Approvals

Approval strategies let you control how much oversight you want over what the ToolRegistry AI agent does.

• Approval Strategies

  • NONE — No prompts; executes immediately.

  • ON_START — Approve the entire plan once before any tool calls. Use this when you want a human end user to review and confirm the plan before execution begins. This gives oversight without interrupting each individual tool call.

  • ON_TOOL_CALL — Approve each external call (fine-grained). Since this can be noisy, it’s recommended to use only when you want your AI agent to approve every tool call.

• Use the one-time token link provided in the server-sent stream to give approvals:

GET /v1/runs/approval?token=...