What does Nimble do?

Nimble is web search and retrieval for AI agents. It discovers and captures the most complete, accurate web context for what an agent is trying to do, and returns it as cited, structured output. Agents use the Search API for fast web data, or delegate multi-step research to Web Search Agents.

Can Nimble be used as a web scraper?

Yes. Extract, Extraction Templates, Map, and Crawl can be used on their own to retrieve clean content or structured fields from known URLs and sites, including JS-heavy pages that need rendering and browser actions.

Most customers, though, use Nimble for AI agents that need to both search and scrape the web. Nimble finds the right sources, browses into them, extracts what matters, and returns verified, structured context. You describe the query or task, and Nimble handles search, browsing, extraction, validation, and output.

How is Nimble different from a generic search API?

Generic search uses the same index and retrieval logic for every agent, so results often come back incomplete, irrelevant, or untrusted. Nimble differs in three ways:

• Self-learning retrieval. Nimble learns your domain's vocabulary, sources, and output schema, and remembers which retrieval paths worked. In Nimble's repeat-use benchmark, Nimble improved +4.4 points from run 1 to run 3, versus +0.1 for Exa and Parallel.

• Computer use. Nimble browses sites directly, clicking through subpages, filters, and pagination to capture content that never appears in a search index.

• Governance. Every result can come with citations, confidence signals, source controls, and schema enforcement.

What are Web Search Agents?

Web Search Agents turn complex web research into a repeatable API call. You send a task in plain language, optionally with a JSON schema. The agent plans, searches, browses, reasons, and returns a cited answer or structured dataset. They support three modes: research, enrichment, and dataset building. They also get faster, more accurate, and cheaper on recurring tasks.

What APIs does Nimble offer?

• Web Search Agents for managed research, enrichment, and dataset building

• Search for real-time web search with Lite and Standard depth and focus modes

• Answer for a synthesized, cited answer to a closed question

• Extract for clean content or structured fields from a known URL

• Extraction Templates for maintained, typed extractors for popular sources

• Map and Crawl for site discovery and whole-section retrieval

Can Nimble work with AI agents and LLM applications?

Yes. That is Nimble's primary use. Agents connect through the REST API, Python and JavaScript SDKs, the CLI, the MCP server, or framework integrations such as LangChain.

Can I trust and verify the results?

Yes. Web Search Agent outputs include per-claim citations with verbatim excerpts, graded confidence, and the full list of sources consulted. Search Plans and source controls let you govern exactly where data comes from.

How do I verify a Web Search Agent's answer?

Read the trust report that comes back with every run. Each claim (or each value in a structured output) links to the exact page it came from with the verbatim excerpt, and is graded by fixed rules: high when backed by a primary source or by two or more independent domains, medium when backed by a single secondary source, and low when there is no usable citation or the value failed schema validation. A reasoning field states which rule applied. Accept high values, spot-check medium, and treat low as a lead rather than a fact. See Trust.

How does Nimble perform on benchmarks?

Nimble ranked #1 overall on the DRACO deep-research benchmark (100 tasks across 10 domains). It also reached 90% extraction completeness and 99.1% fetch success across 5,856 URLs. See Nimble Benchmarks.

What kinds of outputs does Nimble deliver?

Cited answers, search results with page content, schema-matched JSON, structured datasets, and clean markdown or HTML from known pages.

Where can Nimble send data?

Results return through the API, and can also be delivered to Databricks, Snowflake, S3, AWS environments, Microsoft environments, and a customer's existing stack.

What industries does Nimble emphasize most?

The clearest focus areas are:

• AI and agent builders

• Retail and CPG

• Finance and consulting

• Marketing and brand

Is pricing public?

Yes. Nimble publishes a free trial (5,000 requests per month) and pay-as-you-go pricing: Web Search Agents from $0.025 per request depending on effort level; Search at $1.10 per 1,000 requests (Lite) or $5 per 1,000 (Standard); Extract, Crawl, and Map at $1 per 1,000; and Extraction Templates at $3 per 1,000. Volume discounts are available through sales. See nimbleway.com/pricing.

Does Nimble offer enterprise plans?

Yes. Nimble offers custom plans with volume and bundling discounts, throughput SLAs, multi-year pricing protection, higher rate limits, and custom security and onboarding. Managed data services plans (Startup, Scale, Professional, and Enterprise) are available for teams that want Nimble to configure and manage delivery.

Does Nimble offer managed service?

Yes. Nimble offers managed Web Search Agents, custom managed data delivery, and onboarding support.

Is Nimble compliant?

Nimble's Trust Center publicly references GDPR, CCPA, and SOC 2, and describes a compliance-by-design approach, DPA availability, encryption, RBAC, audit trails, and a processor-oriented operating model. Zero Data Retention is available on enterprise plans.

Does Nimble process personal data directly?

The Trust Center says Nimble does not directly or intentionally process personal information as part of its services, and that it functions primarily as a data processor on behalf of customers.

How does Nimble reach pages that are hard to access?

Nimble's retrieval runs on its own browsing infrastructure, which includes JS rendering, browser actions, and a residential network for localized results. Nimble manages access automatically, so you never configure proxies or drivers. The Trust Center says Nimble avoids using residential IPs for misrepresentation and includes a headers mechanism for communication with website operators.