Skip to main content

Cron: Scheduled Prompts

AdaL can run prompts on a schedule — every 30 seconds, every 5 minutes, every hour. Set it and forget it. Your scheduled prompt goes through the exact same pipeline as manual input, so @file references, tool usage, and multi-step reasoning all work identically.

Quick Start

# Schedule a prompt to run every 5 minutes
/cron add 5m Check the build logs and report any new errors

# Open the interactive cron manager
/cron

# Stop the running cron
/cron remove

Commands

CommandDescription
/cronOpen interactive dialog — view status, add, or remove
/cron add <interval> <prompt>Schedule a recurring prompt
/cron removeStop and remove the active cron

Interval Format

Use <number><unit>:

UnitMeaningExample
sSeconds30s — every 30 seconds
mMinutes5m — every 5 minutes
hHours1h — every hour

Minimum: 10 seconds · Maximum: 24 hours

Use Cases

🏋️ Training Monitoring

Monitor model training, catch failures early, and handle post-training workflows automatically:

/cron 30m Check the training progress and checkpoint using wandb CLI.
If there's a problem, fix it and resume training. When the validation
checkpoint of epoch 3 has been saved, stop training and download the
checkpoint. After downloading, convert and upload to HF, and evaluate
the checkpoint with eval.sh simultaneously. Report eval results to
my email.

AdaL acts as an autonomous training supervisor — checking progress, intervening on failures, and executing the full post-training pipeline (download → convert → upload → evaluate) when milestones are reached.

🔄 Data Pipeline Monitoring

Run and babysit data processing pipelines with automatic error recovery:

/cron 10m Run the whole data processing pipeline, and monitor it.
Fix and resume until it successfully completes.

Instead of watching terminal output for hours, AdaL monitors the pipeline, diagnoses failures, applies fixes, and restarts — repeating until the full pipeline succeeds.

Build a research bibliography incrementally, one paper at a time:

/cron 5m Read @LITERATURE.md for my problem statement and relevant
research. Search web and Google Scholar to find and add the single
most relevant paper that is not already in the list.

Every 5 minutes, AdaL reads your current literature list, searches for new relevant work, and appends the best find. After an hour, you have ~12 new curated references without lifting a finger.

More Ideas

# Health check a deployed service
/cron 1m curl https://api.example.com/health and report if status != 200

# Watch for CI completion
/cron 30s Check if the GitHub Actions build has completed and report the result

# Periodic git activity summary
/cron 1h Summarize git commits in the last hour and note any large diffs

# Monitor disk usage during a long operation
/cron 5m Check disk usage on /data. If above 90%, identify and suggest files to clean up

Interactive Dialog

Type /cron to open the visual manager:

Level 1 — Status Overview

  • Shows: active/inactive status, interval, prompt preview, run count, skip count, countdown to next fire
  • Actions: Add New or Remove (if active)

Level 2 — Interval Presets (after selecting "Add New")

PresetValueBest For
30 seconds30sQuick check-ins, CI polling
5 minutes5mRegular updates, literature search
30 minutes30mTraining monitoring
1 hour1hHourly summaries
CustomAny interval you want

After selecting a preset, your input is pre-filled with /cron add <interval> — just type your prompt.

How It Works

  • Fires immediately: When you set a cron, it runs the prompt right away, then repeats at the interval.
  • Smart queueing: If AdaL is busy when the timer fires, one execution is queued and runs as soon as AdaL finishes. Additional fires while queued are skipped — no prompt flooding.
  • Fresh @file content: @file references are re-read on every trigger, so you always get the latest file contents.
  • Full pipeline: Cron prompts go through the same processing as manual input — @file expansion, image detection, tool usage, multi-step reasoning.
  • Chat markers: Each cron event (started, finished, queued, skipped) appears as a visual marker in your chat history.

Limitations

  • One cron per session — setting a new one replaces the current one
  • Session-scoped — the cron clears when you exit AdaL
  • Not persisted — cron configurations don't survive restarts (yet)

Tips

  • Start with a longer interval and shorten once you're confident the prompt works as expected.
  • Use @file references for context that changes between runs (logs, configs, data files).
  • Write detailed prompts — cron prompts can be multi-sentence. The more specific you are about success/failure conditions and next steps, the better AdaL handles autonomous operation.
  • Check status with /cron to see run/skip counts and time until next fire.