AI API Quickstart & Authentication

Use Bearer API key authentication with the OpenAI-compatible Unbound API. Discover model IDs and send your first text request.

Canonical production origin: https://unbound-me.com. This response is served by the production environment. Check /public/runtime for registration and demo-payment availability. Check deployment status before running examples; documentation is not an uptime guarantee.

Base URL: https://unbound-me.com/v1

For a free connection check without a real key, use the public test API. Its replies are simulated and never billed.

Use a customer inference key from Dashboard → API keys. Authentication: Authorization: Bearer $UNBOUND_API_KEY. Clients may also send the full inference key through the x-api-key header (including Anthropic-compatible Messages clients). Do not combine different credentials. Key IDs, MCP connection tokens and master administrator credentials are not inference credentials. Never send an API key in a URL.

Discover models

GET /v1/models requires an inference key. Public names and current USD rates: model catalog.

Text request

curl 'https://unbound-me.com/v1/chat/completions' \
  -H "Authorization: Bearer $UNBOUND_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"glm-5.3-753b-unbound","messages":[{"role":"user","content":"Explain how to review my own application for security issues."}],"max_tokens":128}'

Text endpoints: POST /v1/chat/completions, /v1/responses, /v1/completions and /v1/messages. Verify each endpoint payload format; OpenAI compatibility does not mean every OpenAI feature or endpoint is implemented.

POST /v1/quote gives a read-only estimate; it reserves nothing. GET /v1/billing returns a key's billing information. Requests recheck available funds before acceptance. A quote is not a price-lock or token reservation.

Errors use an error object with message, type, code and param, plus request_id. Responses include X-Request-ID; rate-limited responses include Retry-After. Malformed JSON returns 400, oversized bodies 413, unsupported content type 415, and provider timeouts 504. A proxy may return a non-JSON error before the application; inspect HTTP status and CF-Ray in that case.

401 means invalid/missing inference credentials; 403 can mean blocked or unauthorized key; 402 means the spending allowance cannot cover the request; 429 means a rate limit. Do not retry financial mutations blindly.

Official partner

Unbound Umbris

We’re proud to partner with Umbris. Visit their site and explore what they’re building.

Discover Umbris