← 研踪 PaperTrailACL Anthology · ACL 2026:acl:2026.acl-demo.27 · 记录 #1
Hindsight: Structured Agent Memory that Retains, Recalls, and Reflects
来源日期（仅到月份） 2026-07
来源记录日期 2026-07
雷达观测 2026-11-01 22:08
摘要
We demonstrate Hindsight, a working memory system for AI agents that organizes long-term memory into four logical networks and exposes three core operations. The world, experience, observation, and opinion networks separate objective facts from subjective beliefs, giving developers visibility into what an agent knows versus what it believes. The retain, recall, and reflect operations handle ingestion, retrieval, and reasoning respectively, with a parallel pipeline that combines vector search, keyword matching, graph traversal, and temporal filtering, backed by PostgreSQL with pgvector. Unlike existing systems such as MemGPT, Zep, and Mem0, Hindsight is the only one that jointly provides fact-belief separation, temporal entity graphs, evolving opinions with confidence scores, and configurable behavioral profiles. On LongMemEval and LoCoMo, Hindsight with a 20B open-source model reaches 83.6% and 83.2% accuracy, outperforming full-context GPT-4o and all prior memory systems; with Gemini-3 Pro, LongMemEval accuracy reaches 91.4%. Our interactive demo lets users build memory graphs through multi-session conversations, inspect how memories are classified, and watch opinions form and change. The system is open-source under the MIT license, available as a Python package (pip install hindsight-all) and Docker image, with 13.3K GitHub stars and 763 forks to date, and in production use at Fortune 500 enterprises. Video demo: https://youtu.be/4M2wS-yEmVA.
论文发现出处
这些页面提供论文线索，不代表同行评审或质量认证。
观测到的版本
平台记录号代表本站观测到的元数据变化，不是作者修订号。来源日期仅到月份或年份时，按该月或该年首日存档。
记录 #1 · 2026-07 · 最新观测版本
Hindsight: Structured Agent Memory that Retains, Recalls, and Reflects
We demonstrate Hindsight, a working memory system for AI agents that organizes long-term memory into four logical networks and exposes three core operations. The world, experience, observation, and opinion networks separate objective facts from subjective beliefs, giving developers visibility into what an agent knows versus what it believes. The retain, recall, and reflect operations handle ingestion, retrieval, and reasoning respectively, with a parallel pipeline that combines vector search, keyword matching, graph traversal, and temporal filtering, backed by PostgreSQL with pgvector. Unlike existing systems such as MemGPT, Zep, and Mem0, Hindsight is the only one that jointly provides fact-belief separation, temporal entity graphs, evolving opinions with confidence scores, and configurable behavioral profiles. On LongMemEval and LoCoMo, Hindsight with a 20B open-source model reaches 83.6% and 83.2% accuracy, outperforming full-context GPT-4o and all prior memory systems; with Gemini-3 Pro, LongMemEval accuracy reaches 91.4%. Our interactive demo lets users build memory graphs through multi-session conversations, inspect how memories are classified, and watch opinions form and change. The system is open-source under the MIT license, available as a Python package (pip install hindsight-all) and Docker image, with 13.3K GitHub stars and 763 forks to date, and in production use at Fortune 500 enterprises. Video demo: https://youtu.be/4M2wS-yEmVA.
查看此版本解读 →查看此版本原文 ↗