Prinevo Memory: A Context Graph for Long-Term Agent Memory

说明与用途

Prinevo Memory 采用带有类型化声明和双时间超越的上下文图，提升长期代理记忆准确率，在 LoCoMo 上达到 83.89%，在 LongMemEval 上达到 82.60%。

问题

Agents that start each conversation from zero lack the ability to retrieve prior facts, reason across sessions, and handle knowledge updates, leading to poor answer quality.

结果

On LoCoMo, 83.89% headline score (77.84% strict) and on LongMemEval, 82.60% headline score (88.80% strict). The system shows a 54‑percentage‑point uplift in early engineering Q&A testing versus no memory.

工作流

• Capture session text into typed claims with entity references
• Resolve entity mentions with confidence checks
• Record claim timestamps and true‑time validity (bi‑temporal supersession)
• Retrieve evidence using hybrid relevance, recency, and entity‑anchored lanes
• Classify question type (cardinality, temporal, answer mode)
• Select appropriate answer strategy (point lookup, exhaustive gather, comparison, aggregate)
• If evidence missing, widen retrieval and retry before abstaining

证据

Prinevo Memory is a domain‑pluggable context graph built around structured capture, time‑aware knowledge, and retrieval strategies matched to the question being answered.

来源

Both benchmark runs used Claude Haiku 4.5 for capture and answering, with Claude Sonnet 5 for judging.

来源

An earlier full LoCoMo error analysis found that only 2% of wrong answers were true extraction misses. In 63% of wrong answers, the gold fact was already present in the retrieved evidence but the answer was assembled incorrectly.

来源

Early engineering Q&A testing has shown roughly a 54‑percentage‑point uplift with memory versus without it.

来源