Continue.dev agentic alternative: roo code

If anyone else has had problems getting continue.dev to work in agentic mode, I found roo code and llama.cpp to work better (with qwen 3 coder 30b) GitHub - RooCodeInc/Roo-Code: Roo Code gives you a whole dev team of AI agents in your code editor.

Thank you for the info, hopefully this will help others in the future.

There is also Cline which Roo is a fork of. Roo has more features, but Cline is a bit more robust.

Also, the new GitHub Copilot in VSCode Insiders Edition can work with any OpenAI-compatible endpoint (vllm, llama.cpp), and I prefer it now over the others.

Yep after using Roo for a while it also had some challenges so I reverted to Aider for code changes and UI-based chat for architectural chats. I’ll give VSCode GitHub CoPilot a shot thanks for the tip.

I haven’t used Aider for a while. Did they finally switch to native tool calling? It used to be my #1 coding tool, especially for smaller changes, but the tool calling wasn’t very reliable and usage of backticks for coding blocks backfired a lot.

I recently discovered KiloCode and found it to be quite good - in fact, better than Roo Code in my experience.

Conceptually, they are almost the same thing, but KiloCode feels more polished in the areas where Roo Code is still rough. Most importantly, it behaves more stably overall.

This comparison is specifically about agentic workflows. In that context, KiloCode seems more reliable and better finished, whereas Roo Code can feel inconsistent at times.

If anyone else has struggled with getting Continue.dev to work properly in agentic mode, Roo Code was already an improvement for me (especially when paired with llama.cpp and Qwen 3 Coder 30B), but KiloCode takes it a step further by addressing some of Roo Code’s remaining shortcomings.

Hopefully this information helps others who are experimenting with agent-based coding setups.

KiloCode is one of my favorites, but I’ve found another up-and-coming CLI agent that uses the least amount of token context to work. It’s called Code Puppy, it’s made by a Walmart developer, Mike Pfaffenberger.

GitHub - mpfaffenberger/code_puppy: Agentic AI for writing code

Have you used Code Puppy?

Yes, I use it often for development but I use the cloud-based coding models. I’ve been trying to wean off of them. Qwen3 Coder 30b runs pretty well on the DGX Spark!