Scheduled Tasks
Purpose
Allow users to schedule recurring or one-time data analysis tasks that execute automatically, producing results as read-only conversations, email reports, or Telegram messages.
Key Principles
- Reuse Existing Infrastructure - Leverage current worker pool, tool registry, and conversation system
- Security First - Follow exact same permission flow as interactive conversations
- Cost Efficient - Default to cheap models, track token usage naturally
- User-Friendly - Results appear as read-only conversations in history
- Fail-Safe - Auto-disable problematic tasks, prevent runaway costs
Example Use Cases
"Every Monday at 9am, summarize last week's sales by region and email to team@company.com"
"Check customer count every hour - if below 1000, email alert to ops@company.com"
"First day of each month at 8am, export transaction data to CSV and email to finance@company.com"
"Daily at 2am, run data profiling on customers table (results in conversation)"
"From Telegram: every weekday at 8am, send me yesterday's signups here"
"From Telegram: monitor error rate every 5 minutes; alert me if it goes above 1% (cooldown 30 min)"
Output channels
- Conversation — result lands as a new turn in a read-only conversation in your history.
- Email — HTML email to one or more recipients (no QRY account required).
- Telegram — bot message to your chat. Auto-resolved when the task is created from the Telegram bot. Conditional alerts (fire only on a condition + cooldown) live here.
See the output channels guide for which channel fits which use case.
Last updated: May 07, 2026