Hindsight

About

Give your agents memory that carries over between sessions and gets better over time

Run it locally with Docker (the README one-liner exposes the API on port 8888 and the UI on 9999) or pip install hindsight-all, and set HINDSIGHT_API_LLM_PROVIDER to any of 25+ providers, including local Ollama or an existing Claude, ChatGPT, Cursor or Copilot subscription. Point an MCP client at http://localhost:8888/mcp/<bank_id>/ to get retain, recall and reflect tools. Things to know: each memory bank has its own MCP endpoint, so use one bank per agent or project; Hindsight Cloud at api.hindsight.vectorize.io skips the deployment if you do not want to run Postgres yourself.

Features

• Memory organized as world facts, experiences and opinions
• Retain, recall and reflect operations
• Built-in MCP endpoint per memory bank
• Python and TypeScript clients plus a 2-line LLM wrapper
• Works with 25+ LLM providers, including local models
• Self-host with Docker, pip or Helm, or use Hindsight Cloud

Install