Quickstart: AI inference API for developers

InferenceDirect provides one OpenAI-compatible API endpoint for multiple LLM providers. Swap the base URL in any existing OpenAI SDK client — Python, TypeScript, or plain HTTP. Three steps from sign-up to your first metered call.

Get access

An account administrator adds your email. Sign in with Google and set up two-factor for the portal dashboards.

Get a token

Humans: the device flow on the right — sign in once, receive a token bound to your team. CI and back-ends: ask an admin to issue a service key (Admin → Service keys).

Make a chat request

Any OpenAI client or plain HTTP works. Pick a narrower billing node per request with /n/{node}. Usage, cost, and budget standing land on your dashboard immediately.

# get a personal token (device flow)
curl -X POST https://api.inferencedirect.com/auth/device
#   -> { verification_url, user_code, device_code, ... }
#   visit the URL, enter the code, then poll:
curl -X POST https://api.inferencedirect.com/auth/token \
  -d '{"device_code": "…"}'
#   -> { access_token: $INFERENCEDIRECT_KEY, ... }

# curl
curl https://api.inferencedirect.com/v1/chat/completions \
  -H "Authorization: Bearer $INFERENCEDIRECT_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-ai/DeepSeek-R1",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

# python (openai sdk)
client = OpenAI(
    base_url="https://api.inferencedirect.com/n/ml-platform",
    api_key=os.environ["INFERENCEDIRECT_KEY"],
)

Base URLs

Base URL	Bills to
api.inferencedirect.com	your key's default node
api.inferencedirect.com/n/{node}	the selected division or team (must be within your key's node)

Node selection precedence: URL path, then X-Billing-Bucket header, then the key's own binding. A request can narrow its node, never escape it.

Common responses

401 Not authenticated

Missing or expired key. Errors keep the OpenAI envelope, so SDK retry/backoff behavior works unchanged.

402 Budget exceeded

A cap on your member, team, or department budget is spent and its enforcement is set to block. Ask your admin, or wait for the period to roll.

429 Rate limited

Too many requests. Back off and retry — SDK defaults handle this automatically.