Open-source agent memory system from Vectorize that lets AI agents retain, recall and reflect so they learn over time, not just remember chats.

Hindsight works by combining Retain, Recall and Reflect APIs: Three memory operations that store experiences, retrieve relevant context, and synthesize patterns into learned knowledge the agent can reuse., Biomimetic Memory Networks: Separates world facts, the agent's own experiences, observations and mental models so agents build understanding instead of a flat chat log., Two-Line LLM Wrapper: Swap your OpenAI or Anthropic client for a wrapped one and memories are stored and retrieved automatically on every call, covering 100+ models via LiteLLM., Benchmark-Leading Accuracy: Reports state-of-the-art performance on LongMemEval, with live per-model accuracy, latency and cost results published on a public benchmark site., 60+ Integrations and MCP Server: Works with coding agents such as Claude Code, Codex, Cursor, GitHub Copilot, Cline, Aider and Zed, plus a built-in MCP server and Python, Node.js, Go, CLI and REST clients. to help users with Persistent Assistant Memory: Giving customer-facing or personal assistants long-term memory of users, preferences and past interactions across sessions., Coding Agents That Learn: Letting Claude Code, Cursor or Codex remember project conventions and lessons learned from previous sessions., Replacing Ad-Hoc RAG: Using structured, reflective memory instead of plain vector retrieval for agents that must accumulate knowledge over time., Enterprise Agent Deployments: Running memory on-premises or in your own cloud with SSO, RBAC and custom SLAs for production workloads., Research on Agent Memory: Benchmarking and experimenting with long-term memory techniques using an open-source, reproducible system..

Key features include Retain, Recall and Reflect APIs: Three memory operations that store experiences, retrieve relevant context, and synthesize patterns into learned knowledge the agent can reuse., Biomimetic Memory Networks: Separates world facts, the agent's own experiences, observations and mental models so agents build understanding instead of a flat chat log., Two-Line LLM Wrapper: Swap your OpenAI or Anthropic client for a wrapped one and memories are stored and retrieved automatically on every call, covering 100+ models via LiteLLM., Benchmark-Leading Accuracy: Reports state-of-the-art performance on LongMemEval, with live per-model accuracy, latency and cost results published on a public benchmark site., 60+ Integrations and MCP Server: Works with coding agents such as Claude Code, Codex, Cursor, GitHub Copilot, Cline, Aider and Zed, plus a built-in MCP server and Python, Node.js, Go, CLI and REST clients..

Hindsight is useful for anyone interested in Persistent Assistant Memory: Giving customer-facing or personal assistants long-term memory of users, preferences and past interactions across sessions., Coding Agents That Learn: Letting Claude Code, Cursor or Codex remember project conventions and lessons learned from previous sessions., Replacing Ad-Hoc RAG: Using structured, reflective memory instead of plain vector retrieval for agents that must accumulate knowledge over time., Enterprise Agent Deployments: Running memory on-premises or in your own cloud with SSO, RBAC and custom SLAs for production workloads., Research on Agent Memory: Benchmarking and experimenting with long-term memory techniques using an open-source, reproducible system..

Hindsight offers a free tier with paid plans for advanced features.

Visit https://github.com/vectorize-io/hindsight to sign up and explore Hindsight.