> Honestly, check out some self-hosted stuff if youre worried about safety and code privacy.

Seconding the recommendation above. Before you dive in, basically you gotta understand that "privacy" isn't just a buzzword here... it's about IP protection. If you're cautious like me, I'd suggest Continue.dev for VS Code paired with a local Ollama instance running Llama 3 8B. It's a bit more DIY, but you're not sending code to some random cloud. Just make sure you have the VRAM for it tho lol. It's way safer than generic plugins for legacy repos tbh!

Honestly, for deep technical stuff like refactoring and unit tests, I've been super happy with Cody by Sourcegraph. It has a way better grasp of the whole codebase than Copilot imo. If you're on IntelliJ, it's basically indispensable now. For documentation, Mintlify is actually elite... it saves sooo much time with one-click docstrings that don't suck. Both integrate really well without feeling like you're constantly fighting the AI lol.

Sooo I was in the same boat literally a few months ago. Copilot is fine for basic boilerplate, but I kept getting annoyed with the context switching. I ended up switching my whole flow over to Cursor and honestly? It's been a game changer for refactoring. It basically feels like VS Code but with a brain that actually understands your whole project structure, not just the file you're looking at.

Here's what I've been using lately:
- Cursor: My main IDE now. The 'Composer' feature is actually insane for building entire components.
- Claude: I keep this open in a tab for debugging complex logic cuz its reasoning feels way more reliable than others.
- CodiumAI: This one is decent for generating unit tests without too much hand-holding.

I guess it depends on your needs, but moving away from just plugins to an AI-native editor made the biggest difference for me. It saves so much time on the 'boring' parts of dev work. gl!

I totally agree with the focus on local setups—the latency you get with cloud-based tools can reallyyy kill your flow when you're doing deep work. Tbh, I've been trying to find solid benchmarks on how these models actually perform under pressure, like when you're asking for a full refactor of a messy legacy class. I'm not 100% sure, but I think the 'SOTA' models on paper don't always hold up when you're looking at execution speed vs. accuracy in a real dev environment. IIRC, some of the smaller, optimized models are actually outperforming the giants in terms of tokens-per-second for simple unit tests, which is basically what you want if you don't want to wait 30 seconds for a suggestion, right? Someone told me that certain quantization methods are making local performance way better lately, but I'm still a bit uncertain if the tradeoff in logic is worth the speed boost for complex debugging. It's honestly such a moving target right now (at least that's how it feels on my rig) and the benchmarks change so fast it’s hard to keep up lol

Honestly, check out some self-hosted stuff if youre worried about safety and code privacy. Basically any open-source models running locally are amazing for complex logic without leaking your data. Seriously worth it!!

Similar situation here - I was getting worried about the cost of all these subscriptions tbh. I went through a phase of trying everything, but I've settled on Codeium Individual because it's actually free for individuals and doesn't feel like tech debt. Honestly, I'm pretty satisfied with how it handles refactoring without me having to constantly fix its mistakes. It works well in VS Code and saved me a ton of cash compared to the others!! peace

I almost bricked a legacy repo cuz of poor context. **Warning:** Watch out for token limits! Honestly, Tabnine is actually elite for deep logic and privacy specs.