Long-Term Memory Consolidation
During idle periods, your agents automatically organize and reinforce their memories.
Over time, an agent's memory accumulates hundreds or thousands of facts. Without maintenance, this memory becomes noisy: duplicates, outdated information, incomplete fragments.
Auto-Dream is inspired by how human memory works during sleep. During the agent's idle periods, the plugin triggers a consolidation cycle. Similar memories are merged, contradictions resolved, and obsolete information is archived.
The result is a more compact, more coherent, and more relevant memory. Recalls become more precise, context more reliable, and the agent's overall performance improves over time.
Idle detection
The system identifies periods without interaction (configurable: night, weekends, inactivity threshold).
Inventory
The entire memory is scanned to identify clusters of semantically close memories.
Merging
Redundant or complementary memories are merged into a single, more complete and better-formulated fact.
Conflict resolution
When two memories contradict each other, the most recent one is kept. The older one is archived with an audit trail.
Cleanup
Obsolete memories (superseded by new information, too old without recall) are archived.
Long-running agent
An agent active for 6 months accumulates thousands of facts. Auto-Dream progressively consolidates them, maintaining usable memory even after months of intensive use.
Distributed team
Multiple users interact with the same agent. Information shared by different people on the same topic is automatically reconciled.
Evolving preferences
A user changes their mind about a tool or method. Auto-Dream detects the contradiction and keeps the most recent preference.
featurePages.autoDream.config
- Available for all EasyClaw v2 agents
- Active by default on Max (daily cycle at 3am)
- Requires a memory backend with vector support