Broker VM (On-Appliance) overview
Programmatic management of a single Cortex Broker VM directly from the appliance, served by the broker appliance's on-board web service. These endpoints are reachable on the appliance's HTTPS listener at https://{broker_host}/public_api/v1/... and operate against the local broker only (no tenant orchestration). This API is also referred to as the direct Broker API.
This surface complements the tenant-side Broker Public API (see the Broker VM (Tenant-Side) API reference), which is served by the Cortex tenant backend and orchestrates brokers remotely. The two surfaces share the /public_api/v1/ prefix but live on different hostnames and use different authentication.
Audiences
First-time bootstrap — a fresh broker ships with a factory default admin password. The on-prem operator resets it, generates a 10-minute Bearer token, and uses the token to register the broker against the Cortex tenant.
Day-2 on-appliance configuration — install custom SSL or CA certificates, configure NIC / NTP / proxy, reconfigure the internal Docker subnet.
Local diagnostics — download a synchronous log bundle from the broker without going through the tenant.
Authentication
Two security schemes are used:
Admin password — present in the request body for the two bootstrap endpoints (reset initial password, generate token). The broker authenticates by comparing the supplied password to the stored admin password hash.
Public API token — a short-lived (10-minute TTL) opaque Bearer token issued by the generate-token endpoint. Pass it as
Authorization: Bearer <token>to every other endpoint in this spec.
Authorization is the only authentication header used by this on-appliance API. There is no separate key-identifier header (unlike the tenant-side API).
Initial-password gate
Every endpoint except the reset-initial-password endpoint is blocked until the factory password has been changed. Calls to a non-exempt endpoint before that step return 403 Forbidden.
Response envelope
All endpoints follow a single envelope:
Success (200 OK):
{ "reply": <payload-or-null> }(replyomitted on endpoints that have no body content).Error (4xx / 5xx):
{ "error": "<message>", "reply": null }.
One endpoint streams application/octet-stream directly instead of the JSON envelope.
Rate limiting
Bootstrap endpoints carry a 5/min/IP cap, and the six on-appliance network-configuration endpoints carry a 1/sec/IP cap. Exceeding the cap returns 429 Too Many Requests. Some endpoints also surface 503 Service Unavailable with a Retry-After header when a prior single-writer operation is still in flight.
Operational warnings
Several network-configuration endpoints are Operationally Dangerous — a bad configuration can leave the appliance unreachable and require console recovery. Treat these as fire-and-poll, not fire-and-expect-immediate-200.
Last updated
Was this helpful?
