Ok so, before jumping in, you gotta realize that getting that 'Copilot-like' feel depends entirely on how the extension handles Fill-In-The-Middle (FIM) logic. FIM basically lets the AI see the code *after* your cursor so it doesn't just hallucinate things you've already written, right? I've tried a bunch of setups and honestly, I was kinda disappointed with some paid options that felt super bloated and laggy. For your situation, here's what I recommend: • Continue for VS Code: This is my daily driver. It's the most stable for real-time completions. You just plug in your DeepSeek API key and it handles the FIM prompts natively. Lowkey the smoothest experience if you want low latency and cheap operational costs.
• Roo Code: Formerly Roo Cline. It's amazing for 'agentic' tasks, but I had issues with it being a bit of a token hog. It's not as good as expected for simple ghost-text completions cuz it's built more for complex refactoring. Basically, if you want that native feel without breaking the bank, stick with Continue. It's basically the industry standard for open-weights integration right now. gl!

> specifically something that handles both chat and real-time inline code completions well. Just sharing my experience: I went through this last year when I first tried getting DeepSeek-Coder-V2 to work. Honestly, it took me forever to figure out which setup felt "right." I ended up comparing Continue VS Code Extension and Codeium AI. Basically, Continue is great cuz you have total control over the API (pro: customizable FIM logic, con: harder for beginners to configure). Codeium is super fast and polished (pro: low latency, con: limited custom API support). Tbh, I'm still a bit of a noob with all these keys and stuff, but I found that Continue gave me those "native" feeling suggestions once I finally tweaked the config file. Tip: Stick with Continue if you realy want to use the DeepSeek API. It’s a bit of a setup, but the logic is way better for complex code things!! gl!

> Which one is currently the best for DeepSeek integration? Tbh, while the others mentioned are great for lightweight setups, if you look at the broader market research right now, Sourcegraph Cody is really leading the pack for context management. It handles DeepSeek-Coder-V2 exceptionally well because it doesn't just do FIM—it uses a more sophisticated RAG (Retrieval-Augmented Generation) pipeline that pulls in your local codebase context way better than basic extensions. If you're looking for stability and a "native-feeling" integration, Cody's custom LLM support is top-tier right now and super reliable. Another one that’s blowing up in the developer community is Void. It’s basically an open-source alternative to Cursor that allows for deep DeepSeek integration without the subscription lock-in. Honestly, the market is shifting toward these more "context-aware" tools rather than just simple autocomplete plugins. It’s sooooo much more efficient when the AI actually understands your entire project architecture instead of just the current file—it's basically the industry standard now. Definitely worth a look if you want that high-end feel!

yo, i stumbled upon this and honestly, i feel u. getting that perfect 'native' feel while staying budget-conscious is the dream. i've been using DeepSeek-Coder-V2 for a while now and honestly, the most stable setup for me has been using the DeepSeek API through Double.bot. it feels way snappier than the others and the FIM (fill-in-the-middle) logic is actually legit. if youre trying to save money, here is what i recommend:
- Stick with the DeepSeek API directly cuz it's dirt cheap and you only pay for what you use.
- Check out Double.bot for the VS Code integration. It handles the chat and inline stuff seamlessly without the lag you see in some other wrappers.
- For heavy lifting, Aider AI is also a beast, tho it’s more of a CLI tool it works great alongside VS Code for big refactors. im super satisfied with this setup... feels very pro but costs pennies compared to Copilot. works well tho! peace.

Would love to know this too

Interested in this too

Regarding what #5 said about "Would love to know this too"

• I have been trying to get a stable DeepSeek setup for weeks and unfortunately it has been pretty disappointing. Most extensions have too much overhead or weird telemetry that I dont trust. For a more reliable DIY approach, I suggest looking into:
• Aider AI Coding Assistant which works via terminal but is way more robust for logic.
• OpenRouter API to manage your keys and fallback models if the primary endpoint lags.
• Tabby ML Self-Hosted for a local completion engine that actually respects your privacy. I actually had to pause my testing because my workstation started acting up yesterday. I spent four hours troubleshooting a persistent buzz coming from my Noctua NF-A12x25 PWM fan. Turns out a tiny bit of dust had gunked up the bearing, which is frustrating for such a high-end part... anyway, I'd stick to the more minimal tools if you want actual stability with DeepSeek.