interest-memory

AI agents typically forget everything between sessions, which severely limits their continuity and practicality. interest-memory is a standalone memory backend designed to solve this problem. At the end of a session, it automatically extracts key interest points from the transcript, verifies and cleans them, and then writes them to a local knowledge base. At the start of the next session, it intelligently recalls and injects relevant context, giving agents true long-term memory. The entire system is extremely lightweight, requiring only a single ~18MB binary and a SQLite file to run, without the need for complex Postgres, Redis, or vector database stacks. The core memory unit is the "interest point," which is automatically merged when semantically similar, and written into the knowledge base as a wiki page via an LLM-driven agent loop, ensuring the knowledge base refines with use rather than bloat. This plugin is ideal for AI agent developers, researchers, and anyone building applications that require persistent memory across sessions. It supports bridging with multiple agent platforms (like Claude Code, Codex, DeepSeek Harness, etc.), providing unified session memory injection and search capabilities, helping agents transform from "forgetful" to "knowledgeable" and enhancing the coherence and efficiency of task handling.