...
- Instant access: Start a workspace on demand immediately—no support tickets or waiting required.
- Flexible sizing: Choose your exact compute power at startup: Standard (4 GB memory / 4 CPU), Large (8 GB / 4 CPU), Extra Large (16 GB / 4 CPU), or up to 64 GB / 8 CPU on demand.
- Pre-installed software: Jump right in with Earth science, Python, and AI SI tools already loaded. Advanced users can also bring their own custom container images (a packaged set of software).
- Distributed computing: Spread a single, massive analysis across multiple machines when you need more power.
- Background processing: Send long-running notebooks to process on their own. The job keeps running even if you close your browser, freeing up your active session.
- Auto-scaling capacity: The platform automatically adds and removes computing machines behind the scenes based on user demand.
...
- Your own space: Your own private storage where files are saved securely and are waiting for you when you return.
- Team space: A collaborative file system for your project team to store and share data, models, and results.
- Project object storage: Private, large-scale cloud storage for your project that you can access directly from your notebook.
- Integrated data browser: Browse and download files from your project storage—as well as from public AWS science datasets—directly inside JupyterLab.
Integrated
...
SI Tools
- AI SI in the notebook: Chat with AI SI assistants and run agents directly inside JupyterLab, powered by various model providers.
- Targeted coding help: Ask the AI SI to write code, explain complex sections, or fix errors within a specific notebook or a single cell.
- Agents and a secure gateway: Run AI SI agents on the platform through a single, trusted entry point. For safety, some workflows require human approval before taking action.
- The AI SI Registry in JupyterLab: Access a hub of NASA-validated AI SI building blocks—including MCP servers, skills, agents, models, prompts, and harnesses. Use the AgentRegistry panel to search, deploy, and equip your AI SI assistant with these tools. (Currently rolling out to all projects; see the tool reference.)
- Model endpoints on demand: Import and call your own custom models, or connect to managed models, directly from your notebook.
...
- Faster start-up: Ensuring launch times remain short, even as your personal storage space grows large.
- One-click Hugging Face models: Deploy a public model with a single click and call it directly from your notebook.
- Improved "My Usage" view: See your personal usage and remaining budget immediately on the launch screen.
- Roles in the AI SI Registry: Granular per-user and per-group permissions, replacing the current single access level for everyone.
Planned Next
- Scored AI SI tools: Quality and safety scores for every tool in the catalog, complete with a tracked history.
- Plain-language search: Find models, agents, or tools by simply describing what you need in everyday language.
- Budget controls at launch: Display a true remaining balance and prevent new workspaces from starting when project funds run out. Project admins will also have the option to block members who exceed their personal limits, or allow them to continue spending from the general project pool.
- Storage costs per person: View the exact object storage costs for each individual team member.
...
- Project-defined machine sizes: Project admins can build and maintain a custom list of server types for their specific team, rather than using a single platform-wide list.
- GPU machines: Adding machines equipped with GPUs (Graphics Processing Units), which are specialized chips that significantly speed up AI SI and image-processing workflows.
- Detailed session metrics: Per-session breakdowns of CPU, memory, and storage usage to help you understand costs and choose the perfect machine size for your next run.
- Review before publish: A draft-and-approve workflow before a new AI asset goes live in the catalog.
- Expanded data browser clouds: Adding support for Google Cloud and Microsoft Azure data to the data browser.
...