Connect one repository and control access
Your practice folder is sitting on GitHub. You can now make it useful without writing any code by letting ChatGPT read the files, find information and answer questions with the repository as its source.
“Connect” sounds more powerful than this first step really is. The GitHub app inside ChatGPT provides read access. It can search content you authorize, but it cannot push edits, create commits or open pull requests. Making changes requires a coding workspace such as Codex, which is the subject of the next instalment.
Open ChatGPT, go to Settings, choose Apps and find GitHub. Select Connect, sign in to GitHub and inspect the authorization screen. When GitHub asks which repositories the app may access, select only the `ai-practice` repository from part one.
One repository gives you a clear boundary. You can see what information has entered the workflow and learn where to revoke access. More repositories can be added later through the installed app settings on GitHub. There is no reason to expose an entire account during a first experiment.
Availability depends on your ChatGPT plan and the mode you are using. OpenAI notes that some people may have GitHub in deep research or agent mode without seeing it in ordinary chat. If the option is missing, check the app directory and your plan before reconnecting anything.
ChatGPT can search file names, read their contents and cite material from the repository you approved. You might ask where a setting lives, what a folder is for or which documents mention a particular topic. The answer can still be wrong, so open the cited file whenever the detail matters.
A repository normally takes around five minutes to appear. If a new or private project remains unavailable, OpenAI suggests sending `repo:username/repository import` in ChatGPT to request indexing, then allowing another five to ten minutes.
Before connecting real work, check that passwords, API keys and personal information have not been committed by mistake. A private repository controls visibility on GitHub, but it does not make exposed credentials safe. Review ChatGPT's data controls too. OpenAI says content from individual accounts may be used to improve models when the relevant setting is enabled, while organization products do not train on that data by default.
Begin with three questions
Start with something you can verify against `README.md`: “Summarize the purpose of this repository in three points and cite the file you used.” Next ask, “Which files are present and what does each one appear to do?” A useful third prompt is, “Without changing anything, suggest one small improvement and name the file that would need editing.”
A dependable request includes the task, scope, limit and evidence. Try: “Inspect only this repository, do not invent missing files, and cite the exact path for every claim.” Boundaries do not make the AI less capable. They make its work easier to check. Our guides to how ChatGPT works and what AI agents actually do explain the machinery in more detail.
You can disconnect GitHub from Settings > Apps or revoke access through GitHub's installed app controls. Next time we will move from reading to proposed edits, using a separate branch and a visible review before anything is accepted.
The conversation starts here
Sign in with a supporter account to comment. Sign in
Nobody has commented yet. Want to go first?