Blog

Building in public. Shipping small.

Memory Vault series

Memory Vault v1.0 is Released

Seven weeks of build-in-public, scope locked, shipped. Open-source, self-hosted AI memory with hybrid search, MCP, knowledge graph, and local LLM chat — Postgres + pgvector, MIT, one-command Docker.

Memory Vault Has a Knowledge Graph

Milestone 7: building a knowledge graph without an LLM. spaCy NER + Postgres + co-occurrence, force-directed Cytoscape visualization. 80% of the way for 0% of the LLM cost.

Memory Vault Has a Dashboard

Milestone 6: a four-page web dashboard — Search, Browse, Ingest, Stats. Token-gated, baked into the same Docker image as the API. One command, one port.

Memory Vault Has a REST API

Milestone 5: every MCP tool is now an HTTP endpoint. Bearer auth, rate limiting, OpenAPI docs. Integrate AI memory into any app, any language.

Give Claude a Long-Term Memory

Milestone 4: MCP integration — four tools that let Claude store and retrieve memories mid-conversation. Connect Memory Vault in two minutes.

Memory Vault Now Runs With One Command

Milestone 3: PostgreSQL + pgvector + hybrid search + embeddings, all running with a single docker compose up. Here's what happens under the hood and the decisions I made.

How Hybrid Search Works — Building the Engine Behind Memory Vault

Pure vector search wasn't enough. Here's how combining semantic similarity, keyword matching, and Reciprocal Rank Fusion built the search engine behind Memory Vault.

I'm Building an Open-Source AI Memory System

Every conversation with an AI starts from zero. I got tired of that. So I built a solution, used it daily for months, and now I'm open-sourcing it.

Database toolkit for .NET