Projects with multiple repositories: Giving AI agents the full context

Highlights and Transcript

Why does multi-repository context matter:

I wanted to show how I code with AI agents on projects with multiple repositories. This is something that is pretty common, right? You might have separate Git repositories for the frontend, backend, or the various microservices.

The useful bit here is being able to give the agent context across all of them, instead of treating each repository as a separate project.

A rate-limiting example:

In this case, I wanted to make a rate-limiting-related change in my project. I was not sure which limits existed on the backend, frontend, or both. So I asked Codex to look through all the limits in the project.

It shared the backend-related rate limits and, after a little more digging, the cookie-related frontend limits as well. Based on that, I had a complete picture fairly quickly. I could decide what I wanted to change, and then ask the agent to make that change on the backend or frontend wherever it was needed.

The important thing is that the agent can search for context across multiple Git repositories. I can also point it to files that belong to different repositories. Better context generally gives better outcomes from the AI agent.

Setting it up in Codex and Cursor:

The setup is quite simple. When creating a Codex project, select the folders that represent the different repositories. That is about it. You can then refer to files across those repositories.

Cursor is pretty much the same. A Cursor Workspace lets you add different folders to one workspace. I am not sure how this currently works in Claude Code. About a month ago, it needed a top-level folder containing all the repositories.

Why access to the complete stack helps:

The key is to have a single AI-agent project that reflects all of the repositories involved. It is a small change, but if you work on a project with multiple repos and are not doing this already, I would strongly suggest giving it a try.

I have been using this for months, and I do not think I will want to go back to separate AI projects for separate Git repositories. Even if you are a frontend or backend developer responsible for only one part of the stack, it still helps to have the other repositories available. The agent can suggest things better, and you get a better understanding of the complete picture as well.

The token-consumption trade-off:

The only downside I have experienced is higher token consumption. That is expected: there are more files for the agent to go through and submit to the LLMs.

But given the quality gain I have seen, and the ongoing reduction in token costs, I think it is worth it.