OKF Agent Memory: Git-Native Memory for AI Agents — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OKF Agent Memory provides a domain-neutral persistent memory layer based on Open Knowledge Format (OKF) v0.2, storing project knowledge as plain Markdown files with YAML frontmatter in a repository's
knowledge/directory - The Go-based tool achieves <300µs search latency and ~4ms graph validation via in-memory BM25 retrieval, eliminating recurring vector embedding API costs and network roundtrips
- Built on OKF v0.2, the format supports full provenance (
sources), trust tiers (generatedvsverified), and lifecycle metadata (status,stale_after) - The standalone CLI ships with a built-in Model Context Protocol (MCP) server over stdio for integration with Claude Code, Cursor, and Codex, with sub-5ms startup and zero external dependencies
- Progressive Disclosure via hierarchical
index.mdfiles and link graphs ensures agents load only the concepts they need, solving context bloat and memory rot - The project includes example bundles spanning microservices architecture, executive coaching, and literature review to demonstrate domain neutrality
- Released under MIT License, the project includes a one-command
okf bootstrapthat scaffolds the full memory stack, agent skill definition, andAGENTS.mdinstructions into any target repository
Why it matters: AI coding agents lose architectural decisions and operational facts every time context windows close, and the dominant workaround—ad-hoc CLAUDE.md files or opaque vector databases—either lacks rigor or introduces recurring API costs. OKF Agent Memory undercuts both by offering sub-millisecond lexical search and full Git diff auditability in a zero-dependency binary, which matters for teams standardizing on Claude Code, Cursor, or Codex and needing version-controlled, inspectable agent memory without vendor lock-in.
Ask SkimNews
