I keep seeing this "whole-repo context" phrase thrown around as Aider's killer feature, especially against Cursor. But for someone with a small project—let's say under 20 files—what does this actually *do* for me that's different?
My understanding is that it means Aider can see every file in your project at once when you ask a question or give an instruction. Not just the open tabs. The claim is it makes the AI's changes more consistent.
But I'm skeptical. My questions are:
* If I have a 10-file Flask app and ask "add a new form for contact requests," is it actually scanning all 10 files to understand the structure before writing the new `contact.py` and updating `routes.py`? Or is it just a marketing term for a slightly larger context window?
* What's the practical impact? Does it just reduce the number of "please open `models.py`" back-and-forths?
* Are there clear downsides? I assume dumping the whole repo into the context burns tokens faster. For a small project, is that cost/benefit even noticeable?
I want a concrete, step-by-step explanation of the difference in behavior between having this on vs. off for a trivial project. Not a vague "it understands your codebase better."
-- CRM Surfer
Your CRM is lying to you.
Good questions. It's not just a bigger window - it's a fundamentally different prompting strategy. With a typical chat interface, you're responsible for curating context. Aider automates that.
For your Flask example, yes, it actually does scan those 10 files first. It builds a map before writing a single line. The practical impact is fewer hallucinated imports and less "please open X" chatter, because the AI already has the project structure in mind.
The downside is exactly what you guessed: token burn. For 20 files it's probably negligible. But if your "small" project includes a giant `node_modules` or a 10k-line `package-lock.json` you'll feel it. You need a good `.aiderignore` file, which is just another thing to manage.
The real test is asking "refactor the database layer." Without whole-repo, it's guessing. With it, the AI can actually trace calls across files. But for simple "add a form" tasks? You might not see the magic.