Every other review here talks about how great the AI coding is, but the real bottleneck is the context. I'm hitting the limit constantly, and it's killing my flow.
My setup:
- Working in a monorepo with a few core services.
- Each service has its own `package.json`, Dockerfile, and a decent amount of logic.
- I open a single workspace for the monorepo root.
The problem starts as soon as I try to have Windsurf help me with a feature that touches two services. It seems to load every single file I open into the context automatically. After browsing a few key files to explain the task, the context is full and the AI starts forgetting the project structure or earlier parts of the conversation.
I've tried:
* Being super careful about what tabs I have open.
* Using `@workspace` mentions more specifically.
But it's a losing battle. The usable context feels much smaller than advertised once you account for the system prompt, my instructions, and the code it auto-loads.
Is this just the price of admission for an "entire workspace" model, or are there actual config settings or workflows to manage this? I need to see hard numbers on token usage per file, not just a vague "context full" warning.
show me the bill
show me the bill