A shared DGX Spark is most useful when colleagues can reach the tools they need without learning how to operate containers or manage model files. Team AI Box provides a browser workspace with local AI applications, team files, messaging and virtual devices on a supported Spark.
This guide helps you set up team accounts, share files and use local AI in everyday work. For an application overview, see what you can do with DGX Spark.
1. Prepare the appliance, not every laptop
Follow the installation guide on a fresh supported DGX Spark. Initial source preparation needs internet access to download models and dependencies and build the package. For a disconnected target, prepare the full offline package on a connected compatible machine first.
Once installed, team members open the appliance’s current LAN HTTPS address in their browsers. A locally generated certificate needs to be verified against the device before it is trusted. Do not ask colleagues to ignore arbitrary security warnings.
2. Give people individual accounts
Use the administrator account to complete setup, change the factory password and create ordinary accounts. Keep administrator access for the people responsible for the appliance. Sharing one administrator login removes the separation that individual accounts are intended to provide.
Decide who can use which applications and files. Before adding sensitive material, confirm that the intended members can access it and unintended members cannot. The public marketing website is not your appliance login and does not host your team workspace.
3. Start with a concrete team task
One useful starting task is preparing meeting material: upload an authorized recording or document, use transcription or OCR where appropriate, review the result and ask the assistant to help draft notes. Keep the source and reviewed output in the team’s agreed folder, then discuss the result in messaging.
Carry out each step in the appropriate application. Before sharing results externally, review names, numbers, translations and generated content.
For visual work, image generation can help explore ideas. For a software demonstration, a Windows or Ubuntu virtual computer provides an isolated guest environment.
Turn AI answers into useful working material
In the AI assistant, organize work by project and return to a previous conversation when you need its context. Ask for a comparison table, an explanation with a formula or a Mermaid workflow diagram. Answers display formatted Markdown as they arrive; code blocks include a copy button so you can reuse the source.
Open supported workspace images and image attachments for a larger preview, or download available file attachments. Tool cards show whether an operation is waiting, running, completed or failed. Scroll back to review earlier messages without losing your place, then jump to the latest output. Images hosted on other websites open only when you choose their external link.
4. Define the local-data boundary
Local inference keeps model processing on the Spark. External model providers are optional, and enabling one changes where its requests are processed. A local-first workspace is not a blanket guarantee that every configured service is disconnected from the internet.
Agree on which files may be uploaded, whether external providers are permitted and who maintains the system. Avoid exposing the appliance directly to the public internet with default credentials. Remote access needs a deliberate, secured network design rather than an assumption that a public website link grants safe access.
5. Plan for shared resources
AI inference, image generation and virtual machines share finite compute, memory and storage. Start with the actual jobs your team needs and observe responsiveness before expanding usage. No fixed seat count follows from a hardware specification alone.
Application status shows whether a service is still starting or ready to use. Open the task center to follow progress and access completed results.
A practical handover checklist
- Document the appliance’s access address and administrator ownership without publishing credentials.
- Change the factory password and create named user accounts.
- Agree on shared-file access and external-provider policy.
- Explain that first model startup and long-running jobs can take time.
- Provide the installation and recovery documentation to the maintainer.
- Keep operating-system and third-party licenses separate from the project’s MIT license.
Explore Team AI Box’s applications or contribute on GitHub. Improvements to documentation, accessibility and translations are welcome alongside code contributions.