OpenAI’s GitHub Integration Brings ‘Deep Research for Your Code Base’

OpenAI on Thursday announced a GitHub integration for the Deep Research tool in ChatGPT, enabling users to link repositories to ChatGPT. Based on the queries, the deep research agent will retrieve pertinent information from the repository to compile reports. 

For context, Deep Research on ChatGPT is a feature that performs comprehensive research across multiple sources to craft lengthy reports. The feature was announced last year and is available in various capacities across all ChatGPT plans. 

You can now connect GitHub repos to deep research in ChatGPT. 🐙

Ask a question and the deep research agent will read and search the repo’s source code and PRs, returning a detailed report with citations. Hit deep research → GitHub to get started. pic.twitter.com/cdrQChfAU3

— OpenAI Developers (@OpenAIDevs) May 8, 2025

OpenAI also shared an example to illustrate the feature’s capabilities. The report involved browsing through OpenAI’s Codex CLI repository on GitHub. In a post on X, Greg Brockman, president and co-founder of OpenAI, called it “Deep Research for your codebase”.

GitHub is a platform that lets developers host code repositories for software programs. It was acquired by Microsoft in 2018 for $7.5 billion. Microsoft has also actively backed OpenAI, with multiple investments. However, several reports have hinted that the relationship may have turned sour. 

Besides, OpenAI releasing the feature signifies another attempt at solidifying its value proposition for developers. Recently, it was reported that OpenAI reached an agreement to buy Windsurf for $3 billion. 

Last month, the company released a new open source code generation tool called Codex CLI—a lightweight coding agent that brings multimodal reasoning to the command line. Users can combine screenshots or sketches with local code context, and the tool is now open source.

OpenAI also mentioned that newly released o3 and o4-mini models achieve higher scores in benchmarks such as Codeforces competition code than their predecessors, citing a 20% fewer errors over OpenAI o1, as tested by external experts.

“We’ve trained them to use tools, which is not something that we had done with our previous reasoning models. They actually use these tools in their chain of thought as they’re trying to solve a hard problem,” said Brockman, adding that in one instance, o3 used 600 tool calls in a row trying to solve a really hard task.