GraphifyFrom the builder
Graph memory benchmark: LOCOMO and LongMemEval
Open-source Graphify treats its code graph as agent memory and publishes scores on LOCOMO and LongMemEval with the harness used to run them. On LOCOMO, recall@10 reached 0.497, well above the memory-layer baselines tested, while QA accuracy was 45.3%, between those two baselines.
On LongMemEval-S, QA accuracy was 76%, tied with dense RAG, and building the index used zero LLM credits. The post explains what the metrics measure and how to estimate token savings on your own repository.
Open-source Graphify publishes results on two public long-conversation benchmarks, along with the harness used to run them, so the numbers can be checked rather than taken on trust.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.