Inference API keys authenticate data-plane HTTP calls. They are not the same as console login sessions.

Control-plane OpenAPI (key management, usage queries, etc.) uses separate access tokens (ot_ prefix). See OpenAPI overview. This page covers inference keys (sk-).

How to obtain a key

Create a tenant inference key in the RouterBrain console (UI path depends on your deployment). The plaintext secret is shown once — store it in your secret manager immediately.

OpenAI-compatible surface

Authorization: Bearer YOUR_API_KEY

Required for authenticated /v1/* endpoints (chat, embeddings, files, speech, etc.).

GET /v1/models does not require a key (public catalog).

Anthropic-compatible surface

Either (prefer the first):

x-api-key: YOUR_API_KEY

or

Authorization: Bearer YOUR_API_KEY

Required for /anthropic/v1/messages and /anthropic/v1/models.

Keys and tenancy

  • Each key resolves to a tenant context for model entitlements, balance, IP policy, etc.
  • See Errors for 401/403 semantics.

Do not

Anti-patternDo instead
Embed production keys in front-end codeProxy via BFF / server
Commit keys to GitEnv vars + secret manager
Share one key across unrelated systemsSeparate keys per service/env

See Security best practices and Key lifecycle.

Related