hindsight is a long-term agent memory system that uses biomimetic data structures so that agents learn over time rather than merely recalling conversation history. It wraps existing LLM clients to store and retrieve memories automatically with minimal code changes.
Project overview
By organizing memories into World facts, Experiences, and Mental Models, hindsight addresses the gap between recalling conversation history and enabling agents to genuinely learn from past interactions.
Project type
AI Agent
Deployment
Refer to project documentation
License
MIT
Best for
Developers and AI engineers building agent applications who need agents to learn from past interactions rather than only recall conversation history.
Teams that want to integrate long-term memory into existing LLM-based agents with minimal code changes.
Key capabilities
Wraps an existing LLM client so memories are stored and retrieved automatically as LLM calls are made, requiring only two lines of code.
Pushes new information into memory banks, extracting facts, temporal data, entities, and relationships via LLM normalization.
Performs deep analysis of existing memories to form new connections, observations, and insights.
Isolates memories and mental models to individual users using custom metadata for personalized chatbot experiences.
Runs the memory system directly inside a Python application without requiring a separate server deployment.
Limitations and risks
An external LLM provider API key is required to process and extract memory information.
Intel Macs require the hindsight-all-slim package instead of hindsight-all for embedded DB mode.
Getting started
Setup requires configuring external LLM API keys and coding to integrate memory into agents. The recommended path is to acquire an LLM API key, run the hindsight server via Docker or Python, and integrate the client in code using retain and recall operations.