A data workspace that runs on your machine
Today we are releasing Kontext Studio, an open-source, local-first AI data workspace. Ask a question in plain language and a team of agents works through your files and databases, then returns SQL results, charts, and reproducible notebooks. Bring your own model, local or cloud, and keep the workspace on the machine where your data already lives.
The source is available at kontext-tech/kontext-studio under the MIT License.
Why we built it
Most analytics tools ask you to move data into their cloud before you can ask a question about it. That trade-off is hard to justify for a laptop full of Parquet files, a local DuckDB database, or a production read replica that should never leave your network.
Kontext Studio takes the opposite approach. The app, API, job worker, and Python kernels all run on one machine. Prompts go only to the model provider you configure, and your data stays in the sources you connect. You get the convenience of an AI workspace without handing over a copy of your data.
What you get
- Ask, don't assemble. A coordinator routes work to SQL, chart, notebook, and Python specialists, and every tool call stays visible.
- Outputs you can keep. Answers become data tables, charts, and notebooks that mix prose, SQL, Python, and diagrams.
- Your sources. A managed local lakehouse plus DuckDB, SQLite, PostgreSQL, and MySQL connections over data you already have.
- Your models. Ollama or LM Studio locally, or cloud providers such as OpenAI, Anthropic, Gemini, and Bedrock.
- Grounded context. Curated Knowledge concepts for metrics, joins, and golden queries, with agent proposals you review before they are applied.
- Local-first by default. The app, API, worker, and kernels run on one machine; nothing is sent anywhere except the model provider you choose.
Install and start
Kontext Studio is published to PyPI as the kontext-studio package, and each
GitHub Release carries the matching wheel. It requires Python 3.12 or newer.
With uv:
uv tool install kontext-studio
Or with pip in a virtual environment:
python -m venv .venv
source .venv/bin/activate
pip install kontext-studio
Then start the local server from the directory that holds your project:
kontext serve
This starts the local API server, serves the web app, and runs the embedded local worker. From there you can create a project, connect a source, and open a notebook.
What's inside the repository
The project is a pnpm and uv workspace containing the React app, the
TypeScript packages published as @kontext-tech/*, the Python packages, and
committed generated API types. ts/schema is the contract hub, so the API
surface and the client types move together.
If you want to work on Studio itself, clone the repository and run:
pnpm install
uv sync
pnpm run dev:local
The developer guide covers the architecture, package boundaries, and the generation and release workflows.
Try it and tell us what breaks
Kontext Studio is early, and the fastest way to make it better is to use it on real data. Install it, connect a source you already have, and open an issue when something is confusing or wrong. Contributions are welcome, whether that is a bug report, a documentation fix, or a new package.