Nimble has launched an OpenAI plugin that brings its web search tool to ChatGPT and Codex, giving users access to Nimble's search system inside both products.
The plugin is designed to help users handle more complex research tasks by finding relevant sources, going beyond standard search results and retrieving specific information from across the web.
Nimble positions the product as a specialist layer for AI-assisted research rather than a general search add-on. It argues that while AI models can reason effectively, the quality of their output still depends on the reliability and relevance of the information they can access.
According to the company, the system learns retrieval paths for particular tasks over time. Rather than restarting each research process from scratch, it refines how it searches for information in a given domain.
"AI systems are becoming remarkably capable at reasoning, but the quality of their answers is still constrained by the quality of the information they can access," said Uri Knorovich, Chief Executive Officer and Co-Founder of Nimble. "Nimble gives ChatGPT and Codex a specialized web research layer that can learn where to look, how to navigate, and which information matters for a particular task. The result is deeper web context, fewer unnecessary searches, and greater confidence in the information AI systems use to make decisions."
How it works
Nimble said its search approach focuses on specialization around a task or domain. Instead of only returning pages from a search query, the system can navigate websites and use entity retrieval methods to find specific data points that may not appear directly in search listings.
The tool is also intended to reduce research overhead by learning which retrieval routes are most useful for a certain type of query. In practice, that is meant to cut the number of search and browsing calls needed to gather context while limiting the amount of irrelevant material an AI system must process.
Another element is source control. Users can choose which sources are used in their workflows, inspect how information was retrieved, adjust retrieval strategies and apply confidence scoring before using the results.
Use cases
Nimble outlined several examples for ChatGPT and Codex users. In software development, Codex can use the tool to review migration guides, release notes, compatibility pages and GitHub issues before a major dependency upgrade is applied to a codebase.
For local business research, ChatGPT can use the system to assess companies across listings, reviews, websites and social profiles, then produce a shortlist with supporting evidence and confidence scores.
Nimble also pointed to competitive intelligence work. In that scenario, ChatGPT Work can research a rival across its website, pricing, product documentation, reviews, job listings and announcements, then turn the material into a structured brief for internal teams.
Those examples reflect a broader push in the AI market to move beyond simple retrieval and towards more involved research workflows. As businesses test AI tools for analysis and operational tasks, vendors are trying to show that their systems can trace sources, gather more complete evidence and give users better visibility into how answers were produced.
Benchmark results
Nimble also published performance claims tied to the DRACO benchmark, which measures deep research systems across 100 cross-domain tasks. The benchmark assesses factual accuracy, breadth and depth, presentation and citation quality.
The company said its Web Search Agents achieved a 74.0% overall score, which it described as the highest result among the systems tested. Nimble added that it has published methodology, results and reproducibility materials as part of an effort to make web research performance more transparent.
The plugin's release comes as AI developers and corporate users place greater emphasis on the quality of external data used by large language models. Better reasoning from models has increased demand for systems that can supply more dependable and better targeted information from the web, particularly for work that requires evidence across multiple sources.
Nimble said the plugin is available for ChatGPT and Codex users, while its search technology also remains accessible through API, SDK and MCP interfaces.