OKF Agent Memory
Every agent session starts from zero, and the decisions from last week are lost unless someone wrote them down in the right place. OKF Agent Memory fixes that with plain files your team can already review. It implements Google's Open Knowledge Format v0.2, so memory lives as Markdown with YAML frontmatter in a git tracked knowledge folder, and every entry records whether an agent wrote it or a human verified it. A short push layer caps the working codex at 400 tokens, while the full knowledge bundle is pulled on demand with sub 300 microsecond BM25 search and served over MCP. Version 0.8.0 shipped on 5 November 2026, and the repo now has 866 stars under the MIT license.
What you get
- Memory as plain Markdown files in a git tracked folder, reviewable like code
- Provenance on every entry: agent written or human verified
- A 400 token working codex, with the full bundle pulled on demand
- Sub 300 microsecond BM25 search, served over MCP to your agents
Memory your team can review in git beats memory locked inside a service. Start with decisions, not chat logs, and let a human verify an entry before the agent treats it as fact.
Repo: https://github.com/okf-memory/okf-agent-memory
Quick poll
What would you use OKF Agent Memory for first?
- Storing team decisions in the knowledge folder
- Letting agents write memory, humans verify it
- Serving memory over MCP to my agents
- Watching it for now
Vote in the comments.
If you found OKF Agent Memory interesting, I write about tools like this every week in The AI-Native Engineering Leader, what actually works for engineering leaders building with AI agents, and what to skip.
Top comments (0)