MetaAI

Use the metaai provider for MetaAI Muse Spark models through the Responses API.

Configuration

metaai:
  api_key: "${META_AI_API_KEY}"
  base_url: "https://api.meta.ai/v1"
  web_search:
    enabled: false
    search_context_size: medium  # Optional: low | medium | high
    # user_location:  # Optional approximate locale bias
    #   type: approximate
    #   country: "GB"
    #   region: "London"
    #   city: "London"
    #   timezone: "Europe/London"

Environment variables:

• META_AI_API_KEY: Your MetaAI API key
• META_AI_BASE_URL: Override the API endpoint

Models

fast-agent --model "metaai.muse-spark-1.1"
fast-agent --model "metaai.muse-spark-1.1?web_search=on"

Muse Spark 1.1 supports text, image, video, and PDF input with a 1,048,576 token context window.

Search grounding

MetaAI search grounding is exposed as the same Responses-family web_search tool used elsewhere in fast-agent:

• Config default: metaai.web_search.enabled
• Model string toggle: ?web_search=on|off
• Interactive toggle: /model web_search and the F8 toolbar control

When enabled, fast-agent sends tools=[{"type":"web_search", ...}] and requests include: ["web_search_call.results"] so raw search hits are available alongside url_citation annotations on the answer.

Optional tool fields from config:

• search_context_size: low | medium | high
• user_location: approximate country/region/city/timezone bias

Current models

Model aliases