Automation and operations
Use these guides to run Agents without an interactive caller and operate the resulting integration. Choose the workflow that matches the outcome you need.
Choose a guide
Run a background Task
Submit immediate asynchronous work, observe its terminal state, and read its transcript.
Schedule recurring work
Configure one-time, interval, or cron-driven Task execution.
Monitor usage and quotas
Measure consumption and maintain monthly request and token headroom.
Troubleshoot a failed integration
Map an observed authentication, Provider, stream, MCP, Sandbox, Task, or quota symptom to a safe action.
Operational model
An executing Task run performs one asynchronous Turn. On-demand submissions enqueue immediately; schedules create runs when their fires are accepted. At most one run for a Task is active, so overlapping fires are skipped. Durable workflow identities and repair scans support recovery, but the platform does not claim exactly-once execution.
Every Turn passes a quota check before execution and records usage after it settles. Use Usage and quotas to inspect those rollups and configure soft monthly ceilings. Use the canonical error contract to branch on stable codes instead of matching messages.
Prerequisites
You need an enabled Agent and a backend TypeScript SDK client. API keys are tenant-scoped and have no per-operation scopes, so the same valid backend credential can access Task and usage operations for its tenant.
Review Tasks and schedules, Usage and quotas, and Limits and reliability before setting production polling, alerting, and recovery policy. Return to the Guides index to choose a different integration outcome.