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

Adding a Provider

You connect a provider by choosing an entry from your platform's provider catalog and supplying its credentials. A provider you add is a personal provider: it belongs to you, its models use IDs like your-username:provider/model, and requests through it are not metered by ACTIVATE. To create a provider that the whole organization draws on with budget allocations, see Organization Providers.

Prerequisites

  • AI is enabled for your organization, and the provider type you want is enabled in the platform's provider catalog. See Platform Setup.
  • You have the credentials for the provider: an API key for most entries, a ChatGPT sign-in for Codex subscription, or a linked GitHub account for GitHub Copilot.

Add an API-Key Provider

Most catalog entries, including OpenAI, Anthropic, Google Gemini, GenAI.mil, and custom providers, connect with an API key.

  1. Navigate to AI > Providers.
  2. Click Add provider and choose the catalog entry.
  3. Enter a name in the Provider name field, or keep the suggested one. The provider ID used in model IDs is the name in lowercase with only letters and numbers kept; when the two differ, the dialog shows it as Provider ID: below the field.
  4. Check the Endpoint URL. It is shown for reference only: "The endpoint is defined by the platform catalog and cannot be changed."
  5. Paste your key into API key.
  6. Click Create.

ACTIVATE opens the new provider and confirms with "AI provider created. Enable its models to make it usable." A new API-key provider offers no models until you enable them.

Organization admins

Organization admins also see an Organization-wide switch in this dialog. Turning it on creates an organization provider instead. See Organization Providers.

Enable Models

The provider's details page lists the models the provider reports in its Models card.

  1. Open the provider from AI > Providers.
  2. In Models, click a model to enable or disable it. Use Filter models… to search, and switch between the enabled and all views to review your selection.

Only enabled models appear in Chat, pw code, and the models list for external tools. Only the provider's owner can change which models are enabled. If your platform administrator restricted the catalog entry to a list of allowed models, you only see and enable models on that list.

Connect a Codex Subscription

The Codex subscription entry connects a ChatGPT subscription instead of an API key.

  1. Click Add provider and choose Codex subscription.
  2. Enter a provider name.
  3. Click Connect with ChatGPT.
  4. Click Open ChatGPT and enter the one-time code shown in the dialog. Only enter this code at auth.openai.com.
  5. Once you authorize, the dialog confirms "Codex subscription connected." and opens the provider.

The available models come from your subscription, so there is nothing to enable. If the authorization lapses, open the provider and choose Reconnect subscription from its menu. You can also connect from the terminal with pw ai codex connect; see Models & Allocations.

Connect GitHub Copilot

The GitHub Copilot entry uses the GitHub account linked to your ACTIVATE account.

  1. If you haven't already, link your GitHub account in Account Settings. Until you do, the dialog shows "Link GitHub in Account Settings before connecting Copilot."
  2. Click Add provider and choose GitHub Copilot.
  3. Enter a provider name and confirm the dialog shows "Uses linked GitHub account @your-username."
  4. Click Create.

Copilot determines which models are available for your account and plan. To recheck access later, choose Reconnect subscription from the provider's menu.

Provider Details

Click a provider to open its details page. Depending on the provider type, it shows:

  • Platform, Sharing, and Owner at the top.
  • Endpoint: the address requests are sent to.
  • Connected account: the subscription account, for Codex subscription and GitHub Copilot providers.
  • TLS verification: shown when the catalog entry uses a custom CA certificate or skips verification.
  • Responses API: shown as Enabled when the provider also serves the OpenAI Responses API.
  • Folded history: shown as Enabled when Fold history is on.
  • Tool calling: Native, Emulated from text, or Disabled. See Tool Calling & Emulation.
  • Models: the models card described above.

Edit a Provider

  1. Open the provider and click Edit.
  2. Change the fields you need:
    • Display name: the name shown in lists and model pickers. The provider ID doesn't change.
    • API key: enter a new key to rotate it. Leave the field empty to keep the current key.
    • Tool calling: choose Emulated, Native, or Off. This row only appears when the catalog entry lets connections choose their mode.
    • Fold history: "Sends the whole conversation as one message, for providers that read only the latest message. Turning this on keeps the provider on Chat Completions."
  3. Click Save changes.

The provider's owner and organization admins can edit a provider. Codex subscription and GitHub Copilot providers are updated with Reconnect subscription instead.

Delete a Provider

  1. In AI > Providers, open the menu on the provider's row (or on its details page).
  2. Choose Delete provider.
  3. Confirm with Delete.

Only the provider's owner can delete it. Deleting a provider removes the connection from ACTIVATE; it does not revoke the key at the vendor.

Related Documentation