API documentation

Overview

ASI crawls the English-language web itself and builds its own index, without relying on another search provider's results. The API gives your agents and retrieval pipelines direct access to that index.

Base URL

https://api.asi.xynatec.com

Requests

HTTPS GET requests. Responses are JSON.

Authentication

None at the moment. No API key or account is needed.

Pricing and SLA

No pricing and no service-level agreement are published. All callers share the same capacity.

Semantic search, native MCP access and pay-per-query with x402 are planned and not available yet. See what's next.

Quick start

Each example searches for forest carbon and asks for five results. Run them on a server or in a terminal, not in a web page; see Browsers & CORS.

curl

curl "https://api.asi.xynatec.com/v1/search?q=forest+carbon&limit=5"

Python

Uses only the standard library. Set a User-Agent that names your application: requests with Python's default Python-urllib user agent are currently blocked with 403 before they reach the API.

import json
import urllib.parse
import urllib.request

params = urllib.parse.urlencode({"q": "forest carbon", "limit": 5})
request = urllib.request.Request(
    f"https://api.asi.xynatec.com/v1/search?{params}",
    headers={"User-Agent": "my-research-agent/1.0"},
)

with urllib.request.urlopen(request, timeout=10) as response:
    data = json.load(response)

for result in data["results"]:
    print(result["title"], result["url"])

JavaScript (Node.js 18 or later)

// search.mjs: run with `node search.mjs`
const params = new URLSearchParams({ q: "forest carbon", limit: "5" });
const response = await fetch(`https://api.asi.xynatec.com/v1/search?${params}`);

if (!response.ok) {
  throw new Error(`ASI search failed with status ${response.status}`);
}

const data = await response.json();
for (const result of data.results) {
  console.log(result.title, result.url);
}

Search

GET https://api.asi.xynatec.com/v1/search?q=forest+carbon&limit=10&page=2

Returns one page of results for a query. All parameters go in the query string, URL-encoded.

Unknown parameters are ignored. If a parameter appears more than once, the last value counts. Parameter values never cause an error: the API does not answer with 400.

Pagination

Results are ranked, and page selects a slice of that ranking, starting at (page − 1) × limit. With limit=10, page 1 holds results 1 to 10 and page 2 holds results 11 to 20. Keep limit the same while you page through a query; changing it shifts the slices.

• Up to 10 pages are available per query, so at most 500 results with limit=50.
• The response has no total count. A page with fewer results than limit is the last one, and a page past the end returns an empty list.
• A request for page 11 or higher returns page 10. The page field in the response always shows the page that was served.

Query syntax

Every word of a query must appear in the page's title, description or main text. Matching ignores case and uses English stemming, so forests also finds forest. The first 8 KB of each page's main text are searchable.

Some characters have a special meaning: quotation marks and apostrophes, +, -, <, > or / at the start of a word, colons, parentheses, square and curly brackets, ^, *, ~, backslashes and backticks. The uppercase words AND, OR and NOT are operators. A word followed by a colon is read as a field name; unless it names an indexed field, the part after it is dropped, so note: forest searches without forest.

To search free text from users or models as plain words, convert it to lowercase and replace every character that is not a letter or digit with a space. Matching already ignores case and punctuation, so the words searched stay the same. Malformed syntax, such as an unclosed quotation mark, never causes an error: parts the API cannot read are dropped, so results can be broader than intended, or empty.

Ranking

Results are ordered by text relevance, scored with BM25. A match in the title counts three times as much as a match in the main text, and a match in the description one and a half times as much. No other signals, such as links, popularity or publication date, are used at the moment.

The index contains pages detected as English. It changes as the crawler finds new pages and revisits known ones, so results for the same query can change over time.

Response

A successful search returns 200 with a JSON object and Content-Type: application/json.

Illustrative example. The values are not taken from the live index.

{
  "query": "forest carbon",
  "page": 1,
  "results": [
    {
      "title": "How forests capture and store carbon",
      "url": "https://example.org/forests-and-carbon",
      "snippet": "Forests store carbon in trees, roots and soil.",
      "source": "example.org",
      "date": "2026-09-18"
    },
    {
      "title": "Measuring carbon in managed woodland",
      "url": "https://www.example.com/woodland/carbon",
      "snippet": "Forest inventories estimate carbon from tree height and trunk width.",
      "source": "www.example.com",
      "date": null
    }
  ]
}

Response fields

Result fields

Results never contain the full text of a page.

Errors and limits

Errors from the API itself are JSON objects with a short code in error. Responses from the infrastructure in front of it have no JSON body, so check the status code before parsing.

After a 503, wait for the time in Retry-After and try again. Retry a few times at most, wait longer each time, and don't send retries in parallel: the capacity is shared with everyone else. A 500 may succeed on a second try; if it keeps failing, please let us know.

Limits

• Query length: 512 characters. Longer queries are cut, not rejected.
• Results: up to 50 per request and 10 pages per query.
• Time: 5 seconds per search.
• Capacity: 32 searches at the same time, shared by all callers.

Browsers & CORS

The API sends no CORS headers, so browsers block calls from web pages on other origins. Call it from your backend, a serverless function or your agent's runtime, and pass the results on from there.

Health check

curl https://api.asi.xynatec.com/health

GET /health returns 200 with the plain-text body ok while the API is running. It does not run a search, so it says nothing about current search capacity.

Crawler

Results come from AsiBot, ASI's own crawler. It follows robots.txt and the robots directives that site owners set, and results only ever share titles, links and snippets. The AsiBot page explains how the crawler behaves and how to opt out. Read more about our approach to crawling.

Contact

Questions about integrating ASI, or feedback on the API: service@xynatec.com.