Sites
These routes manage hybrid sites. The site agents themselves never call them: they talk to the site gateway with their client certificates.
| Method and path | Who | What it does |
|---|---|---|
GET /sites | members of the site's organization | Sites, with health, version, capabilities and backlog. |
POST /sites | organization admin | {"id", "name", "description", "tenant_id"}. The id is 3 to 63 characters. |
GET /sites/{site_id} | members of the site's organization | One site, with its tokens and placed projects. |
PATCH /sites/{site_id} | organization admin | {"name", "description", "upload_error_text"}. |
DELETE /sites/{site_id} | organization admin | Remove a site. |
POST /sites/{site_id}/enrollment-tokens | organization admin | {"ttl_hours": 24} (1 to 720). Returns the single-use token once, with the Helm command and the site CA certificate. |
DELETE /sites/{site_id}/enrollment-tokens/{token_id} | organization admin | Revoke a token not used yet. |
POST /sites/{site_id}/rotate | organization admin | Ask the site to rotate its certificate at its next heartbeat. |
POST /sites/{site_id}/revoke | organization admin | Revoke the site, at once and for good. Queued work is cancelled. |
GET /sites/{site_id}/work | members of the site's organization | The site's work items. |
POST /sites/{site_id}/work | project developer or admin; organization admin for work with no project | {"kind", "spec", "project_id"}. See below. |
POST /site-work/{work_id}/cancel | the same | Cancel a work item that hasn't finished. |
GET /sites/{site_id}/events | members of the site's organization | The site's events: enrollment, rotation, work, health. |
GET /projects/{project_id}/placement | viewer | The site a project is placed on, if any. |
PUT /projects/{project_id}/placement | project admin | {"site_id"}. While placed, document uploads, model deploys and code agent deploys for the project are refused in the control plane. |
DELETE /projects/{project_id}/placement | project admin | Remove the placement. |
Work kinds
kind | spec |
|---|---|
rag_ingestion | {"knowledge_base", "path", "chunk_chars", "overlap"}: index files under the site's data folder. path is relative to it. |
tool_execution | {"method", "url", "headers", "body", "credential_ref", "timeout_seconds"}: an HTTP call to a host on the site's allow list. credential_ref names a file in the site's own credentials Secret. |
model_deploy, code_agent_runtime | {"name", "image", "port", "replicas", "args", "env", "resources"}: a workload served in the site's namespace. |
workload_remove | {"name"}: remove a workload the site deployed. |
Results come back redacted: counters, sizes, digests, names and status. A tool result is site://<site>/<work id> with its status, size and digest. The content stays at the site.
Was this page helpful?