Printed because it was nearly cut
A two-day-old repository for agent memory has 353 stars and no vector database
okf-agent-memory was created on 5 September and stores an agent’s project memory as Markdown files with YAML frontmatter in a knowledge directory. It is indexed by BM25 in memory at each invocation and exposed over MCP. Its README reports search under 300 microseconds against no stated baseline, and full-corpus graph validation around 4 milliseconds.
A star count is not news and this paper does not print them. What earns the space is the shape of the choice: memory as diffable files in the repository, retrieved lexically, against the embedding service and vector store that the same problem usually attracts.
It is MIT licensed, written in Go with no external dependencies, and okf bootstrap scaffolds the directory, an AGENTS.md and a skill folder into an existing repository. The convention it enforces is search-before-write, which exists to stop the same concept being recorded three times under three names.
Two days is not a track record and the numbers in that README are the author’s own, measured on unstated hardware. The reason to look at it this week rather than in six months is that the format is plain text you already know how to review, so evaluating it costs a reading rather than a migration.
What it changes If you are about to stand up a vector store for agent memory, read this first and decide deliberately rather than by default. The benchmark numbers are the author’s and are not independently checked.
The performance figures are the project’s own, published in its README, on unstated hardware and not independently reproduced.
Publisher's claim, unverified