Hindsight
Open-source agent memory system from Vectorize focused on agents that learn: retain, recall and reflect operations over memory banks, with an MCP server and a managed cloud.
Tags
Description
Builds observations, mental models and knowledge pages on top of raw memories; Vectorize reports state-of-the-art LongMemEval results and publishes live benchmarks at benchmarks.hindsight.vectorize.io (paper: arXiv 2512.12818). Run the server with Docker or embedded in Python, then use the Python/TypeScript clients, an LLM wrapper, framework and coding-agent integrations, or the per-bank MCP endpoint (http://localhost:8888/mcp/{bank_id}/ when self-hosted). Self-hosted is MIT and free; Hindsight Cloud is pay-as-you-go on token usage with free credits to start; Enterprise on request.
Agent access
Related
- Mem0: Memory layer for AI agents and apps: drop-in persistent memory that lets agents learn from past interactions.
- Zep: Unified context layer that combines business data, documents and conversations into governed context and memory for agents.
- Letta: Platform for stateful agents with memory, identity and the ability to learn over time (formerly MemGPT); terminal UI, App Server, desktop app and Agent SDK.
- Graphiti: Zep's open-source framework for building real-time knowledge graphs for AI agents.
- Supermemory: Memory and context engine for AI agents with a Memory API; can run fully locally or hosted.
Sources
- https://github.com/vectorize-io/hindsight
- https://vectorize.io/pricing
- https://hindsight.vectorize.io/developer/mcp-server
- https://hindsight.vectorize.io/llms.txt
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)