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

# LLM Routing

> AI-based smart routing via any OpenAI-compatible API

BotMux can use an LLM to make intelligent routing decisions based on message content.

## How It Works

1. A message arrives at a bot
2. Rule-based routes are checked first
3. If no rule matches, the LLM receives the message text, sender info, and a list of all bots with their descriptions
4. The LLM returns a routing decision: `{target_bot_id, target_chat_id, action, reason}`

## Supported APIs

Any **OpenAI-compatible** endpoint works:

* OpenAI
* Ollama
* LM Studio
* Any compatible API

## Configuration

Configure via the web UI at **LLM Config**:

| Field         | Description                                |
| ------------- | ------------------------------------------ |
| API URL       | OpenAI-compatible endpoint URL             |
| API Key       | Authentication key                         |
| Model         | Model name (e.g., `gpt-4o`, `llama3`)      |
| System Prompt | Custom system prompt for routing decisions |
| Enabled       | Toggle on/off                              |

## Bot Descriptions

Each bot can have a **description** explaining what it does. These descriptions are sent to the LLM to help it make better routing decisions. Set descriptions via the web UI or API (`/api/bots/description`).

## Reverse Routing

Source-NAT return path works automatically for LLM-routed messages, just like rule-based routes.