Save, retrieve, and assemble context from a knowledge graph of memories. Semantic search meets graph traversal — under a token budget your LLM can actually use.
Working, long-term, and archived tiers with automatic promotion and decay. Memories consolidate and reinforce over time.
Entities and relationships extracted from memories. Graph traversal augments vector search for deeper context assembly.
Blend semantic similarity, importance, recency, access frequency, and graph proximity into a single ranking score.
Build LLM prompts from retrieved memories under a token budget. Ready-to-inject context blocks, not raw dumps.
Store facts with importance, category, entities, and tags. Near-duplicates are reinforced, related memories consolidated.
Vector search + entity-name matching + graph traversal. Composite scoring ranks results by relevance.
Token-budgeted context blocks ready for your LLM prompt. Includes source attribution and graph provenance.