KITE claims SOTA on LoCoMo and LongMemEval with no vectors and ~1.6K tokens of reader context.

memoket/memoket-kite, created Aug 12, is an Apache-2.0 agent memory engine that drops embeddings, vector DBs, and rerankers entirely for a topic-indexed single file. README reports 93.51% on LoCoMo and 85.60% on LongMemEval-S using gpt-4.1-mini across all stages, average reader context ~1.6K tokens per question. The pitch is determinism and receipts: a question compiles into a readable plan naming who said what and when, and returns empty when the answer was never stated instead of handing back a nearest match. Single-source benchmark claims, but the "return empty" behavior is the right design.