Web Search

POST /v1/web-search

Search the web and get structured results with titles, URLs, and metadata. General search is served by Exa or Tavily. Set type to academic to search peer-reviewed papers instead, or pin X-Quantized-Provider to choose a provider yourself.

Headers

Request body

Choosing what to search

type says what kind of source you want, so you don’t have to know which provider
serves it. academic searches Consensus,
which indexes peer-reviewed papers.

type only picks a provider when you have not. It never overrides a header you sent,
and it never widens what your license is allowed to reach: if your license type is
pinned to a set of providers that excludes Consensus, type: "academic" returns a
400 rather than falling back to a general provider.

Any value other than general or academic is rejected with a 422 before a
provider is chosen, so no credits are spent.

Getting page content

Set include_text: true when you need something an agent can actually read. Without
it you get metadata only, which is enough to rank or cite a result but not to answer a
question from it.

How much text you get back depends on the provider, because each behaves differently:

Either way the text is truncated to max_characters. The default of 4000 keeps a
five-result search near 20KB; raise it if you are feeding a model with a large context
window, but note a single page can exceed 140,000 characters uncapped.

Requesting text does not cost extra. If you need the complete page rather than a
capped extract, use /v1/fetch with the URLs you got back.

Examples

curl -X POST https://api.quantized.us/v1/web-search \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Python 3.13 new features",
    "num_results": 3
  }'

import httpx

response = httpx.post(
    "https://api.quantized.us/v1/web-search",
    headers={"Authorization": "Bearer sk-quantized-YOUR-KEY"},
    json={"query": "Python 3.13 new features", "num_results": 3},
)
data = response.json()
for result in data["results"]:
    print(f"{result['title']}: {result['url']}")

Academic search

curl -X POST https://api.quantized.us/v1/web-search \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "does caffeine improve memory",
    "type": "academic",
    "num_results": 3
  }'

import httpx

response = httpx.post(
    "https://api.quantized.us/v1/web-search",
    headers={"Authorization": "Bearer sk-quantized-YOUR-KEY"},
    json={
        "query": "does caffeine improve memory",
        "type": "academic",
        "num_results": 3,
    },
)
for paper in response.json()["results"]:
    print(f"{paper['title']} ({paper['published_date']}) — {paper['id']}")

The response shape is identical to a general search. url is the paper’s page on
consensus.app, id is its DOI when it has one, and text is the abstract:

{
  "results": [
    {
      "title": "How Does Caffeine Influence Memory?",
      "url": "https://consensus.app/papers/how-does-caffeine-influence-memory-.../",
      "id": "10.31234/osf.io/5rf6x",
      "published_date": "2021-05-13",
      "author": "Ruochong Zhang, C. Madan",
      "text": "Caffeine is a widely used nootropic drug...",
      "score": null
    }
  ],
  "usage": {
    "search_time": null,
    "credits_used": 1000000
  }
}

Response

{
  "results": [
    {
      "title": "What's New in Python 3.13",
      "url": "https://docs.python.org/3/whatsnew/3.13.html",
      "id": "r1",
      "published_date": "2025-01-15",
      "author": "Python Software Foundation",
      "text": "Python 3.13 is the latest stable release...",
      "score": 0.94
    },
    {
      "title": "Python 3.13 Release Notes",
      "url": "https://www.python.org/downloads/release/python-3130/",
      "id": "r2",
      "published_date": "2024-10-07",
      "author": null,
      "text": null,
      "score": null
    }
  ],
  "usage": {
    "search_time": 0.42,
    "credits_used": 8000
  }
}

Response fields

Every field except title and url can be null, and which ones are populated depends
on the provider. Exa supplies published_date and author; Tavily supplies score
and a snippet in text but never published_date or author on a general search;
Consensus supplies the abstract in text but never score. Treat all of them as
optional rather than assuming a provider.

Errors

A failed search is not billed.