Workspaces on demand
Jupyter or VS Code, started in your Kubernetes and opened in the browser. No laptop to set up, no data on a laptop.
← Overview
Notebooks
Jupyter and VS Code, behind your SSO
A workspace opens in one click, in the browser, at the size you choose. You are signed in through your SSO, your tables and storage are already wired in, and the notebook that works becomes a scheduled job without being rewritten.
Key capabilities
Explore without installing anything
The right size, GPUs included
You choose an instance type when the workspace opens; project quotas define what is available.
Everything already wired
The project’s S3 storage, Trino SQL and experiment tracking are available from the start, with no credential to copy.
A coding assistant on your model
Completion, explanation and code generation in Jupyter and VS Code, served by graal’s LLM gateway and the model you chose.
From notebook to scheduled job
A notebook runs as is as a job, with its parameters. The executed notebook, cells and outputs included, is kept with the run.
Versioned in Git
Notebooks and scripts live in the project’s Git repository: history, review and rollback, like the rest of your code.
How it works
From exploration to production
Step 01
Open a workspace
You pick Jupyter or VS Code and an instance type. The workspace starts in the project, under your permissions.
Step 02
Explore
You query catalog tables, read project files and track your trials as experiments.
Step 03
Schedule
The notebook you keep becomes a job: you give it a trigger, and it joins a workflow.
Single sign-on, all the way to the workspace
The workspace opens behind graal’s application gateway, which checks your SSO session before letting a single request through. No port is exposed, no token is copied by hand: the right to open a workspace is a project permission, and it is withdrawn like any other.
Standards and integrations
Tools your teams already know
- JupyterLab
- VS Code
- Python
- Trino
- S3
- MLflow
- Git
- OIDC
Governance
A workspace is a project resource
- Access through your SSO, with no workspace-specific password
- Each workspace belongs to a project and only reaches its data
- Openings, shutdowns and executions are recorded in the audit trail
Frequently asked questions
Can our data scientists install their own libraries?
Yes, in their workspace. For a job, libraries are declared at project level and served from your own registry.
Does our code go to an AI provider?
Only if you chose so. The assistant goes through graal’s LLM gateway, which can serve a model hosted on your premises.
Do we have to rewrite a notebook to schedule it?
No. It runs as is, with its parameters, and the executed notebook is kept as an artifact of the run.
From notebook to job, in minutes
A Jupyter workspace, a catalog query, then the same logic scheduled as a job.