Anaconda's Kilo Desktop Targets Local AI Work on NVIDIA RTX Spark PCs

Hardware built for local models

NVIDIA RTX Spark brings local AI computing to Windows laptops, including the new Microsoft Surface Laptop, according to Anaconda. The platform combines NVIDIA RTX graphics, the full NVIDIA AI stack and up to 128GB of unified memory, aimed at developing, creating and gaming on a single device.

Anaconda says RTX Spark pairs NVIDIA Grace CPU cores with a Blackwell RTX GPU, and that some configurations include up to 128 GB of unified memory. Because the CPU and GPU share that memory, there is room for larger models and more of the context they need while running. The platform also runs the full NVIDIA CUDA platform, the same software stack used across NVIDIA hardware, which Anaconda supports with Python artifacts.

On the software side, Anaconda's Windows on Arm support provides native Python, 2500+ conda packages and vetted libraries.

What Kilo Desktop does

Kilo Desktop is described as one app for building with AI, where users can plan, write, test and debug code. Its built-in local model server runs open models on RTX hardware, so prompts can stay on the user's machine. With local inference selected, prompts are processed on the device.

The app also sets up conda environments inside the application, so an RTX Spark device is ready for Python work from the first prompt, according to Anaconda. Users can create or import environments, including from an environment.yml file, and keep code, notebooks and data together with the model chat.

Anaconda suggests starting with a real task, such as exploring a dataset, fixing a failing script or building an assistant that searches project files.

Nemotron models and the setup process

The post highlights the NVIDIA Nemotron family, which it describes as open models built for reasoning and tool use. These matter when an agent has to work through several steps: a suitable Nemotron model can interpret a request, decide which tool to call, and use the result to choose its next action.

Users should pick a model version that fits the machine's memory and import GGUF model files directly into Kilo Desktop's built-in server, which adds them to the agent's model picker. For agent work, Anaconda advises choosing a model that supports tool calling, then turning on coding tools and instructions in its settings.

The listed steps are:

• Open Kilo Desktop, sign in and select the project folder.
• Select the project's conda environment or import its environment.yml file.
• Import a compatible Nemotron model, start the model server and select the model in chat.
• Turn on coding tools for a model that supports them and give it a small task, such as explaining a Python file.

Anaconda says the goal is a first local response in under five minutes once Kilo Desktop, the model file and the project's packages are installed. Initial downloads and setup take longer, depending on the connection and the chosen model.

An example workflow and what it means

Anaconda offers a hypothetical case: a product team wants to know why returns went up and starts with a folder of CSV files. A developer could ask the agent to inspect the columns, flag missing values and write a Python notebook comparing return rates by product and month. The developer reviews the code and results, then asks the agent to look into the largest changes. Python handles the calculations, while the local model directs the work and explains what the numbers show.

The company says the same approach works for a document assistant, a coding agent that runs a project's tests, or a workflow that turns new local files into a report. For teams, Anaconda recommends saving Python requirements in an environment.yml file and recording the model version and settings next to the code, so teammates with compatible hardware can recreate the workflow.

For developers with RTX Spark machines, the pitch is a way to run the whole development process locally on Windows, with prompts staying on the device. The post does not give pricing or performance benchmarks.