Model Memory

Persistent memory for OpenCode: knowledge (permanent) + session (temporary).

Why You Need This

LLMs forget everything between sessions and after context compaction. This tool gives them persistent memory.

• Long-term knowledge (knowledge.md): Architecture, gotchas, patterns, configs
• Session memory (session.md): Plans, todos, progress — survives compaction
• Session isolation: Multiple sessions/agents can work in parallel without conflicts

Benefits: No rediscovering the same patterns. Faster debugging. Lower token usage. Work resumes seamlessly after interruptions.

Quick Start

1. Copy .opencode/ folder to your project
2. Add AGENTS.md to your project root
3. Done — OpenCode will automatically save discoveries and track work

Features

• Two-tier storage: Permanent knowledge + temporary session state
• Session isolation: Multiple sessions can coexist (v5.1+)
• 18 commands: 6 for knowledge, 12 for session
• Word boundary search: "log" won't match "catalog"
• Field weighting: Category matches rank higher than content
• Auto-recovery: Session pointer survives context compaction
• Human-readable: Plain Markdown files
• No database: Pure file-based
• Zero setup: Auto-installs Python via uv
• Cross-platform: macOS, Linux, Windows

Categories

Knowledge (10): architecture, discovery, pattern, gotcha, config, entity, decision, todo, reference, context

Session (7): plan, todo, progress, note, context, decision, blocker

Statuses (4): pending, in_progress, completed, blocked

License

MIT