§APISearch APILive · Builder

▮ Built for AI

One web index.Six endpoints.

The NOSIBLE Search API turns the live web into ranked, grounded answers for AI agents.

GET API KEY API REFERENCE

▮ What you can ask it

Geopolitical risk
Iran · 2026·03
"Where does Brent settle if Hormuz closes for 14 days?" → fast-search.
Company risk
Theranos · 2014·12
"List every public request for independent validation of the Edison." → bulk-search.
Liquidity risk
Evergrande · 2020·10
"Trace prior-lender pullback signals before the December 2021 default." → agent.

§01/ 05

▮ Endpoints

Index, crawl, and agent

Four ways to read the index as a search-engine results page, one route to fetch and parse a single URL, and one agent that explores the index on your behalf. Same crawl, same ranker, same retention discipline behind every call.

300K+Sources

BillionsWebpages

<1sFast-search p50

One stack

One web index. Six endpoints.

Pick the call shape that matches your traffic. Same crawl, same ranker, same retention discipline behind every one.

//01SERP

`/search/v2/fast-search`

Production SERP. Same hybrid-3 ranker, lifted caps, built for live application traffic.

//02SERP

`/search/v2/bulk-search`

Long-tail retrieval delivered out-of-band: up to ten thousand ranked results, encrypted.

//03SERP

`/search/v2/rich-search`

Fast-search plus opt-in enrichments per hit: profile, signals, embeddings, no second call.

//04Crawl

`/search/v2/scrape-url`

Turn any URL into clean, structured JSON: same extraction the index runs.

//05Agent

`/search/v2/search`

Cybernaut-1 writes the queries for you. Agentic SERP, ranked search results back.

§02/ 05

▮ Index endpoints

Ranked search results. The SERP shape your code already speaks.

Free, fast, bulk, and enriched: same request shape, four different cost / depth profiles. Pick the one that matches your traffic, scale up later without rewriting the call site.

Caps · max results × max probes

Endpoint	n_results	n_probes
`fast-search`	10 – 100	5 – 50
`bulk-search`	1,000 – 10,000	5 – 300

Defaults: fast-search 10 / 30 · bulk-search 1,000 / 10. Source-verified against the OpenAPI spec.

//01SERPPOSTHeader · Api-Keyschema · FastSearchSchema

`/search/v2/fast-search`

The everyday workhorse. n_results scales to 100, n_probes to 50, and n_contextify up to 1024, enough to feed a long-context model without follow-up calls. Sub-second p50 latency under Api-Key auth, ready for production RAG.

Request body

questionstring · required1–500 chars.
algorithmenum · hybrid-3Default workhorse ranker. Lexical, baseline, hamming, hybrid-1/2/3, and company are supported.
n_resultsint · 10–100 · 10Up to a hundred hits per call.
n_probes · filters · expansions · …—Full parameter list. See the OpenAPI spec.

Response

Inline ranked results.

Best for

Live retrieval inside an LLM application, RAG pipelines, dashboards, and anything where you want a real page of results in one round trip.

Open in spec ↗

//02SERPPOSTHeader · Api-Keyschema · SlowSearchSchema

`/search/v2/bulk-search`

When you want thousands of results in one query (backfills, factor construction, dataset assembly), bulk-search runs deep retrieval (up to 300 shards, up to 10,000 results) and ships the answer as an encrypted, zstd-compressed JSON download.

Request body

questionstring · requiredSame as the SERP endpoints.
algorithmenum · hybrid-3Same workhorse ranker as the live SERP endpoints.
n_resultsint · 1,000–10,000 · 1,000Long-tail window: up to ten thousand ranked results.
n_probes · filters · expansions · …—Full parameter list. See the OpenAPI spec.

Response

{
  "download_from": "<S3 URL>",
  "decrypt_using": "<Fernet key>"
}

Encrypted, zstd-compressed JSON download. Fetch, decrypt, decompress.

Best for

Backtests, training-set assembly, monthly analyst sweeps, and any workload that wants the full long tail rather than the top page.

Open in spec ↗

//03SERPPOSTHeader · Api-Keyschema · RichSearchSchema

`/search/v2/rich-search`

Extends fast-search with opt-in enrichments (publisher profile, ad-targeting cues, historical track record, quantitative signals, and 1-bit-quantised base64 embeddings) so downstream agents can score or rerank without a second call.

Request body

…all FastSearch fields—Same retrieval contract.
enrich_profilebool · trueAttach website / publisher profile.
enrich_targetingbool · trueAttach ad-targeting metadata.
enrich_historybool · trueAttach publishing track records.
enrich_signalsbool · trueAttach quantitative signals.
enrich_vectorsbool · trueAttach 1-bit-quantised base64 embeddings.

Response

Enriched ranked results: base fields plus whichever enrichment blocks you requested (see spec).

