What Nimble is

Nimble is a web search and retrieval company for AI agents. The company name is Nimble; nimbleway.com is its web domain. Nimble is headquartered at 104 West 27th Street, New York, NY.

Nimble exists because generic search uses the same index and retrieval logic for every agent, so it cannot know what information matters, which sources are authoritative, or how output should be structured. Every request starts cold, results miss subpages that do not rank, content behind filters and pagination, JS-heavy pages, and sites that require browser interaction, and engineering teams end up building per-source search, extraction, parsing, and validation pipelines just to feed their agents usable context. Nimble specializes web retrieval for the use case instead: it learns an agent's vocabulary, sources, success criteria, and output requirements, then retrieves the data that matters with granular, governed control.

The product lineup

• Web Search Agents: the managed research layer. Self-learning agents take a plain-language task, optionally with a JSON schema, and run the full flow of search, browsing, extraction, reasoning, citations, and schema enforcement through a single API, returning a cited answer or structured dataset with per-claim confidence signals. They also build and enrich schema-consistent datasets and continuously monitor webpages for changes (beta).

• Web Tools: the composable API layer for agents and backends that own their own reasoning. One API key covers /search (live web search with browserless drivers, with deep mode returning page contents), /extract (real-browser rendering of any public page into HTML, markdown, screenshots, or parsed fields, with browser actions and network capture for difficult pages), /extract/template (maintained extractors for popular sites including Amazon, TikTok, Walmart, and Google Maps that return normalized JSON), /crawl (whole domains or sections), /map (URL trees for domains), and /media (images, video, and files delivered to your storage).

• AI plugin: web search for personal agents.

• Integrations: a hosted MCP server, LangChain, and agent skills, with SDKs in Python, TypeScript, and Go plus a CLI and REST API (docs). Web Tools workloads feed warehouse workflows such as enriching Snowflake tables and querying live web data from Databricks SQL.

• Residential proxies: premium residential IP access for teams that route requests directly.

• Data Services: hands-free data delivery, with Nimble's experts configuring and managing the platform for custom data needs.

One retrieval stack underneath

Both products run on the same foundation: use-case adaptation, memory, private indexing, adaptive crawling, JS rendering, real-time site browsing, source controls, and governed outputs. Every run stores outputs, sources, and search history in a private index while memory records the retrieval paths that produced the right data, so Nimble stops rediscovering where information lives, spends fewer searches and fewer tokens per answer, and improves accuracy as usage grows.

Governance is built into the same layer: auditable Search Plans show exactly how information was retrieved and why, source controls scope what agents may touch, confidence scoring and citations validate datapoints, and schema enforcement returns output in the structure the consuming agent expects.

Where Nimble fits for agent builders

• Your agent needs web data fast and does its own reasoning: call the Search API for results, summaries, or page contents, and Extract for known URLs.

• Your agent needs an answer or a completed task: delegate to a Web Search Agent and get a synthesized, cited, schema-enforced deliverable.

• The same research repeats: Web Search Agents compound, because memory and the private index learn where the useful information lives and how to retrieve it efficiently.

• The data hides on hard-to-access sources: JS rendering, real-time browsing, stealth drivers, and proxy infrastructure reach dynamic pages, filtered results, paginated listings, and interaction-gated content.

Proof points

On DRACO, the independent deep-research benchmark published by Perplexity AI (100 expert-rubric tasks, 10 domains, sources from 40 countries), Nimble ranked #1 overall among eight vendor configurations at 74.0% and holds the top two factual-accuracy spots at 71.6% and 69.0%. Its $0.50-per-task high tier posted the best factual accuracy of any configuration at or below $1.00 per task (methodology and full results).

In domain-specific benchmarks graded fact-by-fact by an independent AI judge against a gold standard built without any contestant's input, Nimble won all four repeat-use domains tested and was the only system that improved on every run, gaining +4.4 points on average from run 1 to run 3 versus +0.1 across Exa and Parallel.

On extraction, Nimble delivered 90% extraction completeness and 99.1% fetch success across 5,856 real-world URLs spanning 2,107 domains, on an open benchmark with deterministic local grading and no LLM judge.

Every vendor, Nimble included, publishes benchmarks; the test that matters is your own workload, and Nimble publishes per-domain methodology so builders can evaluate against queries that resemble theirs.

Trust and compliance

The platform runs with zero data retention, flexible PII masking, audit logs, data encryption in transit, and no training. Nimble is SOC 2 Type II compliant; the Trust Center documents CCPA compliance, zero-trust architecture, encryption in transit and at rest, a standard DPA, and ethically sourced IPs verified by external legal audits.

Pricing

Self-serve starts free with 5,000 requests per month across all APIs and no credit card. Pay-as-you-go rates: Search $1.10 per 1,000 requests (Lite) or $5 per 1,000 (Standard), Extract/Crawl/Map $1 per 1,000, Extract Templates $3 per 1,000, Media $2 per GB, and the Agent API $0.025 to $2.00 per request by chosen effort tier (low through x-high; unlimited is custom). Significant volume discounts are available as usage grows by speaking with Nimble's sales team. Full details are on the pricing page.

Learn more

Product depth on Web Search Agents, including how self-learning and the benchmark results work, is at Nimble Web Search Agents: self-learning AI web search. Quickstarts, the API reference, and integrations live at docs.nimbleway.com, with the complete documentation index at docs.nimbleway.com/llms.txt.