hindsight
Hindsight: Agent Memory That Learns
Hindsight is an agent memory system that enables AI agents to learn and recall information over time, achieving state-of-the-art performance on long-term memory benchmarks. Unlike traditional retrieval-augmented generation (RAG) systems that simply recall conversation history, Hindsight uses biomimetic data structures—world facts, experiences, observations, and mental models—to help agents form consolidated beliefs and understanding. It offers three core operations: retain (store memories), recall (retrieve relevant information), and reflect (deep reasoning over memory). The system integrates with 60+ platforms including coding assistants, agent frameworks, and no-code tools through native SDKs for Python, Node.js, and Go.
Key Features
Use Cases
- 01Personalizing AI chatbots by storing per-user preferences and conversation history in isolated memory banks
- 02Giving coding agents long-term project memory built from git history and past sessions
- 03Building AI project managers that reflect on risk mitigation needs across historical data
- 04Creating sales agents that analyze why certain outreach messages succeeded or failed
- 05Developing support agents that identify documentation gaps based on recurring customer questions
- 06Enabling autonomous AI employees to learn complex tasks and adapt behavior based on feedback
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
hindsight — FAQ
What is Hindsight and how does it differ from RAG?+
Hindsight is an agent memory system focused on making agents learn, not just remember. Unlike RAG systems that retrieve conversation history, Hindsight consolidates facts into observations with evidence, builds mental models, and uses four parallel retrieval strategies (semantic, keyword, graph, temporal) to deliver state-of-the-art accuracy on long-term memory benchmarks.
How do I install Hindsight?+
You can run Hindsight via Docker (recommended), bare metal with pip install hindsight-api, Kubernetes with Helm, or use the managed Hindsight Cloud. For embedding without a server, install hindsight-all via pip. All options require an LLM provider API key (OpenAI, Anthropic, or 25+ others).
Which AI clients and frameworks does Hindsight work with?+
Hindsight integrates with 60+ platforms including coding agents (Claude Code, Codex, Cursor, GitHub Copilot), agent frameworks (LangGraph, LlamaIndex, CrewAI, Pydantic AI, OpenAI Agents SDK), no-code tools (n8n, Zapier, Dify), and apps (ChatGPT, Perplexity, Obsidian). It also exposes a Model Context Protocol (MCP) endpoint per bank.
Do I need API keys or special prerequisites?+
Yes, you need an API key for at least one LLM provider. Hindsight supports OpenAI, Anthropic, Gemini, Groq, Bedrock, local models via Ollama/LM Studio, and 25+ others. Some integrations like openai-codex, claude-code, and github-copilot can use existing subscriptions without separate API keys.
Is Hindsight free and open source?+
Yes, Hindsight is MIT-licensed open source software. You can self-host it for free using Docker, pip, or Kubernetes. There's also a managed Hindsight Cloud option with usage-based billing, free credits to start, and no fixed monthly fees.
How do I add Hindsight to an existing agent in two lines of code?+
Use the LLM wrapper: install hindsight-litellm, then wrap your existing OpenAI or Anthropic client with wrap_openai() or wrap_anthropic(). Memories are automatically stored and retrieved on every LLM call with no other code changes required.
How do I install hindsight?+
Open the source repository on GitHub and follow its README. hindsight is a skill — MCP Agents Market links you directly to the official repo.
Is hindsight free?+
hindsight is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.