API Access
Overview
How applications, scripts and tools authenticate to Multi-Router, and how an API key decides what a caller can reach.
An API key is the credential an application, script or tool uses to call Multi-Router. It identifies the caller and decides what the caller can reach: which endpoints, which routers, and which resources it can read or change. A key belongs to one organization and reaches only that organization's routers, prompts and other resources. Applications use one API key for every request, while the routing, models, prompts and provider credentials stay on the routers the key calls.
API keys and routers
An API key and a router do different jobs. The key says who is calling and what they may do; the router says how a request is served: which model and provider, which prompt, and what Traffic records. Changing a router's model or prompt changes the behavior for every key that calls it, and issuing a new key gives a new caller access with no change to any router. See Routing.
The API address
The Multi-Router API is served at https://api.multi-router.ai, with its endpoints under /v1, such as POST /v1/chat/completions and GET /v1/routers. The same address also serves the Claude Proxy's Bedrock and Vertex endpoints and the Gemini Proxy's Gemini endpoints, which accept the same keys. Each request carries its key in the Authorization header:
Claude Code, Codex CLI and Gemini CLI use the same key through a Claude Proxy, Codex Proxy or Gemini Proxy, and some of them send it in a different header. See Authentication.
What a key can reach
Three settings on each key decide what it can reach:
| Setting | Decides | Page |
|---|---|---|
| Scopes | Which endpoints the key can call, such as inference only, or reading and changing routers. | Scope |
| Routers | Which routers the key can send inference requests to: any router, or a chosen list. | Router Access |
| Expiry | The date after which the key stops working. | API Key |
Every key also has a rate limit on inference requests. See Rate Limit.
Kinds of credentials
Multi-Router accepts three kinds of credentials for API calls, each created in its own place:
| Credential | Created by | Used for |
|---|---|---|
| API key | A person on the API Keys page of the dashboard. | Applications, scripts and tools, with the scopes and routers chosen for the key. |
| Managed key | Self-Service Provisioning, when a person sets up a tool from the Portal. | That person's tool, on their own router. |
| Machine identity | A workload's own cloud identity, such as an AWS role or a Kubernetes service account, matched to a machine binding. | Workloads that call Multi-Router with their cloud identity in place of a stored key, with inference access. |
See Self-Service Provisioning and Machine Identity.
Security
API keys are stored encrypted and checked on every request. Provider keys stay inside Multi-Router, out of callers' reach. A disabled or expired key is refused, and a change to a key, such as disabling it or changing its scopes, takes effect within about a minute. Scoped keys reach only the endpoints their scopes name, which keeps endpoints added in the future closed to them, and keys limited to specific routers receive a "not found" response for every router outside their list, which keeps the rest of the organization private.
Related topics
The other pages in API Access cover each part of a key in detail:
| Page | Covers |
|---|---|
| API Key | Creating, copying, editing, disabling and deleting keys. |
| Scope | Full control, scoped keys, and every scope and the endpoints it covers. |
| Router Access | Limiting a key to specific routers and aliases. |
| Authentication | The headers that carry a key, and authentication errors. |
| Rate Limit | Request limits per key and the 429 response. |