Skip to content

Scheduled Tasks#

Agents don't have to wait for you to start a conversation. Scheduled tasks let you define work that runs automatically — on a timer, on an event, or on a recurring schedule.

How It Works#

A scheduled task is an agent invocation with a trigger. The trigger fires, the agent wakes up, does the work, and posts results to a channel. You check in when it's convenient.

Scheduled tasks are defined in channel configuration or via the Python SDK. They can be as simple as "run this prompt every morning at 8am" or as complex as a multi-step workflow that reads data, runs analysis, and posts a formatted summary.

Trigger Types#

  • Interval — run every N minutes, hours, or days
  • Cron — standard cron expression for precise scheduling
  • Event — trigger on platform events (new data uploaded, training run complete, etc.)
  • Manual — one-shot tasks that run when explicitly invoked

Results Go to Channels#

Scheduled task output posts to the channel it's associated with. Your channel becomes a running log of autonomous work — model performance checks, data quality alerts, overnight training results.

Example: Nightly Model Health Check#

# Define a scheduled task in the channel config
schedule:
  cron: "0 7 * * *"  # Every morning at 7am
  prompt: |
    Check the performance of the production model against yesterday's data.
    Report val_AUC, flag any degradation > 2%, and suggest next steps.
  channel: "#model-monitoring"

The agent wakes at 7am, pulls yesterday's data, runs the evaluation, and posts the result to #model-monitoring. You read it with your coffee.