Fast Search

Retrieve ranked web evidence for a question with NOSIBLE's production hybrid search engine.

POST/search/v2/fast-search

Related NOSIBLE resources

research & data

Inspect the Search data dictionary or read NOSIBLE research for implementation context and methods.

Authorizations

apiKey

Api-Keystringrequiredheader

Your NOSIBLE API key. Create and manage keys in the API Dashboard.

Example: nos_sk_...

Body

application/json

questionstringbody

The question the search should answer. 1–500 characters; use this as the primary information need.

Example: What is driving AI chip demand?

algorithmstringbody

Search algorithm. hybrid-3 is the recommended default.

Example: hybrid-3

n_resultsintegerbody

Number of ranked results to return.

Example: 10

n_probesintegerbody

Number of shards to evaluate; Fast and Rich Search accept 5–50.

Example: 30

n_contextifyintegerbody

Context words returned per result. Fast and Rich Search accept 64–1024.

Example: 256

expansionsstring[]body

Optional semantically equivalent query expansions.

Example: ["AI semiconductor capex"]

sql_filterstringbody

Optional filter expression over the Search index.

Example: SELECT loc FROM engine WHERE words >= 30

instructionstringbody

Instruction passed to the embedding mode. Keep it short and purposeful; it is not a replacement for the question.

Example: Retrieve evidence about semiconductor demand.

min_similaritynumberbody

Reject results below this similarity threshold. Must be between 0 and 1; it is a score threshold, not a probability.

Example: 0.35

must_includestring[]body

Terms that must appear in each result. Use sparingly: every additional hard constraint can reduce recall.

Example: ["NVIDIA"]

must_excludestring[]body

Terms that must not appear in a result.

Example: ["job listing"]

brand_safetyenumbody

Content-safety class: Safe, Sensitive, or Unsafe.

Example: Safe

languageenumbody

Language classification used to refine retrieval, such as EN, FR, or DE.

Example: EN

continentenumbody

Continent filter for the indexed document.

Example: Europe

regionenumbody

Geographic region filter for the indexed document.

Example: Western Europe

countryenumbody

Country filter using the API's canonical country values.

Example: Germany

sectorenumbody

GICS sector filter.

Example: Information Technology

industry_groupenumbody

GICS industry-group filter.

Example: Semiconductors & Semiconductor Equipment

industryenumbody

GICS industry filter.

Example: Semiconductors

sub_industryenumbody

GICS sub-industry filter.

Example: Semiconductor Equipment

iab_tier_1enumbody

IAB tier-1 content category.

Example: Business

iab_tier_2enumbody

IAB tier-2 content category.

Example: Business Services

iab_tier_3enumbody

IAB tier-3 content category.

Example: Investing

iab_tier_4enumbody

IAB tier-4 content category.

Example: Stocks

companiesstring[]body

Up to three company names used to refine retrieval.

Example: ["NVIDIA", "TSMC"]

collectionenumbody

Collection to search. everything searches the full available collection; this-week restricts to recently added data.

Example: everything

deduplicatebooleanbody

Deduplicate results that represent the same headline or story.

Example: true

How to use this endpoint

guidance

Use Fast Search when you already know the question and want a fast, ranked evidence set. Start with hybrid-3, then add filters only when they express a real business constraint. Persist query with response so an investment decision can be reproduced later.

Response body

application/json

These are the fields you can build against. Nested names use dot notation; optional sections are called out in their descriptions.

messagestring

Human-readable summary of the work performed, including how many indexed records were evaluated and how long retrieval took.

Example: Evaluated 1,204,332 out of 15,311,040 records in 0.184 seconds.

queryobject

The effective query after defaults, filters, and search controls have been applied. Persist this object with results when you need reproducibility.

Example: { "question": "What is driving AI chip demand?", "algorithm": "hybrid-3", "n_results": 10 }

query.questionstring

Question used for retrieval.

Example: What is driving AI chip demand?

query.algorithmstring

Scoring algorithm actually used.

Example: hybrid-3

query.n_resultsinteger

Effective result cap.

Example: 10

query.n_probesinteger

Effective shard/collection probe count.

Example: 30

query.n_contextifyinteger

Context size used for returned snippets.

Example: 256

responseobject[]

Ranked result documents. Results are ordered by the selected algorithm's relevance score.

Example: [{ ... }]

response[].url_hashstring

Stable internal document hash. Use it as an opaque identifier; do not construct URLs from it.

Example: a1b2c3d4...

response[].urlstring

Canonical URL of the matching document.

Example: https://www.reuters.com/technology/...

response[].netlocstring

Source host extracted from the canonical URL.

Example: reuters.com

response[].publisheddate-time | null

Indexed publication timestamp when the source provided one.

Example: 2026-07-20T12:00:00Z

response[].visiteddate-time | null

Timestamp at which NOSIBLE fetched the document.

Example: 2026-07-20T12:04:11Z

response[].languagestring

Indexed document language classification.

Example: en-en

response[].authorstring | null

Author metadata when available.

Example: Jane Doe

response[].titlestring

Document title as indexed by NOSIBLE.

Example: Chipmakers increase AI infrastructure investment

response[].descriptionstring

Indexed page description or metadata summary when available.

Example: A review of data-center investment trends.

response[].best_chunkstring

Highest-scoring context window for this result.

Example: Demand for accelerated computing...

response[].contentstring

Contextified text returned for the result. Its size follows n_contextify.

Example: Demand for accelerated computing...

response[].semanticsobject

Semantic ranking metadata for the result.

Example: { "similarity": 0.9821 }

response[].semantics.similaritynumber

Semantic similarity score for the result. Use it for ranking and thresholding, not as a probability.

Example: 0.9821

query.instructionstring

Instruction used to guide retrieval.

Example: Retrieve evidence about semiconductor demand.

query.expansionsstring[]

Effective query expansions used to broaden recall.

Example: ["AI accelerator demand", "data-center GPU spending"]

query.must_include/must_excludestring[]

Hard lexical inclusion and exclusion constraints applied to results.

Example: { "must_include": ["NVIDIA"], "must_exclude": ["jobs"] }

Responses and errors

HTTP

200Request succeeded. The response body is shown in the panel on the right.

400invalid_request — The request is syntactically valid JSON but a value, date, filter, or combination of fields is invalid.

401unauthorized — The required API key or Bearer credential is missing or invalid.

422validation_error — The request shape is understood but one or more values fail validation.

429rate_limited — The request exceeded the account or public-window rate limit. Respect Retry-After when supplied.

502upstream_error — An upstream retrieval service failed while processing the request.

504upstream_timeout — An upstream retrieval service did not respond within the request timeout.