Scheduled Routines
Tool set: schedulesDefault: Disabled
Create reliable recurring routines or an exact one-time run. Every occurrence has a stable execution record and idempotency key.
Tools
| Tool | Description |
|---|---|
| create_schedule | Create a routine with validated timing and action config |
| list_schedules | List timing, timezone review state, status, next run, and run count |
| update_schedule | Update a routine and recompute its next occurrence |
| pause_schedule | Pause future occurrences |
| resume_schedule | Resume a paused schedule |
| delete_schedule | Delete an idle routine |
Timing Modes
| Type | Contract | Example |
|---|---|---|
| Interval | Minimum one minute; anchored and missed runs coalesce | "every 5m" |
| Cron | Five fields plus a required IANA timezone; DST-aware | "0 9 * * 1" in "Asia/Manila" |
| Once | Required future ISO timestamp with Z or an explicit offset | "2026-08-12T09:00:00+08:00" |
Action Types
| Action | Config |
|---|---|
| run_prompt | { prompt: "Summarize today's tasks" } |
| send_email | { to, subject, body } |
| create_task | { tabId?, title, description?, status?, priority? } |
| fire_webhook | { url, payload } |
Email and webhook actions use the agent safety policy for the schedule context. They require owner approval by default; approved direct actions resume from the captured action snapshot. Outbound tools inside a prompt run are blocked when approval would be required because they have no safe resume path.
Reliability
Convex schedules the exact occurrence. Transactional claims prevent duplicate execution, prompt runs settle only after their agent job completes, and a bounded watchdog fails only an occurrence stranded in the running state.
Limits
| Limit | Value |
|---|---|
| Routines per agent | 20 |
| Minimum interval | 1 minute |
| Maximum interval | 365 days |
| Maximum run limit | 10,000 |