SC26AI agents lab + dinner on the river · Nov 17
Parallel Works
View as Markdown
Share:

Organization Providers

Requiresorg:admin

An organization provider belongs to the organization rather than to a user. It is the only kind of provider ACTIVATE meters: each request is priced at the rates an organization admin sets for the model and charged to a budget allocation. Use organization providers when the organization pays for a vendor key centrally and wants spend tracked per allocation.

Organization providers appear under Organization providers on the AI > Providers page, with the organization as their owner. Their models use IDs like org:provider/model.

Create an Organization Provider

  1. Navigate to AI > Providers.
  2. Click Add provider and choose a catalog entry.
  3. Enter a Provider name and paste the vendor's API key.
  4. Turn on Organization-wide.
  5. Click Create.

ACTIVATE confirms with "Organization AI provider created. Enable its models to make it usable." and opens the provider. A new organization provider is shared with the whole organization, but it offers no models until you enable them.

The Organization-wide switch is offered for API-key catalog entries and for GitHub Copilot. Azure AI Foundry providers are always personal.

Enable Models and Set Pricing

An organization provider only exposes models that an organization admin has enabled with a price.

  1. Open the provider from Organization providers.
  2. In Models, click a model that isn't enabled. The Enable model dialog opens.
  3. Enter the rates in dollars per 1M tokens:
    • Input tokens and Output tokens. Leave both at 0 for free usage; usage is still tracked.
    • Cached input tokens: what the provider charges for a prompt-cache hit. 0 means the input rate applies.
  4. Optionally set Context window (tokens), which sizes context management such as pw code auto-compaction. Leave it blank to use the value the provider reports or the model catalog lists.
  5. Click Enable model.

The Pricing card lists each enabled model as Free or Metered with its rates. Click Edit on a model to change its pricing, and click an enabled model in Models to disable it. If a model shows Usage not reported, the provider didn't return token usage for it, so metered billing can't be applied.

Allocations

Every request to an organization provider must name a budget allocation that allows AI usage and that the user can bill:

  • Chat sends the allocation chosen in its Allocation picker.
  • pw code takes --allocation or /allocation. See Models & Allocations.
  • AI keys are bound to an allocation when they are created. See AI Keys.
  • Other clients using a platform API key send the allocation name in the X-Allocation header.

A request with no allocation is rejected with "X-Allocation header is required for org provider requests". Requests are also rejected when the allocation doesn't allow AI usage or has no budget left. To set up allocations, see Allocations.

Control Access

Organization admins can always use organization providers. Other members need the provider shared with them.

  1. In Organization providers, open the provider's row menu and choose Manage access.
  2. Turn Allow all organization members on or off.
  3. In Groups, enter group names separated by commas.
  4. Click Save access.

Members only see the organization providers they can use. See Managing Permissions for how access is checked.

Edit or Delete

Only organization admins can edit or delete an organization provider. Choose Edit to change the Display name, rotate the API key, or set Tool calling and Fold history (see Adding a Provider), then click Save changes. Choose Delete provider to remove it.

For a GitHub Copilot organization provider, choose Reconnect Copilot from its row menu to recheck the linked account.

Related Documentation