An AI agent changing between sessions does not need a perfect transcript; it needs a small, trustworthy record of decisions and unfinished work. This project makes that record ordinary Markdown files, so it remains readable and editable outside any one model or client. The trade-off is deliberate: users decide what becomes durable memory rather than treating every conversation as reliable knowledge.
What Sets It Apart
- The notes are the source of truth. Local Markdown files can be opened in Obsidian or any text editor, making the memory inspectable and portable rather than hidden in a provider-specific database.
- Several agent clients can share one vault. Supported clients can draw on the same notes and tasks, which is useful when switching between coding agents without rebuilding project context.
- Memory is explicit, not magical. The system helps agents save and retrieve selected information, but does not automatically turn every chat into verified facts. This keeps durable notes intentional, while placing responsibility on the user and agent to record important outcomes.
- Local checks do not require a continuously running server or extra model calls. The AI client still handles model interactions, and its normal provider costs and data policies apply when context is sent to it.
Who It’s For
Great fit if you work across AI coding agents and want an inspectable, user-controlled record of decisions, sources, and next steps. Look elsewhere if you expect automatic capture of every conversation, guaranteed semantic recall, or a managed cloud memory service: retrieval uses local keyword matching, and important information must be deliberately recorded.
