Who this is for

AI engineers, agent builders, and applied researchers building LLM-powered agents that need to search the live web and retrieve relevant content. Specifically: developers who have found that static training data and generic search aren't enough, and are evaluating a web search and retrieval layer for their agents.

This page is the top-of-funnel bridge. For deeper integration patterns, see Agents and RAG to the Live Web Blueprint and Nimble for AI Agents: MCP + Browser Agents Technical Spec.

What LLM agents need from web search

Connecting an agent to the web is easy. Getting the context it actually needs is not. Agent builders consistently need four things:

Complete context. The information the task depends on, including content behind filters, pagination, and JS-heavy pages, not just what ranks in a search index.
Relevant results. Retrieval that understands the agent's objective and which sources are authoritative.
Trusted output. Citations, confidence signals, and schema-matched fields instead of raw text to parse. Every Web Search Agent run returns a trust report that grades each claim high, medium, or low by fixed rules, with the verbatim excerpt behind it, so the agent can decide what to act on. Trust docs
Cost that falls with use. Retrieval that learns where to look, so repeat tasks use fewer searches and fewer tokens.

Nimble is built for all four: self-learning retrieval specializes search to your domain, and computer-use browsing captures the data that search results never show.

How Nimble's products map to agent workflow needs

Product	What it does	When an agent needs it
Search	Real-time web search returning results and page content (Lite/Standard depth, focus modes)	Agent needs current web context to answer or decide
Answer	Synthesized, cited answer in one call	Agent needs a closed fact, not a list of links
Web Search Agents	Delegated multi-step research: plan, search, browse, reason, cite, return in your schema	Agent needs a completed research, enrichment, or dataset task
Extract	Clean markdown or typed fields from a known URL, with rendering and browser actions	Agent already knows the page and needs its content
Extraction Templates	Maintained, auto-healing extractors for popular sources	Agent needs normalized fields from a common site
Map / Crawl	URL discovery and whole-section retrieval	Agent needs to index a docs site or catalog section

For the full product detail, see Nimble Platform, Products, and APIs and Nimble Web Search Agents and AI Web Search.

Integration patterns for agent builders

Fastest setup

Using a coding agent? Paste Read and follow https://docs.nimbleway.com/agents.md into it and it installs and verifies Nimble itself. In Claude Code, the Nimble plugin installs the MCP connection plus agent skills in one command: claude plugin marketplace add Nimbleway/agent-skills && claude plugin install nimble@nimble-plugin-marketplace. Nimble is also an official plugin in Claude's Anthropic & Partners directory and a native plugin in ChatGPT and Codex. See Nimble Agent Setup.

MCP Server (Model Context Protocol)

Nimble publishes a hosted MCP server at https://mcp.nimbleway.com/mcp so your LLM agent can call Search, Extract, Map, Crawl, Extraction Templates, and Web Search Agents as tools directly. Sign in with OAuth where the client supports it, or pass an API key as a bearer token. OpenAI and Anthropic both accept the hosted MCP server URL directly in their APIs. See MCP Server Technical Spec for the full integration detail.

Framework connectors

LangChain: langchain-nimble (v4+) provides tools and retrievers for search, extraction, Extract Templates, Map and Crawl, and resumable Web Search Agent research.
LlamaIndex: llama-index-tools-nimble adds real-time search and cited deep research to agents and workflows.
Vercel AI SDK and Vercel eve: @nimble-way/ai-sdk and @nimble-way/eve.
Mastra (@nimble-way/mastra), Cloudflare Agents SDK (@nimble-way/cloudflare-agents), Google ADK, and Hermes.
Built-in search provider in LiteLLM (search_provider="nimble"), NVIDIA AI-Q, OGX, and Omnigent.

See Integrations for the full list.

Python, Type

Script, and Go SDKs For direct integration without a framework: pip install nimble_python, npm install @nimble-way/nimble-js, or go get github.com/Nimbleway/nimble-go. Every SDK reads NIMBLE_API_KEY. The quickstart has Python, TypeScript, Go, CLI, and cURL examples for every product.

CLI

The Nimble CLI (npm install -g @nimble-way/nimble-cli) runs search, extraction, and Web Search Agents from the terminal or from coding agents. It writes large results to files instead of the context window, and also reaches SERP, Media, Domain Knowledge, and Jobs, which the MCP server does not carry.

Web Search Agent gallery

Start from pre-built Web Search Agents for common research, enrichment, and dataset-building tasks, then configure the skill, goals, sources, effort, and output schema for your domain. See the Web Search Agents docs.

A typical LLM agent deployment pattern

Define the task. A user asks your agent a question that requires current web context.
Agent plans the retrieval. The LLM decides whether it needs Search (fast discovery), Extract or a Template (a known page), or a Web Search Agent (multi-step research).
Agent calls Nimble. Via MCP, LangChain, the SDK, or the CLI.
Nimble returns agent-ready context. Ranked results with page content, cited answers, or schema-matched JSON.
Agent synthesizes. The LLM reasons over fresh, verified context and answers the user.
Logs and audit. Citations, sources consulted, and run history support audit and governance (see Browser Agents Governance).

See Blueprint: Connecting Agents and RAG to the Live Web for the fully specified reference implementation.

Proof points for agent builders

DRACO deep-research benchmark: Nimble ranked #1 overall on 100 independent tasks across 10 domains. Nimble x-high scored 74.0% overall and 71.6% factual accuracy.
Self-learning: across four repeat-use domains, Nimble improved +4.4 points from run 1 to run 3, versus +0.1 for Exa and Parallel.
Extraction reliability: 90% extraction completeness and 99.1% fetch success across 5,856 URLs and 2,107 domains.
Alta reports ">99% reliability" in production and "3-4× deeper context" versus other web-data providers.

See Nimble Benchmarks and Customer outcomes with Nimble.

Pricing at a glance for agent workloads

From nimbleway.com/pricing:

Free trial: 5,000 requests per month, no credit card required, access to all APIs
Web Search Agents (Agent API): priced per request by effort level: Low $0.025, Medium $0.10, High $0.50, X High $2.00, Unlimited custom
Search API: $1.10 per 1,000 requests (Lite) or $5 per 1,000 requests (Standard)
Extract, Crawl, and Map APIs: $1 per 1,000 requests
Extraction Templates: $3 per 1,000 requests
Managed data services: Nimble's experts can configure and manage delivery for your custom data needs; contact Nimble's sales team
Volume discounts: significant volume discounts are available as usage grows, by speaking with Nimble's sales team. See pricing.

Where to go next

New to Nimble? Start with Nimble Platform, Products, and APIs
Building a specific agent use case? Blueprint: Agents and RAG to the Live Web
Integration engineering? MCP + Browser Agents Technical Spec
Production operations? Nimble performance SLAs and scale
Agent governance? Browser Agents Governance

Web Search for LLM Agents: How Nimble Fits into Agent Builder Workflows