Nimble launches Web Search Agents to cut AI research token costs

Generic web search was never built for AI agents, and Nimble is betting that the mismatch is costing companies real money. The New York startup rolled out Web Search Agents today, a product that trains itself on a customer's specific domain rather than treating every query the same way. A lead-enrichment bot and a market-research bot end up needing very different information, Nimble argues, so why hand them the same undifferentiated pile of search results and let the AI sort through the noise.

That sorting is expensive. When an agent gets back a wide, unstructured list of links, it burns tokens deciding what matters and chewing through pages that turn out to be irrelevant. Nimble's harness tries to shortcut that process by adapting its retrieval strategy to the task at hand, blending its own proprietary indexes with live scraping of current sites. The company's own benchmarks claim a 21-point jump in answer quality alongside a 51% drop in tokens burned per query — numbers that, if they hold up in independent testing, would matter a lot to anyone running agents at scale.

The pitch here isn't speed. Nimble is explicitly targeting agents that grind away for hours on business-critical research, the kind of job where missing one important source is far more costly than taking an extra thirty seconds. CEO Uri Knorovich frames the core problem as accuracy and cost, not latency, and the system reportedly checks its own outputs along the way to cut down on rework. Rox, an AI-native CRM company, says it saw a 20-fold reduction in token spend after switching over, while Qodo's head of product described using the tool to tune a Claude-based agent toward specific competitor signals instead of a generic market overview.

Nimble ships the product through an API, an SDK, and Model Context Protocol, with a free trial for developers who want to bolt it onto existing agents or build new applications — anything from quick lookups to deep research and structured dataset generation. The company says it already handles more than 90 million searches daily across a customer base spanning Fortune 500 firms and newer AI-native startups. Founded in 2021, Nimble has raised $75 million total, including a $47 million Series B in February led by Norwest Venture Partners with backing from Databricks Ventures, Target Global, and several other funds.