Schedules, reminders, and webhooks
Nori triggers start or continue Sessions work without requiring a person to send a live chat message. They use the same fleet, agent resolution, integrations, skillsets, lifecycle, and session history as interactive runs.
Trigger types
Section titled “Trigger types”| Type | Use it for | Timing source |
|---|---|---|
| Cron | Recurring maintenance, reports, audits, or checks | Cron expression in the configured timezone |
| Reminder | One task at a future time | A single scheduled timestamp |
| Webhook | Reacting to an external system event | Authenticated inbound HTTP request |
What a trigger stores
Section titled “What a trigger stores”A trigger records the task prompt and scheduling or webhook configuration. It can also select an agent, exact model, and skillset instead of inheriting every fleet default.
At run time, the broker resolves those choices against the current agent registry, model catalog, organization configuration, and available fleet capacity. A removed or disabled agent does not remain launchable just because an older trigger named it.
Direct and chat-oriented runs
Section titled “Direct and chat-oriented runs”Unattended runs receive instructions to make reasonable documented assumptions, finish within the active turn, and return a useful result instead of waiting for a person who is not present.
Chat-oriented trigger paths preserve a more conversational behavior when the run is expected to continue through a communication surface.
Time zones
Section titled “Time zones”Recurring schedules use the organization’s default trigger timezone unless the trigger specifies another supported IANA timezone. Keep the timezone explicit for jobs whose business meaning depends on local time or daylight-saving changes.
Approval and administration
Section titled “Approval and administration”Trigger creation and material changes are admin-gated. A non-admin request can enter a proposed-change flow for approval rather than silently creating an unattended agent.
Use the Triggers screen in the dashboard to review definitions, run state, and recent outcomes.
Failures and recovery
Section titled “Failures and recovery”The trigger executor distinguishes agent errors from transport and machine failures. It can retry bounded transient failures and recover on a replacement machine while preserving the same logical run.
Authentication failures caused by a just-rotated credential receive one credential redistribution and retry. Repeated authentication failure or another terminal agent error marks the run failed.
When a trigger was created from a Slack or Discord channel, Nori stores that origin so it can post a best-effort failure notice. Repeated failures notify on the transition into failure instead of sending the same alert on every run.
Writing a reliable trigger prompt
Section titled “Writing a reliable trigger prompt”An unattended prompt should contain:
- The exact repository, system, or dataset in scope.
- The desired output and where it should be written or reported.
- The verification required before completion.
- The actions that need approval or must remain read-only.
- A fallback instruction for missing data or unavailable integrations.
Example:
Every weekday at 09:00 America/New_York, inspect yesterday's failed CI runs inthe api repository. Group failures by likely root cause, link each run, and posta concise summary. Do not change code or rerun jobs. If CI data is unavailable,report the authentication error and stop.When to use a trigger
Section titled “When to use a trigger”Use a trigger when the starting event is deterministic and the task can produce a useful result without waiting for clarification. Use Slack or web chat when a person expects to steer the work while it runs.
For session behavior after a trigger fires, see What is a Nori session? and the lifecycle reference.
this page as markdown: /sessions/guides/triggers.md· index for agents: /llms.txt