Lemmalog
by Jordy Zomer · Framework · Text & code
Datalog engine storing an agent's investigation state as facts and rules, with retraction, provenance and time-bounded facts; built for vulnerability research
AccessOpen codeLicenseMIT
Evidence
- Aug 28LongMemEval F1 0.463 (accuracy 0.575) versus SimpleMem 0.480, PropMem 0.550 and GPT-4.1 full context 0.197; about 2,700 tokens per query
- Aug 28Commenters say stale or disproven facts persisting in LLM memory is a real problem in long Claude research projects; invalidation doesn't propagate