← Back to all projects

Hindsight

Memory that consolidates repeated facts into beliefs with evidence, rather than storing every mention separately

MIT
Stars
56.6k
Forks
6.8k
Open issues
160
Last commit
30 Oct 2026

What is Hindsight?

Storing an agent's conversations and searching them back is the easy half. The hard half is that the same thing gets said ten times in slightly different words, and a later correction has to win over an earlier statement. Hindsight is built around that: related facts are merged into a single belief that carries its evidence and updates as new evidence arrives, so a change of mind reads as an update rather than as two contradictory records. Retrieval runs four strategies at once — meaning, keywords, relationships and time — because a question about what changed last month is not answered by similarity. It is a server you run rather than a library you import, with an MCP interface and clients in several languages.

What can you do with Hindsight?

  • Merge repeats into one belief with evidence — The same fact stated many times becomes a single record that carries what supports it, instead of ten near-duplicates competing in search results.
  • Search four ways at once — Meaning, keywords, relationships between things and time are queried in parallel, so a question about when something changed is answerable.
  • Separate what happened from what is true — The agent's own actions are stored apart from facts about the world, which keeps "I told the user X" from being retrieved as though it were X.
  • Give a memory bank a character — Each store is configured with what it should care about, rules it must follow and how sceptical it should be — so retrieval is shaped by the job, not generic.
  • Attach it to an agent you already have — It runs as an MCP server and ships clients for several languages, so an existing agent gains memory without being rebuilt around it.
  • Run it embedded or as a service — The same system runs inside a Python process for a prototype, or as a container on Kubernetes when several agents share one memory.

Before you choose Hindsight

  • Consolidation and multi-strategy retrieval are model and compute work, so this is heavier per write and per query than storing text and running a vector search over it.
  • The headline accuracy comparison is published by the vendor, which notes that the other systems' scores in it are self-reported — treat the chart as a starting point and measure on your own data.

Star history

29 Aug to 30 Oct · +34.9k

21.7k56.6k

Frequently asked questions

Is Hindsight free for commercial use?

Hindsight is released under the MIT licence — OSI-approved open source, which permits commercial use.

How can Hindsight be deployed?

Hindsight is available as Self-hosted / Runs locally / Managed cloud.