The hardest web research tasks are not solved by simply running more searches. AI systems need to find the right sources, navigate to the information that may not appear in search results, retrieve precise data, and give users confidence that the final answer can be trusted.

Now available as a native plugin, Nimble brings specialized web search to ChatGPT and Codex, helping OpenAI users build deeper knowledge of the domains they care about.

With Nimble, web search can specialize around a specific task or domain to:

• Go deeper for more complete web context. Nimble learns how to retrieve the specific information your task requires and self-learns over time to capture rich insights. Instead of simply searching and retrieving pages, Nimble implements computer-use to intelligently navigate websites and uses entity retrieval to find the precise data you are looking for.
• Reduce the cost of web research. Nimble learns the retrieval paths that work, reducing the number of search and browsing tool calls needed to gather the right context.
• Trust your data with web search governance. Control which sources your AI uses, inspect how information was retrieved to make adjustments, and use confidence scoring to evaluate the data before acting on it.

How Nimble specializes web search for your domain

Nimble's Web Search Agents improve with every run, ultimately becoming "experts" at your specific research task. They create a search plan, adapt crawling and retrieval to the task, and persist useful sources, search history, and responses in memory.

Those learnings feed back into future searches, so the agent gets better at knowing where to look, how to navigate the web, and which retrieval paths produce the most relevant results. Over time, this specialization helps Web Search Agents go deeper into your domain to improve the completeness of your results, while reducing unnecessary search calls and token spend.

Three use cases for OpenAI + Nimble

Software Development: Codex Researches a Production Dependency Upgrade

Codex uses Nimble to research a major Next.js upgrade before touching the codebase. The workflow spans official migration guides, release notes, third-party package compatibility pages, and GitHub issues to identify breaking changes, deprecated APIs, and known integration problems.

Nimble gathers that context into a focused research layer for Codex, helping it make code changes based on current, cross-source evidence rather than isolated documentation pages.

Local Intelligence: ChatGPT Maps the Best Coffee Shops in San Francisco

ChatGPT uses Nimble's Local Places skill to discover and rank independent coffee shops near Bryant Park for a client meeting. Nimble searches across local listings, reviews, business websites, and social profiles to verify details like seating, Wi-Fi, ratings, and overall fit.

The result is a structured shortlist with confidence scores, supporting reviews, social presence, and an interactive map that ChatGPT can use to recommend the best options.

Competitive Intelligence: ChatGPT Work Builds a Competitive Landscape

A strategy team asks ChatGPT Work to research a competitor and build an actionable brief. Using Nimble, Work analyzes the company's website, pricing, product documentation, reviews, job listings, announcements, and other sources to understand its positioning, capabilities, and recent moves.

Work turns that research into a structured competitive brief, creates a dedicated Slack channel for the competitor, and posts the findings there for the team to review and discuss.

Benchmarking Deep Research Performance

To measure research quality, Nimble tested its Web Search Agents on the DRACO benchmark, which evaluates deep research systems across 100 cross-domain tasks for factual accuracy, breadth and depth, presentation, and citation quality. Nimble's highest-effort configuration scored 74.0% overall, the highest result among the systems tested; see the full methodology, results, and reproducibility in Nimble's public DRACO benchmark.

Better Web Context for ChatGPT and Codex

Nimble gives OpenAI users a way to move beyond generic web search and specialize web research around the tasks that matter to them.

By learning where to look, how to navigate the web, and what information to retrieve, Nimble helps ChatGPT and Codex gather more complete context with fewer tool calls and greater control over the data behind every answer.

Nimble is now available through the Plugins tab in ChatGPT and Codex.