Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/vectorize-io/hindsight/llms.txt

Use this file to discover all available pages before exploring further.

Hindsight is an open-source agent memory system built to make AI agents smarter over time. Unlike simple conversation history or vector search, Hindsight uses biomimetic memory structures — organizing information into world facts, experiences, observations, and mental models — so agents can reason about what they know, not just what they’ve stored.

Quick Start

Add persistent memory to your agent in under 60 seconds

Installation

Deploy with Docker, Helm, pip, or embedded in Python

API Reference

Explore retain, recall, and reflect endpoints

Integrations

Connect with LangChain, LlamaIndex, CrewAI, OpenAI Agents, and more

​
How Hindsight works

Hindsight exposes three simple operations that handle all memory interactions:
1

Retain — store information

Pass text, conversations, or documents to Hindsight. The LLM extracts facts, entities, and relationships and stores them in a structured knowledge graph — not raw text.
client.retain(bank_id="my-agent", content="Alice works at Google as a software engineer")
2

Recall — search memories

Query memories using four parallel strategies: semantic similarity, keyword (BM25), graph traversal, and temporal reasoning. Results are fused and reranked automatically.
results = client.recall(bank_id="my-agent", query="Where does Alice work?")
3

Reflect — reason with memory

Generate disposition-aware responses grounded in retrieved memories. Optionally configure a mission, directives, and personality traits per memory bank.
answer = client.reflect(bank_id="my-agent", query="Tell me about Alice")

​
Key capabilities

Biomimetic memory hierarchy

Facts, experiences, observations, and mental models mirror how human memory consolidates knowledge over time

TEMPR multi-strategy retrieval

Semantic, keyword, graph, and temporal search run in parallel — fused with RRF and reranked with a cross-encoder

Automatic consolidation

Related facts are deduplicated and synthesized into observations, with evidence tracking and freshness trends

Mission & disposition

Configure per-bank personality, directives, and reasoning style to shape how the agent responds

20+ LLM providers

Works with OpenAI, Anthropic, Gemini, Groq, Ollama, local models, and any OpenAI-compatible endpoint

Built-in MCP server

Every memory bank exposes an MCP endpoint — connect Claude, Cursor, or any MCP-compatible tool directly

​
Core concepts

Memory architecture

Understand the hierarchy of memory types and how they interact

Retain

How Hindsight extracts and structures facts from raw content

Recall

How TEMPR’s four-way retrieval finds the right memories

Memory banks

Configure mission, directives, and disposition per agent

​
SDKs & clients

Python SDK

pip install hindsight-client

Node.js SDK

npm install @vectorize-io/hindsight-client

Go client

go get github.com/vectorize-io/hindsight/hindsight-clients/go

CLI

curl -fsSL https://hindsight.vectorize.io/get-cli | bash

Build docs developers (and LLMs) love

Get started for freeTalk to us