Loading...
Beyond MemGPT's two-tier model: hybrid memory architectures that fuse vector, graph, and lexical retrieval; the Mem0 architecture for production-scale agent memory; Voyager-style skill libraries that turn execution traces into reusable code.
MemGPT solved the capacity problem — how to store more than fits in the context window. Memory II is about the quality problem — when you have memory, how do you find the right piece, and how do you turn experience into reusable knowledge?
Two families of techniques:
This chapter covers:
Click any topic to jump in
Vector + graph + lexical retrieval, fused via RRF. Each mode wins different queries.
Declarative facts (Mem0) and procedural skills (Voyager)
Production memory framework: atomic facts, relationships, summaries — hybrid retrieval at all three levels.
Persistent procedural memory — turn successful execution traces into named, retrievable skills.
When to learn a new skill, when to use an old one. The library-as-codebase discipline.
Recall@k, consistency, latency, and the 'did memory help?' test that matters most.
Vector retrieval was the default for agent memory in 2023 — embed every memory, search by cosine similarity. It works for many cases but misses three: (a) facts that share entities but use different words, (b) exact matches buried under semantically similar noise, and (c) chains of relationships that no single embedding captures. Hybrid memory combines vector, graph, and lexical (BM25-style) retrieval and learns to weight them per query.
Vector retrieval. Embeds memory and query into a shared dense space; returns top- by cosine similarity. Wins when: the user's question and the relevant memory share concepts but not words ('the doctor I saw last month' → memory about Dr. Lee). Loses when: the user wants an exact identifier ('order #1234') — a vector store may return 'similar' orders rather than the exact match.
Graph retrieval. Stores memory as a knowledge graph: entities (people, projects, topics) connected by typed relationships ('works_with', 'mentioned_in'). Traverses from a query entity outward by relation. Wins when: the question is multi-hop ('what did Alice say about Bob's project?'). Loses when: the query is unstructured natural language with no clear entities.
Lexical retrieval (BM25 / keyword). Classic IR — score by term frequency × inverse document frequency. Wins when: the user uses specific keywords (acronyms, identifiers, proper nouns). Loses when: the user's terminology differs from what's in memory.
No single mode wins all queries. Hybrid retrieval runs all three in parallel and fuses the results.
The dominant fusion method: Reciprocal Rank Fusion (RRF). For each retriever, get the top- results in ranked order. For each candidate document, sum across all retrievers (with a constant to dampen the head). The fused ranking is by the summed score.
RRF's appeal: scale-free. It does not require calibrated scores from each retriever (which is hard — vector cosines and BM25 scores live on different scales). Just the rank matters. A document that's #1 in vector AND #3 in BM25 is almost certainly the best answer. A document that's only #1 in vector but #50 in BM25 is suspect.
Production systems weight the retrievers (vector_weight = 0.6, lexical_weight = 0.3, graph_weight = 0.1) — typically learned from a retrieval-quality eval set.