Install and start Kontext Labs

Kontext Labs runs on your own machine. The local server, the data engine, and the web app all start together, and everything you create stays in a workspace directory on disk.

Before you start

You need Python 3.12 or newer and uv.

Install

uv tool install kontext-labs

To pin a specific release, use the version number, for example 0.261.0:

uv tool install "kontext-labs==0.261.0"

To upgrade later:

uv tool upgrade kontext-labs

Start the server

Run the server from the directory that should hold your workspace:

kontext-labs serve

The first start creates a workspace in .kontext/ and prints the URL to open. Kontext seeds a placeholder profile called Local Admin and one project called Local Project, so you can start working immediately.

To open the browser automatically:

kontext-labs serve --open

Use --host, --port, or --lake-path to bind a different address, change the port, or keep the workspace somewhere other than .kontext/.

Replace the placeholder profile

The placeholder profile is enough to get started, but it is worth replacing before you add data. The setup page is available at /setup on the same address as the app, for example http://localhost:8000/setup.

1. Kontext starts with a placeholder profile. Open /setup before you add any data to replace it with your own name and email.
2. Enter a first name, an optional last name, and an email address.
3. Select Create local profile. Kontext stores the profile on this machine and signs you in.
4. Kontext opens your workspace with one project and the local profile you created.

Connect a model provider

Kontext does not bundle a model. After signing in, open Model providers and add a local or cloud provider, then choose the default model for your work. See Connect a model provider for the supported providers and settings.

CLI commands

Most work happens in the browser, but kontext-labs also exposes the local workspace from the terminal:

Command	Description
kontext-labs serve	Start the local server and UI
kontext-labs project list	List projects in the local workspace
kontext-labs project create <name>	Create a project
kontext-labs project import <files…>	Import data files into a project
kontext-labs job list	List available job types
kontext-labs job run <type>	Run a job synchronously

project import accepts CSV, Parquet, JSON, and other data files. Pass --project to target a project other than the first one, and --subdir to choose where the files land under the project’s data directory.

Supported platforms

Platform	Architecture
macOS	arm64, x86_64
Linux	x86_64
Windows	x86_64

Python 3.12 or newer is required.