Why Automation Triggers?
Without triggers, agents only respond when someone sends them a message. With triggers, agents can:- Run daily reports at 9 AM
- Monitor conversations for specific topics and react
- Set up their own reminders and scheduled tasks
- Act proactively based on patterns they observe
Three Types of Triggers
1. Scheduled (Cron)
Time-based triggers that fire on a schedule. Schedule types:- At — Fire once at a specific date and time, then auto-delete
- Every — Fire repeatedly at a fixed interval (minimum 60 seconds)
- Cron — Standard cron expressions with optional timezone support (5 or 6 field)
2. Event-Based (Whenhook)
Triggers that evaluate conversation messages against conditions. When a message matches, the trigger fires. Two-layer evaluation: Layer 1 (Prefilter): Fast keyword matching prevents unnecessary LLM calls. Keywords are mandatory — every whenhook must have at least one. Layer 2 (Semantic): An LLM evaluates whether the message truly matches the condition description. The confidence threshold (default: 0.7) controls how certain the LLM must be before firing. Example:- Keywords:
["deploy", "release", "ship"] - Condition:
"User is discussing a production deployment that needs monitoring" - A message saying “we just deployed v2.3 to prod” matches keywords and the LLM confirms it’s about a production deployment → trigger fires
3. Proactive Automations
When enabled, agents can create their own scheduled and event-based automations using built-in tools. This lets agents set reminders, schedule follow-ups, or watch for conditions — all without human configuration. Agent-accessible tools:scheduling_create/scheduling_list/scheduling_get/scheduling_cancel— Manage scheduled automationswhenhook_create/whenhook_list/whenhook_remove— Manage event-based automations
Actions
When a trigger fires, it executes an action:- AgentRun — Sends a prompt to the agent, which processes it as a normal conversation turn
- WorkflowRun — Starts a workflow execution with specified input
Safety and Limits
Automation triggers include multiple layers of protection:Layer 1: Creation Quotas
| Limit | Value |
|---|---|
| Automations per agent | 50 |
| Automations per account | 500 |
Layer 2: Daily Firing Budget
| Limit | Value |
|---|---|
| Firings per agent per day | 100 |
| Firings per account per day | 500 |
Layer 3: Safety Controls
| Control | Value |
|---|---|
| Auto-disable after consecutive failures | 5 |
| Minimum interval for “Every” schedules | 60 seconds |
| Minimum event cooldown | 300 seconds (5 minutes) |
| Default event cooldown | 3,600 seconds (1 hour) |
| Default confidence threshold | 0.7 |
Active Hours
Any automation can be restricted to fire only during specific hours:- Start hour and end hour (0–23)
- IANA timezone (e.g.,
America/New_York) - Automations outside active hours are skipped
Expiration
Automations can have an optionalexpires_at timestamp. After expiration, the automation is automatically disabled.
Execution Flow
Scheduled Triggers
Event Triggers
Event evaluation runs asynchronously after a conversation turn completes — it doesn’t block agent responses.Configuration
Each agent has a TriggerConfig with independent toggles:- Enabled (master switch) — Must be on for any triggers to work
- Scheduled (Cron) — Enable time-based triggers
- Event-Based (Whenhook) — Enable conversation event triggers
- Proactive Automations — Allow the agent to self-create automations
Next: Learn how to configure automation triggers for your agents.