vectorize-io/hindsight
How Hindsight gives AI agents long-term learning memory
Stop using basic RAG for your AI agent memory.
Basic conversation RAG fails when agents need long-term learning.
First, simple vector retrieval only finds matching keywords. It doesn't synthesize past failures or build persistent mental models.
Three core operations turn raw observations into learned knowledge.
Second, Hindsight uses three operations: retain raw observations, recall relevant context, and reflect to consolidate lessons into structured knowledge pages.
Deploy locally in one Docker command with support for 25+ LLMs.
Third, you can spin up the full server in one Docker command. It exposes a fast API on port 8888 and a visual dashboard on port 9999, connecting to twenty-five LLM providers.
Instant skills integration for Claude Code, Cursor, and MCP.
Fourth, you can add it directly to Claude Code and Cursor with one command, giving your coding assistant instant access to project memory and documentation.
Free. MIT licensed. Twenty-eight thousand stars on GitHub.
It's completely free and open source under the MIT license, with over twenty-eight thousand stars on GitHub.