# Hindsight

> Updated 2026-09-28 · type: tool · category: ai-infrastructure · status: active · rev 1

Hindsight gives agents retain/recall/reflect over four memory networks — world, experience, opinion, observation — via per-bank MCP endpoints.

- Open source: yes (MIT)
- Self-hostable: yes
- Pricing model: free
- Best for: A team building long-lived agents that need durable, structured memory — separating world knowledge, lived experience, opinions, and observations into distinct banks reached over MCP and self-hosted on a PostgreSQL or Oracle AI Database backend.
- Not for: A throwaway or single-turn agent that needs no persistent memory, or a team unwilling to run and operate a database-backed memory service.
- Last verified: 2026-09-28

- **Canonical:** https://gtmstacker.com/registry/tool/hindsight/
- **Source:** [vectorize-io · GitHub](https://github.com/vectorize-io/hindsight)
- **Tags:** ai-infrastructure, agent-memory, mcp-agents, retrieval, self-hostable, memory
- **Repository:** https://github.com/vectorize-io/hindsight

## Is Hindsight open source?

Yes, Hindsight is open source under the MIT license.

## How much does Hindsight cost?

Hindsight is free to use.

## Can I self-host Hindsight?

Yes, Hindsight can be self-hosted (the source is available under the MIT license).

## Alternatives & related

- [motif](https://gtmstacker.com/registry/tool/motif/)
- [Itsuki](https://gtmstacker.com/registry/tool/itsuki-memory/)
- [Cognee](https://gtmstacker.com/registry/tool/cognee/)


---

Hindsight is an agent memory system exposing retain/recall/reflect over four memory networks — world, experience, opinion, observation — with per-bank MCP endpoints and 60+ integrations. Open source: yes (MIT); self-hostable via `docker run` with PostgreSQL or Oracle AI Database backends; free OSS (vectorize.io offers an optional hosted version). It has ~39.9k stars and is active (~3,257 commits).

## What it does

Hindsight gives an agent a structured, long-term memory rather than a single undifferentiated store. It separates memory into four networks — world (facts about the domain), experience (what happened), opinion (judgments formed), and observation (what was seen) — and exposes three operations across them: retain (write), recall (read), and reflect (synthesize). Each memory bank has its own MCP endpoint, so an agent addresses the right kind of memory directly, and the project ships 60+ integrations including Claude Code, Cursor, CrewAI, LangGraph, and n8n. You self-host it with `docker run` backed by PostgreSQL or an Oracle AI Database. Open source: yes (MIT); self-hostable; free OSS.

## Provenance

- MIT per repo; ~39.9k stars; ~3,257 commits; active. retain/recall/reflect over four memory networks (world/experience/opinion/observation); per-bank MCP endpoints; self-hostable via `docker run` (PostgreSQL / Oracle AI Database); 60+ integrations (Claude Code, Cursor, CrewAI, LangGraph, n8n) (github.com/vectorize-io/hindsight, verified 2026-09-28).
- vectorize.io offers an optional hosted version; the OSS core is free and self-hostable.
- Vendor claim: "state-of-the-art on long-term memory tasks" — the vendor's own claim (vendor-claim), no benchmark cited here, not independently verified.
- Curated from the GTM Stacker signal registry (2026-09-28 pass); license/facts independently verified 2026-09-28.

## Why it matters for a GTM stack

A GTM agent that forgets is a liability: it re-asks answered questions, loses account context between sessions, and cannot build a durable view of a prospect. Hindsight offers a shaped memory for exactly that — separating world facts, lived experience, formed opinions, and raw observations, each addressable over MCP — so an outbound or account agent can retain, recall, and reflect across a long relationship, self-hosted on your own PostgreSQL. Open source: yes (MIT); self-hostable; free OSS. The honest read: the four-network model plus per-bank MCP is a serious, well-adopted design (~39.9k stars, 60+ integrations), but "state-of-the-art on long-term memory" is the vendor's unbenchmarked claim, and self-hosting means running a real database — evaluate recall quality on your own data before it becomes the agent's source of truth.