Blazing fast search

When I started building medi, my command-line notes manager, the goal was a fast, local-first workflow. But how fast is fast? Could it stay fast at scale? With a database of thousands of notes, would search grind to a halt?

A user on Reddit asked the same question, so I decided to find out. At first, I hacked together a quick bash script, but I wanted something more solid in the test suite. The results were better than I expected.

The test

I wrote a simple performance test that does three things:

1. Generate Data: Create 1,000 notes, each ~50KB of random text (~50MB total).
2. Build Index: Run medi reindex to build a full-text search index from scratch.
3. Run Search: Time a “cold” search for a single word, including application startup.

The results

The numbers speak for themselves:

• Reindexing 1,000 notes (50MB of text): 2.7 seconds
• Searching the entire database: 106 milliseconds

Analysis

Rebuilding a search index for 50MB of text in under three seconds is a great sign, even if the index gets corrupted, it can be rebuilt in no time.

The real win is search speed. That 106ms includes everything: starting the binary, loading the index from disk, parsing the query, running the search, and printing the result. The actual search is probably in the single-digit milliseconds. From the user’s perspective it’s instant.

This test also validates the architecture choice: sled for storage + tantivy for search is reliable and fast. It gives me confidence that medi will stay responsive with 10_000 notes, just as it is with 10.

If you’re building a Rust app that needs local search, I can highly recommend tantivy. You can also check out medi and its implementation on my GitHub, the specific test can be found at test_performance()