Draft for review. This page is under review before the Operational Readiness Review. Panels marked as open questions are still waiting for an answer. Send corrections to the Research Platform team.
Read time: 9 minutes. One section per tool. Each section says what it is, where to find it, an example, and what to watch out for. Prefer a walkthrough in order? Read your first day instead. The team will decide which of the two pages to publish, or publish both.
The tools at a glance
| Tool | It answers… | Available to |
|---|---|---|
| JupyterLab and notebooks | Where do I work? | Everyone |
| Machine sizes and images | How much power do I get? | Everyone |
| Storage | Where do my files go, and what survives a restart? | Everyone |
| Your own kernel | How do I get my own packages? | Everyone |
| Notebook Intelligence | Can AI help me code? | Everyone |
| Data browser | How do I get at cloud data? | Everyone |
| Dask | How do I go faster? | Everyone |
| NBQueue | How do I run something long? | Everyone |
| RTC Launcher | How do we edit together? | Projects with collaboration on |
| Resource Tracker | What am I spending? | Everyone |
| AI Registry | What shared AI pieces can I reuse? | Rolling out |
| Git | How do I manage code versions? | Everyone |
1. JupyterLab and notebooks
What it is. The workspace in your browser. A notebook is a page of cells: you write code in a cell, run it, and the result appears under it.
Where. rp.sciencecloud.nasa.gov, after login.
Example. Open the Launcher, click a Python notebook, type print("hello"), and press Shift+Enter.
Watch out. Your workspace costs money while it runs. Stop it when you finish.
2. Machine sizes and software images
What it is. At launch you choose the size of your machine and the software set.
Sizes. From a small 4 GB workspace up to 64 GB with 8 CPUs, or a GPU machine. (A GPU is a chip that is fast for AI and image work.)
Software. Ready-made Earth-science, Python, and AI stacks. A project can also bring its own image — a packaged set of software.
Example. Training a small model? Choose the GPU machine. Reading one file and plotting it? Choose the small workspace.
Watch out. A bigger machine costs more per hour. Take the size the job needs, not the biggest one on the list.
3. Storage — home, EFS, and object storage
What it is. Places to keep files, with different sharing rules — and different behaviour when your workspace restarts.
| Place | Path | Who sees it | Survives a restart? |
|---|---|---|---|
| Home folder | ~ = /home/<your name>, for example /home/ramon.linan | Only you | ✅ Yes |
| Your own EFS folder | /efs/<your username>, in email form — for example /efs/ramon.linan@navteca.com | Only you | ✅ Yes |
| Project folder on EFS | /efs/<project>, one for every project you belong to | Everyone in that project | ✅ Yes |
| Object storage — yours | s3://<bucket>/home/<your username> | Only you | ✅ Yes |
| Object storage — project | s3://<bucket>/<project> — the project you chose at launch | Your project | ✅ Yes |
| Everything else | /tmp, system folders, packages installed outside your home folder | — | ❌ No. Lost. |
Example. You download a 200 GB dataset for the team. Put it in the project's object storage, not in your home folder. Keep your own notebooks in your home folder. Put a shared intermediate result in /efs/<project>.
Good to know.
/efs/lists a folder for every project you belong to. Permissions stop you from opening other people's folders.- The file browser and the terminal show the same files in a different layout. The permissions are the same in both.
- Object storage in a remote AWS account (your institution's, or a mission's) is not charged to your project. See How you pay for it.
How big is your home folder? It starts at 10 GB, and it can be made larger — ask the platform team. Do not ask for more than you need: a very large home volume makes your workspace slower to start. Put large data in /efs/<project> or in object storage instead.
Watch out. You pay every month for what you store in EFS, in your home volume, and in object storage — used or not. Delete what you do not need.
3b. Your own kernel from a conda environment
What it is. A private set of packages that appears in JupyterLab's kernel menu. Use it when the ready-made software set lacks a package or a version you need.
How to build one:
mamba create -n myscience python=3.12 ipykernel -y conda activate myscience mamba install -c conda-forge xarray netcdf4 -y
Then pick it with Kernel → Change Kernel. It appears by itself — the platform discovers your environments for you.
Good to know.
- The platform keeps your environments in
~/users_conda_envs, inside your home folder, so they survive a restart. - Always include
ipykernelin the environment, or there is no kernel to pick. condaandmambaare both installed.mambais faster.- If a kernel does not show up, register it by hand:
python -m ipykernel install --user --name myscience --display-name "My science env"
4. Notebook Intelligence — AI help in the notebook
What it is. An AI assistant built into JupyterLab. It can chat, complete code as you type, and run small agent tasks such as editing a notebook, running cells, editing a Python file, reading and writing files in your workspace, and running commands.
Where. Inside JupyterLab, next to your notebook.
Example — explain. Select a cell you do not understand and ask for an explanation.
Example — write. In an empty cell, type what you want: "load this CSV, drop empty rows, plot the monthly mean". Read the code it writes, then run it.
Example — fix. A cell throws an error. Ask what the error means and let the assistant correct the cell.
Example — one notebook, one cell. You can point the assistant at a whole notebook, or at a single cell, instead of copying text into a separate chat window.
Models. Several providers are supported, including GitHub Copilot, OpenAI, Anthropic Claude, Ollama, and local or OpenAI-compatible models. Connecting your own is covered in Notebook Intelligence and your LLMs.
Watch out. The assistant can be wrong. Read the code before you run it, and check the science.
