An agent’s memory is only useful if it survives a model or CLI change. This project puts that memory in ordinary Markdown files, shifting continuity from a particular assistant’s hidden state to a knowledge base the user can inspect and keep. The more consequential design choice is that new sources can revise existing notes, rather than leaving the vault as a growing pile of summaries.
What Sets It Apart
- Memory lives outside the agent. Multiple supported coding agents can work with the same vault, so changing tools does not require rebuilding accumulated context.
- Ingestion maintains knowledge, not just a transcript archive. Existing pages can be updated, and contradictions and cross-source patterns can be surfaced. That is useful when project facts evolve, but it also means generated edits deserve review.
- Freshness is treated as a storage rule. The project’s Open Knowledge Metabolism approach distinguishes timeless facts, dated claims, and pointers to changing information. This helps prevent volatile details from silently becoming stale notes.
- The shared Crew logbook is a smaller, separate option. Agents can record completed work in a plain folder without requiring Obsidian; human review is required before a suggested lesson becomes a rule.
Who It Suits
Great fit if you already keep substantial Markdown notes and want coding agents to reuse decisions, project context, and research across sessions or platforms. Look elsewhere if you want a polished standalone chat application, a fully managed memory service, or hands-off automation without reviewing edits. The core vault workflow can run without research API keys, but some research integrations need them; background and scheduled behavior also depends on platform-specific setup.
