Dev.to
7/16/2026

My note server has no search engine. Search still answers in 0.4 ms.
Short summary
Vellum MCP is a self-hosted markdown note server that achieves sub-millisecond search via a ranked in-RAM linear scan instead of a search engine or vector index. For a 2,000-note vault, exact-term queries return in 0.38ms with custom diacritic folding, typo tolerance scaled to word length, and context-aware ranking. The author argues that for small-scale personal knowledge bases, a hand-written scan is faster and simpler than bolting on a search library, and can be swapped behind a Searcher interface when scale demands it.
- •Linear scan over in-RAM notes achieves sub-ms search for ~2,000-note vaults
- •Custom matcher handles diacritics, typos, and context-aware ranking better than generic engines
- •Searcher interface allows swapping in a real engine when corpus grows large enough
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



