Kenneth W. Bingham

case study · AI system · Model Context Protocol · open source

Memory that makes an AI assistant cheaper and more careful.

kravitz-mind is a memory server for AI assistants. It decides what the assistant is handed, what it is allowed to save, and how it checks itself, without touching the model. Any client that speaks MCP can use it, including Claude Code.

The problem

An AI assistant with memory usually reads that memory the expensive way. A whole index of notes is loaded into every session, whatever the question is, and a whole note is read when one line would have answered it. Every one of those tokens is paid for.

Memory also lets mistakes stick. If an assistant saves whatever it is told, sourced or not, a guess written down once comes back later looking like a fact.

Why it is hard

What I built

Five tools the assistant calls over MCP:

The gate is deterministic, with no AI judgment in it. A summary must fit on one line. A note with no source is stored but always shown as unverified. A near-duplicate is refused with the name of the subject it repeats. An update must name the version it replaces, so a stale update is dropped.

The proof

Measured on 116 real working notes with 30 questions asked in everyday wording. Tokens are estimated at four characters per token, with the same rule on both sides. The baseline loads the full index once per session.

Anyone can rerun the benchmark on the example notes in the repository, or point it at their own.

What it does not claim

Technologies and skills

7–34×fewer tokens, short sessions
30/30right note in the top three
16tests, all passing
0outside packages