Replace the agent draft version config
Replaces the whole draft version config: harness, instructions, model, skills, tools (the grants the version may call), spaces, network, approvals, delegation and tasks. The body is the FULL config — omitted fields fall back to their defaults, they are not merged with the stored draft. Nothing a run sees changes until the draft is published. Authorization: organization administrators only. A private agent is visible only to its owner; every other caller gets 404. Response: { data: { agent, draft, published } }.
Authentication
Bearer authentication of the form Bearer <token>, where token is your auth token.
Path parameters
Request
Automatic execution keeps the agent loop in the platform and starts isolated compute only when a tool needs it. Default: auto. Legacy native and sandbox values remain accepted for existing integrations.
The standing instructions the agent runs under. A native version must also fit the engine ceiling (8000 chars) to publish.
Saved memory the agent reads and updates across sessions. Private agents only: publishing an org agent with memory enabled answers 422 memory_requires_private. Default: on for sandbox, off otherwise.