Skip to main content
These capabilities exist on the platform but are service-to-service and are not wrapped by the SDK, to keep the client surface safe:
  • Brain /v1/internal/* — workflow execution, generation, LLM/image/video/voice egress. Requires the shared internal service key.
  • Render service — stage, still, and sequence rendering, simulation, audio extraction. Requires a service token.
  • File write/sync transport (POST/DELETE/PATCH on workspace files) and reconcile/publish — these mutate R2 and the manifest and are reserved for authorized tooling.

​
The supported client path for generation

If you need generation or rendering from a client, run it through an agent (project) on the platform, and read the result back via the brain feed and files surface:
  1. The agent’s workflow generates content on the platform (billed to the workspace).
  2. Publish promotes it to the frozen release.
  3. Your client reads it: brain.getFeed, brain.getItems, brain.streamR2File.
There is no client-side “run a render” verb, by design.

​
Reading data you do not own

A workspace key reads exactly one workspace’s published data. There is no cross-workspace read, no listing of other workspaces, and no member data on the brain surface. To read another account’s data, that account mints a key and hands it to you.