Run ensembles from your own code

The AI Crucible API runs the same multi-model strategies as the app: several models work the question and one answer comes back. API access is included in the Pro plan.

Let Autopilot choose

Set model to crucible/auto and Autopilot picks one model or an ensemble, the strategy, the models and the judge for each prompt. You set only Speed (fast, balanced, thorough) and Cost (lean, standard, max), in ai_crucible or the x-crucible-speed and x-crucible-cost headers. The response reports the plan it ran in ai_crucible.plan.

Quickstart

curl https://ai-crucible.com/v1/chat/completions \
  -H "Authorization: Bearer $AI_CRUCIBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"crucible/auto","messages":[{"role":"user","content":"Should a five-person startup move from Heroku to AWS this year?"}],"ai_crucible":{"speed":"balanced","cost":"standard"}}'

Endpoints

• POST /v1/chat/completions: Create a chat completion using an ensemble strategy
• POST /v1/responses: Create a response from a single input string
• POST /v1/evaluations: Evaluate model responses with AI judges
• GET /v1/usage: Get usage statistics

Resources

• OpenAPI spec: Every endpoint, request and response schema, for code generators and API tools.
• Postman collection: Import, set baseUrl and apiKey, and send your first request.
• Integration guide: Walkthrough of the endpoints, strategies, streaming and usage tracking.
• Sample app: A small chat app that calls the API, to copy from.
• Bring your own OpenRouter key: In the app, the Connect plan runs ensembles on your own OpenRouter account. The API itself needs Pro.
• MCP tools: Give ensemble models access to your tools through the Model Context Protocol.