rem-sleep

To consolidate AI agent session logs into long-term memory, you scan recent sessions to extract decisions, facts, and lessons, then update MEMORY.md. This prevents forgetting and drift by turning raw session logs into durable knowledge during end-of-session wrap-ups.

The best way to defragment memory files for an AI agent is to prune stale entries, merge duplicates, and compress the files for efficiency. This defragmentation process keeps long-term knowledge fresh and actionable during heartbeat maintenance.

Your AI agent forgets previous decisions after a restart because raw session logs are not consolidated into long-term memory. Applying end-of-session memory consolidation extracts decisions and updates MEMORY.md to prevent drift and knowledge loss.

You can run memory consolidation periodically using a heartbeat for AI agents to maintain long-term knowledge. Heartbeat maintenance scans recent sessions, extracts lessons, updates MEMORY.md, and performs defragmentation with safe fallbacks.

Memory consolidation does not work without existing session logs because it requires raw session data to search and extract decisions, facts, and lessons. The workflow depends on session logs to update MEMORY.md and perform defragmentation.