SC26Join us at Booth 5405 · Nov 16–19
Parallel Works

AI Providers

An AI provider is a connection from ACTIVATE to a model service. Providers supply the models you use in Chat, in pw code, and in external tools that authenticate with an AI key. Every request goes through the platform's OpenAI-compatible API, which checks access before it reaches the provider, so provider credentials stay in ACTIVATE.

To see your providers, navigate to AI > Providers in the sidebar.

How a Provider Becomes Usable

Three layers decide whether you can use a model:

  1. Platform setup. A platform administrator turns AI on for your organization and enables provider types in the platform's Provider catalog. You can only connect to a provider type that is enabled there. See Platform Setup.
  2. A connection. A user, or an organization admin, connects a provider from an enabled catalog entry and enables the models it should offer. See Adding a Provider and Organization Providers.
  3. Access. The connection is shared with groups or the whole organization through the aiprovider:use permission. See Managing Permissions.

If the Add provider button is missing, no catalog entries are enabled for your organization yet. Ask your platform administrator.

Provider Types

When you click Add provider, the menu lists the catalog entries your platform administrator has enabled. The built-in entries are:

Catalog entryHow you connect
OpenAI, Anthropic, Grok, Groq, Together AI, Fireworks AI, Mistral, DeepSeek, Google Gemini, Z.ai, Z.ai Coding Plan, GenAI.milAn API key from the vendor. The endpoint is fixed by the catalog.
USAiAn API key. Your platform administrator sets the agency-specific endpoint in the catalog.
Codex subscriptionSign in with ChatGPT. Models come from your subscription.
GitHub CopilotThe GitHub account linked in your Account Settings. Models come from your Copilot plan.
Azure AI FoundryACTIVATE provisions the Azure resources in one of your networks. See Azure AI Foundry.
Self-hosted sessionNot added from this menu. A model server running in an ACTIVATE session becomes a provider when the session exposes it. See Self-Hosted Sessions.

Platform administrators can also add custom providers: private or internal OpenAI-compatible endpoints that appear in the Add provider menu next to the built-in entries. In your provider list, a connection to a custom provider is labeled OpenAI compatible. See Custom OpenAI-Compatible Providers.

Personal and Organization Providers

The providers page groups connections into My providers, Shared with me, and Organization providers. Who owns a connection decides how its usage is billed and who can use it.

Personal providerOrganization providerSelf-hosted session
Created byAny userAn organization admin, with Organization-wide turned onThe session's owner
Model IDowner:provider/modelorg:provider/modelsession:owner:session/model
UsageNot metered by ACTIVATE. The vendor bills the API key or subscription behind it.Metered at the rates the organization admins set, and charged to a budget allocationNot metered
Allocation neededNoYes, from an AI key or the X-Allocation headerNo
Who can use itThe owner, plus the groups or organization it is shared withOrganization admins, plus the members and groups it is shared withThe owner, plus the groups or organization the session is shared with

Only usage on organization providers appears on the AI Usage dashboard.

Model IDs

Every model is addressed by an ID that names its provider:

  • owner:provider/model for a personal provider, for example jsmith:openai/gpt-5.2.
  • org:provider/model for an organization provider, for example org:openai/gpt-5.2.
  • session:owner:session/model for a model served from a session, for example session:jsmith:my-vllm/llama-3.1-8b.

Here provider is the provider ID, not its display name. Copy exact IDs from Your models on the Connect Tools page, or list them with pw ai models ls.

Suspended Providers

A provider whose subtitle reads Suspended by platform policy can't serve requests. This happens when a platform administrator disables its catalog entry, disables AI, or removes your organization from AI access. Its configuration, credentials, and permissions are kept, and it resumes when access is restored.

Next Steps

  • Chat: Using the chat interface
  • AI Keys: Programmatic access through the OpenAI-compatible API
  • AI Usage: Spend and tokens on organization providers