hindsight

Why we included this project

Most agent memory systems just store what happened in a conversation. Hindsight is built to do more: it learns from interactions and adjusts how it behaves based on feedback, so an assistant can get better at open-ended tasks the longer it runs. You get a self-hostable server with Python and Node clients, and an embedded mode that needs no separate service, so wiring persistent memory into an existing agent is a small code change. A UI and a documented API give you control over when memories are stored and recalled. It also holds state-of-the-art results on the LongMemEval benchmark, which is worth knowing if you are comparing memory layers. For simple one-off workflows, though, this is probably more machinery than you need.