OkfAgentMemory
AI coding agents typically lose context between sessions: each new conversation starts from scratch, forcing repeated re-explanation of project history and decisions, which costs both time and tokens. okf-agent-memory gives agents persistent memory stored directly in the same git repository as the code, implementing Google's Open Knowledge Format (OKF) v0.2. It uses an embedded, in-memory BM25 search engine (a standard text-relevance ranking algorithm) that returns results in under 300 microseconds, exposed through an embedded MCP server, without requiring any external database or added dependency. The project reports that this approach cuts the amount of context needed to remind an agent of prior work by 80%, reducing token cost accordingly.
The project shows meaningful real-world traction: as of research, it has 578 GitHub stars, was last updated the same day as research, and carries only 3 open issues — a stronger adoption and maintenance signal than most newly bookmarked tools.
This is placed in Tools as a concrete, installable memory system for AI coding agents. It sits in Assess because, despite strong external adoption signals, there is no confirmed first-person use in the user's own pipeline, which the Trial gate requires. Position is inner, reflecting the project's comparatively high star count, active maintenance, and low open-issue count relative to typical Assess-ring candidates.
