Meilisearch Cloud

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build the search I actually need instead of hosted Meilisearch — and note the engine is open source, so this is running it rather than writing it.

**First, measure your corpus.** Under a million documents on one machine, SQLite's FTS5 answers in single-digit milliseconds and needs nothing new running at all. Reach for a search engine when you want typo tolerance and faceting out of the box, not because a search box implies one.

STACK
- Meilisearch in a container, or SQLite with FTS5 if the corpus is small
- Node 20+ with Fastify for the indexing and the search endpoint
- Caddy in front, **with the engine bound to localhost** — a search engine on a public interface is a database on a public interface

INDEXING
- Index at write time, never scan at read time. A document changes, one record is updated
- A full reindex command that works and is tested, because you will need it after a schema change
- **Fields weighted: a title match outranks a body match, by a lot.** That single rule is most of perceived relevance
- Store only what the results page displays, plus what you filter on. An index carrying whole documents is an index that stops fitting in memory

SEARCH
- Prefix matching so results appear as you type, and typo tolerance scaled to word length — one edit on a short word, two on a long one
- Facets as declared attributes with counts, which is one query
- **Permission filters compiled into the query**, never applied to the results afterwards. Filtering after is how a private title leaks into an autocomplete
- Synonyms as a small table you maintain from your own failed searches

THE PART THAT MAKES IT ACTUALLY GOOD
- **Record every query, the result count, and whether anything was clicked.** That log is your relevance data and the only thing that will ever tell you the search is working
- Failed searches ranked: that list is a content backlog and a synonym list at the same time
- Look at it monthly. Ten synonyms added from real queries beat any amount of tuning by intuition

OPERATIONS
- A snapshot on a schedule, off the machine. The index is rebuildable from your database, so this is convenience rather than the backup that matters
- Memory is the constraint: this engine is fast because it keeps a lot in memory, so measure with your real corpus before sizing
- Health endpoint that runs one query through the real path

AND THE STATIC ALTERNATIVE
- For a documentation or content site, Pagefind builds an index at build time and searches entirely in the browser. No server, no query cost, works offline. If that fits, it is the cheapest good answer available

WHAT MATTERS MOST
Measuring the corpus first, and logging the queries. The first tells you whether you need any of this; the second is what makes it improve.