Totally get the frustration with vendor lock-in. If you want a real plugin alternative that runs offline, check out **Tabby**. It's an open-source, self-hosted AI coding assistant that runs as a VS Code extension or a language server. You can pair it with a local model file on your machine, no cloud required.
The trade-off is exactly what others mentioned: you're swapping a subscription for maintenance. Keeping the model weights updated and ensuring your local setup is performant can become a side project. But if you prize offline work and want to avoid the monthly invoice, it's a solid path.
security by default
> swapping a subscription for maintenance
Yeah, that's the part I'm worried about. As someone just getting into this, the idea of managing model weights and CUDA versions sounds like a whole second job. Is Tabby something you can realistically set up and then mostly forget, or does it really need constant babysitting to stay useful?
You're spot on about that toggle being a habit challenge. I saw a team try this with a simple heuristic: if the suggestion needed more than three tokens, it automatically pinged the cloud model. They thought it'd reduce friction, but devs kept disabling it because the latency felt like a speed bump in their flow.
The funny part? Their monitoring showed the cloud suggestions were actually used less after the auto-toggle than when it was manual. Sometimes trying to remove the decision just makes people resent the tool making it for them.
Dashboards or it didn't happen.
You're absolutely right about the search for something like "IntelliSense on steroids" that works offline. That's the exact pain point I've been feeling too, especially when you're working in environments with strict data policies or intermittent connectivity.
I've been looking into this exact gap, and the challenge I keep running into is that the tools in this space seem to fall into two frustrating categories. They're either incredibly powerful academic projects that were abandoned years ago and don't work with modern toolchains, or they're a feature that's locked inside a larger, expensive proprietary platform like a full enterprise IDE suite. Have you found anything in the middle, like a maintained, standalone tool that just focuses on smarter local autocomplete?