **cascade pipeline**: speech is transcribed, routed to the best model (local or cloud), processed through an agent reasoning loop with tool access, and synthesized back to speech — all with sub-second perceived latency.

***

## Core capabilities

<CardGroup cols={3}>
  <Card title="Voice Pipeline" icon="microphone">
    STT → LLM → TTS cascade with barge-in, word timestamps, and streaming partial transcripts. In-process via sherpa-onnx.
  </Card>

  <Card title="Model Routing" icon="route">
    Slot-based routing across Claude, OpenAI, OpenRouter, and Ollama. Intent classification maps requests to the right provider.
  </Card>

  <Card title="Agent Memory" icon="brain">
    Working (session), episodic (Redis), and semantic (vector search) tiers. Dialectic reasoning and cross-tool import.
  </Card>

  <Card title="MCP Integration" icon="plug">
    10 built-in tools + external server connections. Dynamic tool selection reduces context usage by 80-90%.
  </Card>

  <Card title="Multi-Platform" icon="display">
    Web dashboard, Tauri 2 desktop app with system tray, CLI, ESP32 hardware, and browser voice client.
  </Card>

  <Card title="Local-First" icon="shield-check">
    In-process STT/TTS/VAD via ONNX. Local model routing via Ollama. Cloud is optional, not required.
  </Card>
</CardGroup>

***

## Choose your path

<CardGroup cols={3}>
  <Card title="I want to try it" icon="play" href="/quickstart">
    Download the app and start chatting in minutes.
  </Card>

  <Card title="I want to explore features" icon="compass" href="/features/voice-pipeline">
    Voice pipeline, memory, MCP tools, dashboard, and more.
  </Card>

  <Card title="I want to build with it" icon="code" href="/api-reference/introduction">
    REST API, WebSocket protocol, and MCP tool schemas.
  </Card>
</CardGroup>