withnative
Native's research into the efficiency of querying Markdown material with SQLite.
What does it do?
Native's article describes a search method that combines SQLite and local MCP with Markdown material. In a benchmark using 40 questions, it compared the tokens an agent needed per correct answer when using SQL with a conventional Markdown vault approach. Readers can examine the published research material and run the benchmark on their own data and computer. The article also discusses research directions for real-time collaboration and self-managed work material.
Key features
- Describe a research approach for querying Markdown material through SQLite and local MCP.
- Compare token usage per correct answer in a benchmark of 40 questions.
- Explain how to run the benchmark on your own data and computer.
- State that the core engine has been released as open source.