Okay, so I was setting up Continue in VS Code and hit this choice: cloud or local? The docs were a bit dense. 😅
Here's my quick take after trying both:
**Cloud Mode**
* Your code/context gets sent to Continue's servers.
* Uses their cloud LLMs (like Claude 3.5 Sonnet, GPT-4).
* Pro: Usually faster, less setup, and you get the fancier models.
* Con: Your code leaves your machine. Might be a dealbreaker for work stuff.
**Local Mode**
* Everything runs on your machine.
* Uses local LLMs (like Ollama with Codellama, DeepSeek Coder).
* Pro: Your code never leaves your laptop. Total privacy.
* Con: Needs more RAM/GPU, and the models aren't quite as sharp as cloud ones.
Basically: **Cloud = power & convenience, Local = privacy & control.**
For my side projects, I use cloud. At my day job (CRM stuff), we're required to use local. Makes sense!
Anyone else have a strong preference? What's your setup?
~E
Trial first, ask later.
I think you've nailed the core trade-off. Your point about cloud mode being "usually faster" is something I'd qualify, though. That speed depends entirely on your network latency and the current load on Continue's API endpoints. I've had moments where local Ollama with a quantized Codellama responded in under a second while the cloud call spun for three. For isolated, single-file tasks, a well-tuned local model can feel snappier.
The real gap, in my experience, isn't just raw speed but reasoning depth on complex, multi-file refactors. That's where the cloud models still pull ahead significantly. Have you tried benchmarking a specific refactoring task, like extracting a common module, in both modes to compare the quality of the diff?