> ## Documentation Index
> Fetch the complete documentation index at: https://docs.snipex.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# How It Works

> A look under the hood at Snippex's extraction pipeline

Snippex turns a plain text query into ranked, verbatim relevant passages in a single API call. Here's what happens internally:

## The pipeline

<Steps>
  <Step title="Web search">
    Snippex searches the web for your query, applying any filters you've specified — recency, language, domain allow/block lists — and retrieves the top-ranked pages.
  </Step>

  <Step title="Content retrieval">
    For each result, Snippex fetches the full page content. This is the raw material for extraction — not just the short meta description you'd get from a typical search API.
  </Step>

  <Step title="Snippet extraction">
    Our extraction engine analyzes the full page content against your query and identifies the passages that are most relevant. These are verbatim excerpts — nothing is paraphrased or synthesized.
  </Step>

  <Step title="Ranking and scoring">
    Each snippet is scored for relevance to your query (0.0–1.0). Results are ranked by source quality; snippets within each result are ordered by relevance score.
  </Step>
</Steps>

## What you get back

For each search result, you receive:

* The **URL and title** of the source page
* A list of **snippets** — verbatim passages from that page, each with a relevance score
* Optional metadata: `publish_date`, `last_updated`
* A `warning` field if extraction partially failed for that result

## Grounded in source content

Every snippet is a verbatim quote from the original page. Snippex does not synthesize, summarize, or generate content. If a passage appears in the response, it exists word-for-word in the source.

This makes Snippex suitable for applications that require **attribution**, **citation**, or **factual grounding**.