Using ai-devkit in multirepository projects

[lusorio]

Hi, not an issue but rather a question for you guys.

When working in a multirepository project, what's the best approach for using ai-devkit? Would it be better to init each project independently, so each would hold its own /docs/ai structure, or would it be best to have a single /docs/ai folder in a dedicated repository?

Currently I'm using cursor and have all projects in a workspace. I'm testing out devkit for the first time and I've only init it in one of the projects for now, and I see when using commands e.g /new_requirement, the ide transforms it to /login/new_requirement which makes me wonder if having ai-devkit in each would be a simpler approach, or maybe simply referencing the project with each command e.g "/new_requirement add a new item in **auth frontend**"...

Excuse me if this question is totally nonsense, I'm still processing a lot of info here 😅

fwiw, hats off for this project, I'm really enjoying while learning it. Thanks for your nice work

[codeaholicguy]

Hi @lusorio, thanks for asking, and glad to hear you are trying it out.

The way I think about ai-devkit is that docs/ai is the context layer for the agent. It contains things like requirements, implementation plans, progress notes, testing notes, design decisions, and other working context. Because of that, it is usually quite project-specific.

So my default recommendation is: keep docs/ai inside the same project codebase.

When the agent is working on that codebase, it can retrieve the most relevant context more efficiently, without mixing too much unrelated information from other repositories.

That said, there is another approach that can work well if your setup is closer to a mono-repo or mono-workspace. For example, if you check out all related codebases under the same root folder, you can initialize ai-devkit at that root. In that setup, docs/ai becomes the shared context for the whole system rather than one specific codebase. This can be useful when the work naturally cuts across frontend, backend, auth, API, infra, etc.

So I would frame it like this:

• Option 1: Init ai-devkit per repository. Best when each repo has its own lifecycle, context, requirements, and implementation plans.
• Option 2: Init ai-devkit at the workspace root. Best when the repositories are tightly connected and you often work across them as one system.

For your Cursor workspace case, I would probably start with the simplest setup first. If each project is relatively independent, init each one separately. If you often need the agent to reason across multiple repos together, then try the workspace-root approach.

I also want to keep ai-devkit flexible by design. The goal is not to force one fixed workflow, but to provide the foundation pieces, docs, skills, prompts, memory, rules, and let you plug them together in the way that fits how you work.

Your question is not nonsense at all. It is actually the right question to ask when applying this to real multi-repo work.

Looking forward to hearing your success story. And if you have any other questions, please feel free to ask.