Multi-Project support is currently in early access. Every Organization includes a Default Project. To enable additional Projects, please contact support.Not all resources are Multi-Project supported yet. See Early Access Limitations for the full list and how unsupported resources behave.
How Projects Work
Default Project
Every Organization has a Default Project. A few things to know about it:- All Organization Members are automatically granted access to the Default Project
- All historical account usage and resources that pre-date Projects are attributed to this Project (work in progress)
- All Playground usage is attributed to this Project
- It has an “Organization Default Key” that is only accessible to the Organization Owner (see API Keys & Authentication)
- Because all Organization Members have access, do not use the Default Project for sensitive resources. Create a separate Project for those.
Managing Project Collaborators
You can manage Project Collaborators from Settings > Project > Collaborators.Adding Collaborators
- Go to Settings > Project > Collaborators
- Click Add Collaborator
- Enter the user’s email address
- Click Confirm
The user must already belong to your Organization, unless they are being added as an External Collaborator.
Removing Collaborators
- Go to Settings > Project > Collaborators
- Find the Collaborator you want to remove
- Click the three-dot menu next to their name
- Select Remove User
- Confirm the removal
Project API Keys
Each Project has its own API keys. These keys authenticate API requests and are scoped to the Project’s resources. For details on creating, managing, and rotating API keys, see API Keys & Authentication.Early Access Limitations
During early access, not all products work correctly across multiple Projects. This applies across in-app experiences, the CLI, SDKs, and the public API.What’s Supported
The following work correctly across multiple Projects:- Instant Clusters — On-demand GPU clusters for training and inference
- API Keys — Create and scope API keys to a specific Project
Not Yet Supported
The following do not yet work correctly across multiple Projects:- Dedicated Endpoints — Always-on inference endpoints
- Volumes — Shared and local storage attached to clusters
- Fine-tuned Models — Custom models trained on your data
- Evaluations — Model evaluation runs
- Files — Training data and uploads
For Organization Members and Project Collaborators
If you use a product that isn’t multi-Project aware, usage attributes back to the Default Project of the Organization. In-app views and API responses for those resources may be inaccurate — they might return all resources belonging to the Organization or only resources in the Default Project, not the Project you’re working in.For External Collaborators
The behavior is more significant for External Collaborators. Because many product decisions are based on the user’s Organization, usage in unsupported products may attribute back to the External Collaborator’s own Organization rather than the Project and parent Organization they are collaborating within. This means billing and resource attribution can be incorrect for external collaborators using products that don’t yet work across multiple Projects. We’re actively expanding multi-Project support to all products. This section will be updated as more products are supported.Common Project Structures
Teams organize Projects differently depending on their needs:| Strategy | Example | Best for |
|---|---|---|
| By team | ml-research, platform-eng, applied-ai | Large Organizations with distinct teams |
| By environment | development, staging, production | Teams that want resource isolation across environments |
| By workload | training, inference, evaluation | Teams that want to separate compute budgets |
| By customer | customer-a, customer-b | Service providers managing multiple clients |
Next Steps
Roles & Permissions
What Admins and Members can do within a Project
API Keys
Create Project-scoped credentials
Cluster Access
Product-specific guide for managing cluster access