Building with AI usually means working across several tools. Frontier labs like OpenAI, Anthropic, and Google build remarkable models, and teams want them for their hardest problems. The catch is that each lab’s agent harness is locked to its own model family, so you cannot switch to the open-source models coming out of NVIDIA, Moonshot, MiniMax, Z AI, and others. Some teams also want to build on local models, and they want more from a building platform than generic code generation. Your notebooks, conda environments, and terminals live somewhere else, and your security team has to approve each of those tools on its own.
Kilo Desktop beta puts that work in one app. You get native access to more than 500 AI models, agentic coding and data science workflows that plan before they build, and built-in management of local models and conda environments, all inside a tool your security team can approve.
Kilo Desktop comes from the team behind Kilo, which more than 5 million developers already use.
One app. Every use case. #
In Kilo Desktop, you can build a service, analyze a dataset, debug a failing pipeline, or ask how an unfamiliar codebase works without switching tools.
Workspaces
Real projects rarely live in one folder. An API and the analysis that justifies it usually sit in different repositories, and most coding agents can only see one of them. A Kilo Desktop workspace is a named group of folders from your machine, so the agent can read the analysis while it writes the service, and nothing has to be copied between them.
Agent modes
Kilo Desktop ships with four agents. Code writes and edits files. Plan reads your project and returns a step-by-step implementation plan without touching a line. When you approve a plan, Code takes over, splits the plan into subtasks, and runs them in parallel with sub-agents.
Ask answers questions about your code without changing anything, and Debug finds what broke and fixes it. You can also build custom agents for your own workflows and share them with your team.
One app. Your entire workflow. #
Notebooks that agents can edit
In most tools, AI help with a notebook means pasting cells into a chat and pasting answers back. In Kilo Desktop, the agent edits the notebook directly, and you watch its cursor move through the cell as it writes. You review the work as it happens, in the notebook, with the data loaded.
Conda environments without a terminal
An agent is only as useful as the environment it runs in. Kilo Desktop creates and manages conda environments inside the app and draws packages from Anaconda’s catalog of more than 19,000 vetted packages. You get the right dependencies before the first prompt, and every one of them has passed Anaconda’s review.
A layout built for review
Chats, notebooks, files, diffs, and terminals all open as tabs, and you arrange them into groups however the work demands. If you keep the diff view open while the agent writes, you see changes as they land, and run the result in a terminal beside the chat without leaving the app. For wide layouts, carousel mode lets you set each panel’s width and scroll across them. Any tab can go full screen or pop out onto a second monitor, and it snaps back into place when you are done.
One app. Massive power. #
500+ models, and a router that picks for you
Every major frontier lab, the leading open-source models, and providers from every geography are available in one picker. If you would rather not choose, Kilo’s auto routers choose for you. Auto Efficient checks your request against Kilo Bench, our own coding benchmark, and sends it to the cheapest model that has proven capable of that kind of task. You pay for what the job needs, and your finance team gets a cost story that holds up.
Bring your own providers
Models you host yourself or subscribe to elsewhere plug in through a preset or a custom endpoint with your own API key.
Local models
Some work should never leave the machine. Kilo Desktop includes a local model server that runs open models directly on your hardware, where your prompts stay with them. One setting controls whether local requests carry Kilo’s full agent toolkit, which makes them capable of working on code, or go without it, which makes them fast and light.
Skills, MCP servers, and custom agents
The Kilo marketplace carries skills, Model Context Protocol (MCP) servers, and custom agents, and you can install any of them for one workspace or for all of them. An agent you shape and approve for one project follows you to the next.
Safe doesn’t mean slow #
Breadth without governance is a liability, and until now enterprises had to choose. Kilo Desktop lets administrators decide which models people can use, which providers and geographies are allowed, what each agent is permitted to do, and which skills it can install. The controls live in the product, so the policy your security team writes is the policy the app enforces. You get the whole landscape, and you draw the boundaries.
Kilo Desktop is available to download today. Try it, and let us know what you think.