Giving an OpenCode Coding Agent Persistent, Editable Memory

Developer @ghilteras has introduced an experimental plugin for the OpenCode editor designed to provide AI coding agents with persistent, editable memory. Unlike opaque remote services, this plugin treats memory as structured Markdown files stored locally, allowing agents to maintain project-specific context and global preferences across sessions. The system utilizes 'memory blocks' with YAML frontmatter to define scope, size limits, and usage descriptions. Additionally, the plugin features an optional, append-only journal for tracking decisions and discoveries, supported by local semantic search powered by Transformers.js. By keeping state on disk, developers can manually inspect and edit the agent's knowledge, addressing common issues where AI models struggle to retain context after restarts or context compaction. The project is open-source and available on GitHub, offering a transparent approach to improving long-term agent reliability in software development workflows.