Your first spotlight

Chunk a document. Rank snippets against a probe glyph.

Spotlight compiles one document into fingerprinted chunks, then scores each chunk against a probe. Use Rank to inspect every chunk, or Query to filter by threshold.

What you will build

1. Create a spotlight document from text.
2. Rank chunks against a probe glyph.
3. Query with a threshold and limit.
4. Optionally return plain strings with textOutput.

Create a document

import { spotlight } from "@koda.oss/glyph";

const doc = spotlight.New(
  "Goodbye moon under quiet stars. Pasta recipe with tomato. Goodbye moon again.",
);

console.log(doc.Size()); // number of chunks

Rank all chunks

import { Create, spotlight } from "@koda.oss/glyph";

const doc = spotlight.New("...");
const probe = Create("Goodbye moon under quiet stars").glyph;

const results = doc.Rank(probe);

for (const hit of results) {
  console.log(hit.score.toFixed(2), hit.text);
}

Each result is a GlyphSpotlightResult with text, glyph, score, and comparison.

Query with threshold

const hits = doc.Query(probe, {
  threshold: 0.15,
  limit: 5,
});

Group probes

Pass a GlyphGroup to match multiple examples (emails, URLs, phrases):

import { CreateGroup } from "@koda.oss/glyph";

const hits = doc.Query(
  CreateGroup(["user@example.com", "support@example.com"]),
  { threshold: 0.2, limit: 10 },
);

Group probes use GroupResultAggregatorSum by default.

textOutput

Return chunk text only, sorted by score:

const snippets = doc.Query(probe, {
  threshold: 0.1,
  limit: 3,
  textOutput: true,
});
// string[]

Try it in the terminal

npm run demo -- spotlight ./docs/core/index.md "LSH banding"