Markdown is 40× more token-efficient with SQLite
Hi HN, I'm Richard, founder of Native. I'm working on open-source context ownership, to stop AI centralisation: <a href="https://github.com/withnative/native" rel="nofollow">https://github.com/withnative/native</a><p>My first focu
In a 40-question benchmark, agents using SQL through a local MCP used over 40× fewer total tokens per fully correct answer than a wikilink Markdown vault.
40× fewer tokens: wrap Markdown in SQLite — Native Native . Product Research Company Open source Product Research Company Open source Ask AI Ask ChatGPT Ask Claude Copy prompt Sign in Research preview 40× fewer tokens: wrap Markdown in SQLite Richard Ng Tokens per fully correct answer Native's SQLite MCP 172k 43.5× fewer tokens Native-to-Markdown 653k 11.5× fewer tokens Markdown + wikilinks 7.5M Cached input Uncached input Reasoning Output TL;DR: Agents do better with SQLite: 97% fewer failed trials for a fixed 1M input-token budget Token spend gets slashed: 40× fewer tokens, ~20× lower modell
In a 40-question benchmark, agents using SQL through a local MCP used over 40× fewer total tokens per fully correct answer than a wikilink Markdown vault.
40× fewer tokens: wrap Markdown in SQLite — Native Native . Product Research Company Open source Product Research Company Open source Ask AI Ask ChatGPT Ask Claude Copy prompt Sign in Research preview 40× fewer tokens: wrap Markdown in SQLite Richard Ng Tokens per fully correct answer Native's SQLite MCP 172k 43.5× fewer tokens Native-to-Markdown 653k 11.5× fewer tokens Markdown + wikilinks 7.5M Cached input Uncached input Reasoning Output TL;DR: Agents do better with SQLite: 97% fewer failed trials for a fixed 1M input-token budget Token spend gets slashed: 40× fewer tokens, ~20× lower modell
Related listings
Typing sounds for Mac: 30+ real recorded mechanical switches, from Cherry MX Blue to Holy Panda. A menu bar app that wor...