Best for

Multi-stage agents that grade sources, sentiment and factor pipelines, and any flow where you would otherwise glue several enrichment APIs around fast-search.

Open in spec ↗

§03/ 05

▮ Crawl

Turn any URL into clean, structured data.

The single-URL counterpart to the index endpoints. Hand the crawler a link and it returns the same structured extraction the index runs on every document it ingests, ideal for grounding LLM applications and resolving citations agents just produced.

//01CrawlPOSTHeader · Api-Keyschema · VisitSchema

`/search/v2/scrape-url`

Turn any URL into clean, structured JSON, the same extraction the NOSIBLE index runs on every document it ingests. One schema across the 300K+ sources we already crawl. Built for LLM applications and agents that need grounded sources, not raw HTML.

Convertsany HTML page into structured JSON: article text, metadata, authors, and publish dates.
HandlesJavaScript-rendered pages via render: true; forces a fresh crawl via recrawl: true when the cached copy is stale.
Powersevery NOSIBLE retrieval endpoint, so what scrape-url returns is what the index ranks against.

Request body

urlstring · requiredThe URL this route will use for processing.
renderbool · falseForce the crawler to render the JavaScript.
recrawlbool · falseWhether to re-crawl the url again. Optional.
htmlstringThe HTML associated with this URL. Optional.

Response

Parsed page object: extracted text, metadata, and link graph (see spec for the full result shape).

Best for

Resolving a citation the agent just produced, point-checking a single source, or pulling a fresh extract for a page the index has not seen yet.

Open in spec ↗

▮ Field note

When you know the query, hit the index. When the query is the thing you need to discover, the agent writes it for you, and hands back the same ranked results.

— NOSIBLE · 2026

§04/ 05

▮ Agent endpoint

You bring the question. The agent writes the queries.

Agentic SERP, not deep research. Cybernaut-1 plans multi-angle strategies, fans out parallel probes, and returns the same ranked search results as the SERP endpoints: the query is what the agent authored for you.

//01AgentPOSTHeader · Api-Keyschema · SearchSchema

`/search/v2/search`

Hand the agent a freeform prompt and Cybernaut-1 writes the queries for you. It plans multi-angle strategies, fans out parallel probes, and deep-dives the promising branches, then returns ranked search results, not a synthesised summary. Same response shape as the SERP endpoints, different planner.

Request body

promptstring · required25–2,500 chars. Describe what you need to know.
agentenum · cybernaut-1Currently only "cybernaut-1".

Response

Same ranked-result shape as the SERP endpoints: the agent authored the queries, you read the results.

Best for

Open-ended retrieval where the right query is not known in advance. Let the agent author the queries and return the same ranked results your SERP code already consumes.

Open in spec ↗

§05/ 05

▮ Client packages

Skip the curl. Use the SDK.

Official clients wrap every endpoint with typed bindings, retries, and key handling. Python for research and pipelines, TypeScript for browsers, Node, and edge runtimes.

Pythonpkg · nosible

Install

pip install nosible

Example

from nosible import Nosible

with Nosible(nosible_api_key="YOUR API KEY") as client:
    results = client.fast_search(
        question="What is Artificial General Intelligence?"
    )
    print(results)

Set NOSIBLE_API_KEY in your environment (format: nos_sk_…) or pass it directly to the client.

nosible-py docs ↗

TypeScript · JavaScriptpkg · nosible

Install

npm install nosible

Example

// ES Module
import { /* exported items */ } from "nosible";

// CommonJS
const { /* exported items */ } = require("nosible");

TypeScript client for the NOSIBLE AI Search API. Imports listed; for the full surface see the package’s TypeDoc / npm listing.

nosible on npm ↗

Raw HTTPpkg · curl

Install

curl --version  # any 7.68+ build

Example

curl -X POST https://www.nosible.ai/search/v2/fast-search \
  -H "Api-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question":"What is Artificial General Intelligence?","n_results":10}'

Replace YOUR_API_KEY with a real key from app.nosible.com. Same request shape across every endpoint: swap "fast-search" for "rich-search", "bulk-search", "scrape-url", or "search" to switch routes.

OpenAPI spec ↗

▮ Get started

START YOUR 90-DAY TRIAL.

Review the products and integration materials, then request the data and delivery method your team needs.

START TRIAL GET API KEY

▮ API v2.1.0 · spec last seen 2026·06

Endpoint

n_results

n_probes

fast-search

10 – 100

5 – 50

bulk-search

1,000 – 10,000

5 – 300

from nosible import Nosible with Nosible(nosible_api_key="YOUR API KEY") as client: results = client.fast_search( question="What is Artificial General Intelligence?" ) print(results)

curl -X POST https://www.nosible.ai/search/v2/fast-search \ -H "Api-Key: YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"question":"What is Artificial General Intelligence?","n_results":10}'