2026-07-31 — AGI runs and capability checks
Spawn agent runs with a one-time capability token; authorize tools and memory; kill or interrupt runs.
Agent runs
You can start an agent run for yourself (personal) or an organisation you belong to:
POST /v1/api/agi/runs
The response includes a capability token once. Store it; the API does not show it again. Use it when checking what that run may do.
Capability checks
POST /v1/api/agi/authorize checks whether a run may call a tool, touch a memory namespace, or charge a wallet context. Personal runs cannot read organisation memory (and the reverse) unless a bridge grant exists.
Kill and confirm
POST /v1/api/agi/runs/:runId/killstops the run and revokes its token.- Interrupts: create with
POST /v1/api/agi/runs/:runId/interrupts, then approve or deny with.../interrupts/:interruptId/decide.
All of these routes need a signed-in session.