Okf Agent Memory

OKF Agent Memory is a git native persistent memory layer for AI coding agents. Instead of a vector database, it stores memory as human readable Markdown files in a repository, structured to the Open Knowledge Format version 0.2, so changes can be inspected with git diff, git blame and code review. The tool is written in Go and ships as a single binary with no runtime dependencies. It searches the knowledge graph in memory using BM25 ranking, with the site citing queries completing in under 300 microseconds. A progressive disclosure approach lets an agent pull only the specific concept it needs rather than loading a large monolithic file into every prompt, and the site reports up to an 80 percent cut in prompt tokens. Trust tiers distinguish decisions verified by a human from drafts generated by an agent, so agents do not overwrite human architectural decisions. An embedded stdio MCP server connects it to Claude Code, Cursor, Windsurf and custom IDEs, and a bootstrap command scaffolds knowledge bundles and enriches the AGENTS.md file in a new or existing repository. The site also targets local models run through Ollama and LM Studio. Installation is by curl script, go install, Homebrew or building from source. A cloud beta with a waitlist is offered. The site publishes no prices.