Anaconda Makes Kilo Desktop AI Assistant Generally Available

Anaconda has made Kilo Desktop generally available, bringing an AI agent into live Jupyter notebooks to edit and run cells, inspect data, and manage environments and packages. The application also offers interactive charts, dashboards, data tables and access to more than 500 models.
- Kilo Desktop is now generally available.
- The agent can edit and run cells in live Jupyter notebooks.
- Users can manage conda environments and governed package installations in the app.
- The app offers more than 500 models, including local and self-hosted options.
- Kilo Enterprise can block specific models and providers.
An AI agent inside the notebook
Kilo Desktop lets users open an existing .ipynb file or ask the agent to create one, with the live notebook displayed beside the chat. The user and agent work in the same document, allowing the agent to write cells, run them, read outputs and use plots as part of its analysis.
According to Anaconda, the agent works in a real kernel and can read tracebacks, diagnose kernels that fail to start, and restart stuck kernels. Users can ask it to profile a dataset, test feature-engineering ideas, or turn an exploration into a clean notebook while keeping the data loaded for review.
Code execution requires approval in Manual mode. Running a cell and restarting a kernel each display an approval card, while package installation has a separate card listing proposed additions and flagging names that resemble misspellings, such as “numppy” for numpy and “requets” for requests.
Environments, governed packages and visual outputs
Users and the agent can create and manage conda environments within Kilo Desktop. The app supports starting with a Python version, importing an environment.yml file, or cloning an environment before making changes. Anaconda says users can review dependency changes, while the “Back up and update” option clones an environment first so an upgrade can be undone.
When signed in, the agent installs from Anaconda’s catalog of more than 19,000 vetted Python packages and from governed channels configured by an organization. Its install tools reject packages from local files and URL channels, while shell commands are permission-gated. New notebooks receive an environment with ipykernel automatically.
CSV, TSV and Parquet files open as tables in the Files pane. The agent can produce interactive Plotly, Bokeh and HoloViz charts, as well as data tables and inline diagrams, which users can save to their workspace or open in a full-size viewer.
Model choice and organization controls
Kilo Desktop provides a single picker for more than 500 models, including models from OpenAI and Anthropic, NVIDIA Nemotron and models running on the user’s own hardware. Users can add an API key, an Amazon Bedrock account or a self-hosted endpoint as a provider. Auto Efficient can select the least expensive model shown by Kilo Bench to be capable of a task.
For sensitive data, the built-in Local Model Server runs open models on local hardware without sending those model requests to a hosted model provider. Users can browse the Anaconda Local Model Catalog or import GGUF models downloaded with Anaconda Desktop.
The release also includes multi-agent sessions, the Workspace Board and the Anaconda Local Model Catalog. Kilo Enterprise lets organization owners block specific models and providers, combining those controls with Anaconda’s governed package channels. Anaconda says more than 5 million developers already build with Kilo, and positions Kilo Desktop as part of its AI Workspaces.


