Public API.

OpenAI-compatible serverless inference (chat, embeddings, image and video) on deeprelay. Authenticate with a Bearer API key, point at https://api.deeprelay.ai/v1, and you're live.

“OpenAI-compatible” covers the chat request and response shapes, streaming, tool calling and the error envelope, not every OpenAI parameter. Some are rejected with a 400 and a few are accepted and ignored. The compatibility table lists each one.

§ 01.1Get an API key¶

Keys have the shape deeprelay_live_<24 chars> and are shown exactly once at creation. We only store a SHA-256 hash. There is no recovery if you lose the raw key.

Option A

From the dashboard

Sign in, open API Keys in the cloud dashboard, and click Create key. Copy the raw key from the modal before closing it.

Option B

From the CLI

Install the deeprelay CLI and run deeprelay login. The device flow mints a key and writes it to ~/.config/deeprelay/credentials.json with mode 0600.

§ 01.2Hello, world (no auth)¶

The health endpoint needs no API key, so it tells a platform outage apart from a problem with your credentials:

curl https://api.deeprelay.ai/v1/health

Paginated responses are JSON envelopes with a data array and a next_cursor field. See the pagination conventions.

§ 01.3Run your first completion¶

The inference surface is OpenAI-compatible, so any OpenAI client works once you override the base URL. Model ids are canonical deeprelay/<slug> names; list them with GET /v1/models.

curl https://api.deeprelay.ai/v1/chat/completions \
  -H "Authorization: Bearer deeprelay_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deeprelay/deepseek-v4.1-flash",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

§ 01.4Connect a coding agent¶

deeprelay ships an MCP server, so Claude Code, Claude Desktop and Cursor can browse the model catalog, run completions and manage webhooks without leaving the editor. Point your client at it with one environment variable:

SHELL

export DEEPRELAY_API_KEY=deeprelay_live_...

Spend is bounded by a session budget the server quotes and you confirm. See the MCP guide for client configuration and the full tool list.

§ 01.5OpenAPI spec¶

The full OpenAPI 3.1 spec is served live at api.deeprelay.ai/v1/openapi.json. Pipe it into Postman, Stoplight, or any OpenAPI tool to explore interactively.