distill-pending

name	distill-pending
description	Process pending transcript distillation
execution	inline

Distill Pending Sessions

Trigger distillation of pending transcripts and check distillation status.

How Distillation Works

1. Automatic: The daemon distills transcripts every 5 minutes (configurable)
2. On Compaction: PreCompact hook triggers immediate distillation before context is lost
3. On Session Start: SessionStart hook triggers distillation of any pending work

Manual Trigger

If you need to manually trigger distillation for the current session:

# Add to queue for daemon processing
echo '{"tool":"distill_trigger","args":{"session_id":"CURRENT_SESSION_ID"},"ts":'$(date +%s)'}' >> /tmp/chitta-queue.jsonl

Check Distillation Status

View registered transcripts and their distillation state:

chitta transcript_list

This shows:

• session_id: Transcript identifier
• last_processed_line: How many lines have been distilled
• distilled: Whether any distillation has occurred
• realm: Project context

Distillation Output

Distilled learnings are stored as memories with:

• [learn] prefix for extracted insights
• SSL format for efficient storage
• Connected via triplets for graph navigation

Configuration

Daemon distillation settings (in chittad --help):

• --distill-interval MINS: Check interval (default: 5)
• --distill-min-turns N: Min turns before distilling (default: 4)
• --distill-model MODEL: LLM model for extraction
• --no-distill: Disable automatic distillation

Troubleshooting

If distillation isn't running:

1. Check daemon is running: chitta health_check
2. Check GPU endpoint is reachable: curl -s http://localhost:11434/v1/models
3. Check transcript is registered: chitta transcript_list
4. Check queue is being processed: ls -la /tmp/chitta-queue.jsonl*