Dhruv Bhutani has been writing about consumer technology since 2008, offering deep insights into the personal technology landscape through features and opinion pieces. He writes for XDA-Developers, where he focuses on topics like productivity, networking, self-hosting, and more. Over the years, his work has also appeared in leading publications such as Android Police, Android Authority, CNET, PCMag, and more. Outside of his professional work, Dhruv is an avid fan of horror media spanning films and literature, enjoys fitness activities, collects vinyl records, and plays the guitar.
Say what you will about AI, as a tech enthusiast, you're bound to be fascinated by the pace of development. I try to keep up with things as much as possible, and lately I've been testing out a range of AI coding tools to help me brush up on my rusty coding skills. Most of them promise a fair amount of flexibility in how much of an assist they'll offer, but the results can be extremely hit or miss. The focus tends to be more on a great interface rather than solid completions, and not many are focused on giving you access to a locally run model to save on cost. More often than not, you are locked to a few LLM providers, and the workflow basically assumes that you're going to use a hosted stack.
That problem is what pushed me towards Kilo. Sure, I wanted an AI assistant that lives inside VS Code. Something that could understand my project and help me complete it when I was stuck. However, I also wanted the flexibility of being able to switch models on the fly. Kilo offers that and more. One, it runs directly within the editor. But, more importantly, it gives you the flexibility of using multiple cloud-based operators as well as local models. While it's too early to say that it's the definitive AI code extension to use, if local AI is of importance, it's certainly worth looking into.
I finally found a local LLM I actually want to use for coding
Qwen3-Coder-Next is a great model, and it's even better with Claude Code as a harness.
Posts 25
By Adam Conway
Designed for how developers actually work
Switch between local and cloud models without breaking stride
The first thing that strikes you when you install Kilo is that it fits like a glove within the VS Code interface. It's not a floating window or panel. Instead, it can interact directly with your workspace, files, and the structure of the project that you're currently working on. That might not be a huge criterion for everyone, but for me, it goes a long way towards integrating it into my workflow.
Elsewhere, the biggest friction I've felt with most AI coding tools is the model lock in. I like experimenting. And while the latest QWEN model might not be able to compete with Claude, it can still be extremely useful when I want privacy, or access with limited or no data connectivity. Just as importantly, for lightweight tasks, I often want to run a lightweight local model that can keep costs down.
Kilo lets you maintain active configuration profiles with API keys and more saved. You can plug in different providers, bring your own keys, or point it to locally running models. You can easily switch these out with a drop-down menu. That separation and ease of switching matters more than you'd imagine. It means that you're not getting bogged down with rebuilding your workflow every time you want to test out a new model.
What also differentiates Kilo is how it differentiates different kinds of work through modes. Instead of one generic assistant trying to perform all kinds of tasks, you can shift between architect mode, coding mode, debugging mode, and, my favorite, an ask mode that lets you query your own code. This is especially useful when you're trying to understand how something works. In practice, this maps very closely with how I move through a new project that I'm working on.
When I'm in the early stages of planning out a feature, Architect mode keeps responses structured and at a high level, since I haven't committed to an approach yet. Switching over to Code mode, the assistant gets right into action and starts generating or modifying files directly. If you don't want to use AI-generated code in your project, you can also use the debug mode, where it specifically helps you identify errors or optimize your code. It's invaluable in keeping the entire cycle of working with code right within the VS Code interface, and reduces workflow friction by leaps and bounds. I now spend less time searching for fixes and more time just moving my project forward.
Does Kilo actually save time
Speed enhancements and a whole lot of flexibility
So, you might be wondering if there's any real benefit to installing Kilo over using something like Copilot. Does Kilo actually save you time during everyday work, or is it just another AI layer that seems impressive but doesn't add much value to your work day.
The answer is -- it depends. So far, in my use, paired up with a local LLM, I've found great utility in writing things like boilerplate code. Local models aren't quite powerful enough to manage writing long code sections just yet, especially not on a laptop-first workflow. More than that, though, I've found a lot of utilities in debug mode which walks you through failure points in your code and suggests fixes. It's not perfect, no AI tool really is, and it's best if you verify all suggestions, but it goes a long way towards expediting the process.
Subscribe to our newsletter for AI coding-tool insights
Get deeper value by subscribing to the newsletter for in-depth coverage, hands-on guides, model comparisons and practical tips about AI coding tools like Kilo -- how to run local models, switch providers, and use architect/code/debug modes for real developer workflows.
Subscribe
By subscribing, you agree to receive newsletter and marketing emails, and accept our Terms of Use and Privacy Policy. You can unsubscribe anytime.
It also helps that, paired with a local LLM, you just end up saving a lot of money if your AI use case revolves more around smaller code snippets rather than full-scale projects. With Kilo, I can easily alternate between the two modes. A smaller local model for simpler tasks, and a stronger cloud model when I need deeper reasoning. That flexibility goes far when using the tool across varying workloads.
Built for real-world workflows
Kilo isn't positioning itself as the best AI code extension. Instead, it comes across as one of the most versatile ways to balance local and cloud coding workflows. Moreover, it focuses on practical workflows that power users run into, and offers tools specifically optimized for them. Be it architecture mode, coding mode, or debug mode, it is able to give you an experience optimized for specific use cases over all-around code assistance. All of that combined make Kilo an excellent add-on for any developer working in VS Code.
Kilo
Kilo AI is an open source VS Code coding assistant that lets you generate, debug, and automate code using either local or cloud LLMs.
See at Official Website
Expand Collapse
