Search API v2.1

Search calls use the merged API origin and the api-key header.

Interactive Search

fast_search returns legacy-compatible Result objects while preserving the complete Search v2.1 response. rich_search returns RichResult objects and preserves each selected enrichment block.

from nosible import Nosible

with Nosible() as client:
    results = client.fast_search(
        question="What is changing in semiconductor capacity?",
        companies=["NVIDIA", "TSMC"],
        collection="this-week",
        deduplicate=True,
        n_results=25
    )

    rich = client.rich_search(
        question="What is changing in semiconductor capacity?",
        n_results=10,
        enrich_vectors=False
    )

Fast Search retains the SDK’s established defaults of 100 results and 128 context tokens, even though the service schema defaults are 10 and 256. Requests below the service minimum of 10 are truncated locally.

Scrape responses use lossless Snippet models. Unknown fields, omitted known fields, and explicitly null known fields retain their original wire representation during from_dict / to_dict round trips.

Asynchronous Search

bulk_search and time_search submit a task, poll the presigned result URL, decrypt the Fernet payload, and decompress current Zstandard or legacy gzip data. Authentication headers, including client-level authentication on an injected httpx.Client, are never sent to the download host. Transient download responses use the configured retry policy.

buckets = client.time_search(
    question="semiconductor investment",
    start="2026-01-01T00:00:00Z",
    end="2026-04-01T00:00:00Z",
    frequency="1w",
    n_results=100
)

Time Search composes the common Search schema with its time fields. It permits at most 500 intervals and 50,000 requested results per call. time_search preserves the downloaded interval envelope; bulk_search returns a ResultSet.

Endpoint coverage

The SDK covers all Search v2.1 routes:

Method	Purpose
`search`	Cybernaut agentic Search.
`fast_search`	Interactive ranked Search.
`rich_search`	Interactive enriched Search.
`bulk_search`	Long-running high-volume Search.
`time_search`	Independent searches across time intervals.
`scrape_url`	Structured page and snippet extraction.
`topic_trend`	Date-keyed topic prevalence.
`save_search`	Create or update a saved Search.
`get_searches`	List saved Searches.
`delete_search`	Delete a saved Search.
`get_limits`	Retrieve current key limits.

fast_searches is the SDK batch convenience for issuing several Fast Search requests. Question lists, list[Search], and SearchSet inputs all receive the shared filters and retrieval options supplied to the batch; non-null fields populated on an individual Search model retain precedence. This includes False for the optional certain filter. Expansion generation is enabled when either the batch or the individual model opts in. Requests run concurrently up to the client’s configured concurrency while results retain input order.