Native is a research-preview product, currently open source, built around a simple premise: the Markdown vault that works well for humans is a poor substrate for AI agents. When an agent has to read notes file by file, context costs balloon and accuracy drops as the collection grows. Native addresses this by relocating the vault into SQLite, turning notes into structured data an agent can query directly.
The core mechanism is query-based retrieval. Instead of walking a chain of notes and reading each in full, an agent sends SQL and gets back only what it needs. The site illustrates the difference with token-per-correct-answer comparisons and a timed accuracy test, claiming roughly 40× fewer tokens and 98% fewer mistakes versus a Markdown wiki. Setup is described as one line once access is granted.
Sharing is the other pillar. Four people can bring separate vaults into one shared space with permissions controlling who sees what, which suits small teams maintaining a common knowledge base. Because the project is open source, technical users can inspect the code before trusting it with their notes, and the site explicitly suggests asking an agent to review the code first.
Native is best suited to heavy note-takers, researchers, and small teams already running agents over a large Markdown collection, especially those comfortable with SQLite and early-access software. Casual users with small vaults will see little benefit. Access is gated through an email waitlist, so plan for
Comments & Ratings
0No comments yet. Be the first to review.