Agents That Remember
EDDI provides a comprehensive persistent memory system that enables agents to remember facts, preferences, and context across conversations. Memory entries are structured key-value pairs with visibility scoping, giving agents the right context at the right time.
Memory Capabilities
- Persistent User Memory: Key-value entries with
global,agent, andgroupvisibility scoping across conversations - LLM Memory Tools: Built-in tools agents can call to read, write, and search their own persistent memory
- Dream Consolidation: Background memory maintenance: stale entry pruning, contradiction detection, and fact summarization
- Token-Aware Windowing: Intelligent context packing with model-specific tokenizer support and anchored opening steps
- Rolling Summary: Incremental LLM-powered summarization of older turns with a Conversation Recall Tool for drill-back
- Property Extraction: Config-driven slot-filling with
longTerm/conversation/stepscoping - Memory Policy (Commit Flags): Strict write discipline marks failed task output as uncommitted (hidden from LLM context) and injects concise error digests for graceful degradation
- Conversation State: Full history with undo/redo support
How Dream Consolidation Works
Inspired by Anthropic's research on background memory consolidation, EDDI's Dream Cycles run on a configurable schedule to maintain memory quality. Stale entries are pruned, contradictions are detected and resolved, and facts are summarized, all with cost ceilings per run. Dreams are scheduled via cron expressions and can use persistent or fresh conversation strategies.