Cron Jobs

Automated memory maintenance — daily transitions, stale pruning, and weekly consolidation

The MemoryCronService (apps/api/src/memory/memory-cron.service.ts) runs scheduled maintenance tasks using @nestjs/schedule.

Schedule Overview

Job	Schedule	What it does
Daily file transitions	`0 3 * * *` (03:00 UTC daily)	Move old daily files: active → warm → archived
Stale memory pruning	`0 4 * * *` (04:00 UTC daily)	Mark inactive task memories as completed
Memory consolidation	`0 5 * * 0` (05:00 UTC Sundays)	Cluster and merge similar memories

Daily File Transitions (03:00 UTC)

Moves daily memory files through the pipeline/lifecycle status chain:

active (0-7 days) → warm (7-30 days) → archived (30+ days)

Logic:

Files older than 7 days: active → warm
Files older than 30 days: warm → archived

Only active files are included in pipeline/context-assembly.

Stale Memory Pruning (04:00 UTC)

Targets task-type agent memories that are likely outdated:

UPDATE agent_memories
SET status = 'completed'
WHERE memory_type = 'task'
  AND status = 'active'
  AND created_at < now() - interval '7 days'
  AND last_accessed_at < now() - interval '7 days'

Tasks that haven't been created or accessed in 7 days are assumed complete.

Weekly Consolidation (Sundays 05:00 UTC)

The most complex maintenance job:

Find candidates — Agents with >20 active memories
Cluster — Group memories with cosine similarity ≥0.80
Filter — Only clusters of 3+ memories
Consolidate — LLM summarizes each cluster into one memory (see pipeline/extraction)
Replace — Old memories marked consolidated, new memory inserted with boosted importance

The consolidated memory gets importance = min(max_cluster_importance + 0.1, 1.0).

Startup Self-Healing

The service implements OnModuleInit to run catch-up tasks on server startup:

async onModuleInit() {
  // Run daily transitions (catch missed 03:00 runs)
  await this.dailyFileService.transitionDailyFiles();
  // Run stale pruning (catch missed 04:00 runs)
  await this.memoryService.pruneStaleMemories();
}

This handles cases where the server was down during scheduled cron times (e.g., deployments, crashes).

pipeline/lifecycle — Status transitions managed by these crons
pipeline/scoring — Similarity thresholds used in consolidation
pipeline/extraction — consolidateMemories() LLM call
operations/configuration — Threshold values and intervals