Resources

Benchmarks, architecture writeups, and learning material — everything we publish about how MemLab is built and how it performs.

BENCHMARKS

How MemLab measures up

We run MemLab against the three public memory benchmarks the community uses to compare systems — LoCoMo, LongMemEval, and ConvoMem — and publish the full methodology, the raw numbers, and the source code that produced them.

Read the benchmark suite

ARCHITECTURE

Inside the gateway

A deep walkthrough of the Go gateway that fronts MemLab — auth, rate limiting, route permissions, request fan-out, and the claim-check pattern that keeps Kafka payloads tiny.

See the gateway design

EXTRACTION

How memories get made

A tour of the in-process extraction pipeline — facts, entities, relations, embeddings, five-tier dedup, and the hallucination guard that runs before anything lands in the graph.

Read the deep dive

GRAPH RAG

Why graph retrieval beats vector retrieval

The case for replacing pure vector RAG with hybrid graph RAG when the answer requires multi-hop reasoning, temporal awareness, or strict explainability — with the worked policing-graph example.

Read the writeup

BOOK NOTES

Knowledge Graphs and LLMs in Action

Listen-friendly chapter walkthroughs of the Manning book that informs much of the design — fifteen chapters, condensed for a single sitting per chapter, in flowing prose.

Open the chapter index

CHANGELOG

What shipped this quarter

New gateway endpoints, new playground features, performance work on the retrieval fan-out, and the bi-temporal upgrade that landed in the knowledge engine.

See recent updates