Skip to main content
A snippet is a verbatim passage extracted from a source page that is relevant to your query. Each search result can contain multiple snippets.

​
Snippet object

{
  "text": "Quantum error correction improves exponentially as the system scales, a key milestone toward fault-tolerant quantum computing.",
  "relevance_score": 0.94
}
FieldTypeDescription
textstringVerbatim passage from the source page
relevance_scorefloatRelevance to your query, from 0.0 (low) to 1.0 (high)

​
Relevance score

The relevance score reflects how closely a snippet answers your specific query — not just how related the topic is. A score of 0.9+ means the passage directly addresses your query. A score below 0.6 means the passage is topically related but may not be a direct answer. Practical guidance:
  • For RAG pipelines: filter to snippets with relevance_score >= 0.7
  • For display: show all snippets — let users judge
  • For strict fact retrieval: use precise depth + filter to >= 0.85

​
Snippet count

Control how many snippets are extracted per result with max_snippets_per_result (default: 3, range: 1–5). 
{
  "query": "...",
  "max_snippets_per_result": 5
}
Note that the actual number returned may be lower than the maximum if the page doesn’t contain enough relevant passages.

​
The warning field

Each result has an optional warning field. It’s set when extraction succeeded but with caveats — for example, if the page content was too short or if some passages couldn’t be reconstructed verbatim. 
{
  "rank": 2,
  "url": "...",
  "snippets": [],
  "warning": "Found relevant passages but failed to reconstruct them verbatim."
}
When warning is set and snippets is empty, the result still appears in the list for transparency — you can decide whether to use the URL directly.