Documentation
One key, one balance, every model. The API is OpenAI-compatible for chat and submit-then-poll for media, these pages cover the whole surface. Raw markdown mirrors of every doc live under /llm for agents.
Quickstart
Get a key, point any OpenAI SDK at the CompanyFabric base URL, and make your first metered call.
Authentication
Bearer authentication with sk-cf- keys: creation, budget caps, revocation, and how metering attributes spend per key.
Chat Completions
POST /v1/chat/completions, the OpenAI wire format across every text and code model, with streaming and exact usage. POST /v1/responses serves OpenAI's Responses API on the same models.
Models & IDs
How model ids work: canonical 2-segment ids, branded aliases, and the companyfabric/* meta-models, plus the /v1/models listing endpoint.
Jev Decisions
Decisions on CompanyFabric. POST /v1/decisions, the endpoint for decision models, TypeSafe's Jev, Perplexity's Decider 27B, Microsoft-Decision-1 and Google's Gemini Decision 1: send a state and typed questions (noul, choice, score), get calibrated probabilities back. Same key and balance as chat.
Web Search
Web search on CompanyFabric. POST /v1/search returns ranked results (title, url, snippet, published date) in one shape: DeepSeek on your CompanyFabric balance, Exa and Parallel on your own provider key (BYOK, 0% fee). companyfabric/auto-search picks the engine per query.
Media Jobs
POST /v1/jobs, the async API for video, image, and audio models: submit with the model in the body, poll until completed, read outputs and exact cost.
Errors
Every error is structured JSON with a code, a human message, and a machine-actionable resolution, built for agents that fix their own failures.
BYOK, Bring Your Own Keys
Route through CompanyFabric on your own provider keys at 0% fee, the vault stores them encrypted and never returns more than the last four characters.