Models

MushroomAgent connects to LLM providers through the llm section in config.yaml. It supports OpenAI and any OpenAI-compatible API.

Basic config

llm:
  api_type: "openai"
  api_key: "sk-..."
  base_url: "https://api.openai.com/v1"
  model: "gpt-5.2"

The API key can also be set via the environment variable LLM_API_KEY — it will be used if api_key is empty in config.

Key parameters

Field	Default	Description
api_type	—	Set to openai
api_key	—	API key
base_url	—	API endpoint URL
model	—	Model name
temperature	0.0	Sampling temperature (0-2)
max_completion_tokens	4096	Maximum output tokens
stream	false	Enable streaming responses
timeout	600	Request timeout in seconds

OpenAI-compatible providers

Any service that exposes an OpenAI-compatible API works — just change base_url:

Service	base_url
OpenAI	https://api.openai.com/v1
Azure OpenAI	Your Azure endpoint
Ollama (local)	http://localhost:11434/v1
Any OpenAI-compatible service	Provider-specific

To use Anthropic Claude, Google Gemini, or other non-OpenAI models, use a proxy service that provides an OpenAI-compatible adapter (e.g., One API, LiteLLM).

Realtime voice models

For voice mode, a separate realtime_llm section configures the streaming model:

realtime_llm:
  model: "gpt-realtime"
  api_key: "YOUR_REALTIME_API_KEY"        # falls back to llm.api_key

Embedding models

For memory and retrieval, configure the embedding section:

embedding:
  api_type: "openai"
  api_key: ""
  base_url: "https://api.openai.com/v1"
  model: "text-embedding-3-small"
  dimensions: 1536

Supported embedding types: openai (and OpenAI-compatible), dashscope.

Reference

For all available fields, see Configuration Reference.