5. Data browser
What it is. A panel that shows cloud storage as folders, inside JupyterLab. You browse and download without a command line and without handling keys. (The extension behind it is jupyterlab-bxplorer.)
What you can reach. Your project's own buckets, and public AWS science datasets.
Download only. The browser brings files in. It does not upload or write back.
Example — your own data. Open the panel, pick your project bucket, open the folder for this year, and download one file into your notebook folder.
Example — public data. Pick a public AWS dataset, find the granule you need, and pull it straight in.
Coming later. Google Cloud and Microsoft Azure.
Watch out. Downloading a very large file into your home folder costs storage every month. For big data, read from object storage instead of copying it.
OPEN: the on-screen label, and whether remote accounts appear here. Needs a screenshot.
6. Dask — many machines for one job
What it is. Dask splits one analysis across many machines and collects the answer. The platform runs it through Dask Gateway, so you ask for a cluster from your notebook.
Example. A year of daily files takes hours in one workspace. A 10-machine cluster finishes it in minutes. You start the cluster, run your code, read the answer, and shut the cluster down.
Watch out. The cluster costs money for every machine, for as long as it is up. Shut it down as soon as the job is done. A forgotten cluster is the most expensive mistake on the platform.
Example notebook — to be linked. The platform team will publish a notebook that starts a cluster, runs a job, and shuts it down. The link goes here.
7. NBQueue — background notebook runs
What it is. Send a whole notebook away to run on its own, on a machine you choose. Your interactive workspace stays free, and the job survives closing the browser.
How you use it. Pick the notebook. Pick the machine. Send it. Watch the status in the same panel. When it finishes, you get the finished notebook, with its output.
Example. A 6-hour job needing 40 GB. Keep a small workspace for yourself, and send the job to a big machine in the background. Go home. Read the output tomorrow.
Watch out. The background machine costs money while it runs, exactly like an interactive one.
8. RTC Launcher — the collaboration tile
What it is. A tile in your Launcher named "Open <your project> Collaboration". It opens a shared session for the project, where several people edit the same file at the same time.
Why a shared session, and not "share my screen". Nobody gets into your private workspace, and nobody runs code as you. Everyone works in a separate shared account, and the files land in the shared project space.
Example. Three people must finish one notebook. Everyone clicks the tile and lands in the same session. You all type in the same notebook, and markers show who is online.
Stopping it. Click the tile again to stop the shared session. The tile shows Open when the session is down, and Stop when it is up.
Watch out. The shared session bills the project, not you, for as long as it runs. An idle session is closed automatically after about an hour, but an open browser tab can keep it alive. So stop it on purpose.
Availability. Only projects with collaboration turned on see the tile. See Working together.
9. Resource Tracker — my usage and budget
What it is. A panel showing your own numbers.
- This month — hours used, compute cost, and storage cost split into shared filesystem, workspace disk, and object storage.
- Funding period to date — used, allocated, and remaining. (The funding block is hidden if you have no allocation.)
- Last updated — when the numbers were prepared.
Example. Before starting a GPU machine for a week, open the panel and check what is left of your allocation.
Watch out. The numbers are read when your workspace starts, and the cost data is prepared once a day. The panel does not tick up while you work. To get fresher numbers, stop your workspace and start it again.
10. AI Registry — shared, NASA-validated AI building blocks
What it is. A hub of AI building blocks that NASA colleagues have shared and NASA has validated. Instead of every team building the same AI plumbing, you take a piece that already works and has passed review.
What it holds.
| Block | In plain words |
|---|---|
| MCP servers | A small service that gives an AI assistant a new skill — for example, search a data catalog, or query a database |
| Skills | A packaged way of doing one job well |
| Agents | An AI worker that carries out a task with several steps |
| Models | A trained model you can call |
| Prompts | A tested instruction that gets a good answer |
| Harnesses | The scaffolding that holds the pieces together |
Where. The AgentRegistry panel in the left sidebar of JupyterLab.
What you see. The panel has two tabs:
- Registry — the catalog. Every entry is a card showing its type, whether it is active, how many deployments exist, whether it is published, whether it is deployed, and when it was last updated. There is a link to its repository.
- Deployments — the ones that are running.
You can search by name or by version. Turn on Semantic search to describe what you need in plain words instead of guessing the name.
The three buttons on each card.
| Button | What it does |
|---|---|
| Add to NBI | Hands the tool to your AI assistant (Notebook Intelligence), so it can use it in your next question |
| Deploy | Starts the tool for you |
| Undeploy | Stops it again |
Example. You want your assistant to answer questions from a USGS data service. Open AgentRegistry → Registry, find the USGS MCP server, press Deploy, then press Add to NBI. Now ask your assistant a question about that data, and it uses the tool.
Why it matters. Every item was shared by a NASA peer and validated by NASA. You are not picking an unknown tool off the internet.
Availability. The panel is in the software image used by projects on the platform's newer image today, and it is being rolled out to every project. If you do not see the AgentRegistry icon, ask the platform team.
Watch out. A deployed tool uses resources while it runs. Press Undeploy when you have finished with it.
11. Git
What it is. The standard JupyterLab Git panel, for code versions.
Example. Clone your team's repository, commit your notebook changes, and push them.
Watch out. Do not commit data or secrets into a repository. Data belongs in the shared folder or in object storage.
What you cannot do
- Install your own JupyterLab extensions. The Extension Manager is switched off. Ask the platform team if you need a new extension in the image.