[DigitalToday reporter Chi-gyu Hwang (황치규)] Nimble has launched a "web search agent" to improve the accuracy of enterprise AI agent web research and reduce token costs. SiliconANGLE reported on July 29 that the product learns a customer's work domain and then carries out complex web research tasks on its own.

The product is focused on overcoming the limits of enterprise agents that deploy general-purpose web search in production environments. Nimble said general search returns broad, unstructured results, forcing agents to filter relevant information again. It said tokens are wasted in the process through unnecessary tool calls and the processing of less important pages.

Nimble said the same search product should not be applied to both a market research agent and a lead enrichment agent. With that in mind, it adopted a structure in which the system learns the knowledge work needed for the task on its own and adjusts search strategies to match the nature of the work.

Nimble views its target market as core research work that runs over longer periods rather than response speed. It said that in business-critical research, it is more important not to miss needed sources even if the answer is slower.

The "web search agent" is provided through the Nimble API, a software development kit and the Model Context Protocol. It also offers a free trial. Developers can connect it to existing agent tools or build applications for low-latency search, deep research and structured dataset generation.