Vectorize Hindsight iconVectorize Hindsight

Open-source agent memory from Vectorize.io: retain, recall and reflect over facts, entities and time, with MCP, Postgres and a managed cloud

60k+ GitHub Stars
MIT License
15,000+ Developers (vendor claim)

Overview

Vectorize.io is a Boulder startup (not Cloudflare's Vectorize vector database) whose main product is now Hindsight, an open-source "agent memory that learns." Agents call three operations: Retain stores information, with an LLM extracting facts, entities, dates and relationships; Recall searches memory four ways at once (vector, BM25 keyword, graph and temporal) and merges the results with reranking; Reflect reasons over stored memories to answer questions that need more than a lookup. Mental models keep standing answers (such as "what does this user prefer?") updated in the background, so reading one needs no LLM call. Hindsight runs on Postgres with pgvector, has Python, TypeScript and Go SDKs, exposes an MCP endpoint per memory bank, and lists 60+ integrations including LangGraph, CrewAI, Claude Code, Codex and Cursor. Vectorize started in 2024 with a RAG pipeline platform and the Iris document extraction model; the RAG platform has closed to new signups, and Iris extraction is now billed inside Hindsight Cloud. Also on this site: Cloudflare Vectorize, a different product from a different company.

The Verdict

Who Should Use Hindsight?

Best For

  • Teams building agents that must remember users and facts across sessions
  • Developers who want self-hosted memory on their own Postgres
  • Coding-agent users who want shared memory over MCP

Not Ideal For

  • Simple chatbots where a short conversation buffer is enough
  • Teams looking for Vectorize's old RAG pipeline platform (closed to new signups)

What's Great

  • MIT-licensed and free to self-host with no usage limits
  • Combines vector, keyword, graph and time-based recall
  • Strong LongMemEval results, which the README says outside groups reproduced

Watch Out For

  • Thoughtworks Radar rates it "Assess": compare with Mem0 first
  • Consolidation adds extra LLM calls; measure if it pays off
  • Benchmark comparisons on the vendor blog are vendor-run

Pricing

Cloud storage costs $0.25 per million tokens stored a month (first 30 days free); Iris extraction is $7.50 per million tokens.

View all features & details

Key Features

  • Retain, Recall and Reflect operations
  • Mental models, observations and isolated memory banks
  • Postgres with pgvector, or embedded pg0 for local use
  • MCP endpoint per bank; REST API and CLI
  • Python, TypeScript and Go SDKs
  • Own models: hs-embed-1, hs-rerank-1, hs-decide-1-flash

Company

  • Vectorize, Inc., founded 2024 by Chris Latimer (CEO) and Chris Bartholomew (CTO)
  • $3.6M seed led by True Ventures (2024)
  • SOC 2 Type 2

Real-World Usage

Adoption

  • 15,000+ developers use Vectorize and Hindsight (vendor claim)
  • Thoughtworks Technology Radar, Nov 2026: Assess

Logos on Site

  • NVIDIA
  • Groq
  • Electronic Arts

How It Compares

Feature Hindsight Mem0 Zep
Approach Retain, recall, reflect with mental models Universal memory layer Knowledge-graph memory
Open source license MIT Apache 2.0 Open source core
Retrieval Vector, keyword, graph and temporal Memory search Graph-based
Thoughtworks Radar advice Assess; compare with Mem0 first Baseline to compare against Not covered here

Sources: Vectorize · Pricing · About · Hindsight GitHub · RAG platform docs (signups closed) · Seed round (2024)

Is this your tool? Add a badge