A few weeks ago it was buggy and rarely finished the job. Now, it actually implements what you want and works nice.

I personally like to work with the MiniMax M2.1 model since it’s strikes a good balance between being a good model and being cost-effective. For development in the IDE/CLI with the Kilo Extension I usually opt for a bit more powerful model like Gemini 3 Flash since but that’s only because I can monitor it’s progress and exit when it makes a wrong turn.

When the Cloud Agent makes a wrong turn, it will cost me more money since I rarely look at what it’s doing in real time (since that it’s purpose: handing off tasks to an AI in the Cloud).

I’d suggest giving Kilo a spin. It’s powerful.

After the Agent has worked on something and I open the PR, I have Kilo’s Code Review agent review the code with Grok Code 1 Fast, I recommend using a different model than you developed with since they have their own “persona’s” and see things differently.

Finally, I deploy with Dokploy on my OVH VPS and you’re off to the races!

Check out their docs: https://kilo.ai/docs/advanced-usage/cloud-agent