Managing Organization Settings
Invite members, add workspaces, create organization-level integrations, secrets, and API keys.
About This Task
All settings you configure on the Organization Settings page apply across the entire organization, including all its workspaces. For example, if you add an integration here, it becomes available to every workspace in the organization.
Access Organization Settings
- Click your profile icon and choose Settings.
- Go to Organization.
- Open the tab for the setting you want to update, for example Integrations for organization-wide connections. To invite people, click Invite members.
Fleet Environments
Fleet environments let you group services under shared hosting settings, so you don't have to configure the size, replicas, and idle timeout individually for every service. On the Environments tab, organization admins create, edit, and delete environments.
When you create or configure a service, choose how to host it:
- Environment: The service inherits the environment's settings. They're shown read-only, with a link back to the Environments tab.
- Custom: You set the service's own size, replicas, and idle timeout. Switching between an environment and custom hosting, or between two environments, restarts the service, so you're asked to confirm the change first.
For steps to create, edit, and delete environments and to choose an environment for a service, see Manage Fleet Environments.
Compute Tab
Organization admins use the Compute tab to choose which instance sizes your organization can use, how many warm instances to keep ready for each size, and which size is the default. For steps, including what happens when you turn off a size that's in use, see Managing Compute Settings.
Invite Multiple Members and Assign Them to Workspaces
When you click Invite members, you can invite one or several people in a single action:
- In the Emails field, type an email address and press Enter, comma, or space to add it as a tag. Repeat for each person you want to invite, or paste a comma- or space-separated list of addresses. Addresses that aren't valid emails are automatically removed from the field, and listed in an error message below the field so you can fix or remove them.
- In Add members to workspaces, select one or more workspaces. All the people you invite in this action get the same role and access to the same workspaces.
- Click Invite. Each invited address is sent its own invitation. If you invited multiple people, you get a separate success or error notification for each invite, so you can tell which invites went through even if one of them failed.
Environments Tab
Use Environments to manage the shared compute that your services run on. For background, see Gateway.
View Environment Metrics
Services that use the same environment share its compute instances. To find out which services put the most load on them, check the environment's metrics. On the Environments tab in Organization Settings, click the chart button in the environment's row. The metrics panel expands.
The metrics panel shows the services in your current workspace that use the environment, with their request counts for the last seven days. The busiest service is listed first. It also shows:
- The total number of services and requests.
- CPU, memory, and replica charts for all pods in the environment. For details, see Check an Environment's Resource Usage.
Click the chart button again to collapse the panel.
Environment Metrics Limits and Permissions
- The metrics panel only includes services in your current workspace. Services and requests from other workspaces that use the same environment aren't counted in the total.
- Viewing request counts needs permission to read search history. If you don't have this permission, you can still see which services use the environment, but request counts are hidden and a message explains why.
- If a service's request count can't be loaded, the environment total shows as unavailable (
n/a) until every service's count is in. - If no service in your current workspace uses the environment, the panel says so.
- The CPU, memory, and replica charts appear only if at least one service that uses the environment is in your current workspace.
Integrations Tab
Use Integrations to connect external services for your organization. Setting up a connection in Integrations makes it available to every workspace in the organization. You can then use components from the connected service in your pipeline.
External Connections
Register OAuth providers so users can authenticate with third-party services such as Microsoft, Google, Slack, or Notion. You can add, view, and delete OAuth providers from this section. For step-by-step instructions, see Configure External Connections.
Prewarm Models Tab
Use Prewarm Models to shorten cold starts for pipelines that use Hugging Face models. When you list the repo ids of the models your pipelines use, we download them onto your organization's spare instances before a service starts, so your service doesn't wait for the download. Prewarming is best-effort, so in some cases a service still downloads models when it starts. For details, see How Prewarming Works. If you leave the list empty, models download when the service starts.
Only organization admins can view and change this setting.
Add Models to Prewarm
- Go to Settings > Organization and click Prewarm Models.
- In the Hugging Face models field, type a Hugging Face repo id, for example
intfloat/multilingual-e5-base, and press Enter. You can also paste several ids separated by commas, spaces, or new lines. Each id appears as a tag. - To remove a model, click the close icon on its tag.
- Click Save changes. The button is active only when you have unsaved changes. When Your prewarm models are saved. appears, your changes are saved. They apply only to spare instances that start after you save, not to instances that are already running.
If the tab shows We couldn't load your prewarm models., click Retry to load the current list before you make changes.
How Prewarming Works
Prewarming is best-effort. It speeds up startup, but it never stops an instance from starting.
- A spare instance becomes available as soon as it starts. It doesn't wait for the model download to finish.
- If a download fails, for example because a model repo can't be reached or the download runs out of memory, the spare instance stays available. Your service downloads the models when it starts, as it does without prewarming.
- If a service claims a spare instance within a few seconds of the instance starting, the download might not be finished, so the service downloads the models when it starts.
Repo ID Requirements
A repo id is either a model name or namespace/name. It can contain letters, digits, underscores, periods, hyphens, and at most one slash. Each segment must start with a letter, digit, or underscore. Leading and trailing spaces are removed, and duplicate ids are saved once. We reject ids that don't follow these rules.
Prewarming doesn't support gated Hugging Face models that need an access token.
To learn how to use Hugging Face models in your pipelines, see Use Hugging Face Models.
Was this page helpful?