Adding a model provider

Connect Workforce Intelligence to a model provider such as Anthropic, OpenAI, Amazon Bedrock, Azure OpenAI, Google Vertex AI, Groq, Mistral or Ollama, and edit it later.

For: Administrators (the admin ability) · Last updated

Coworkers think with the models of the providers you connect. You bring your own provider keys, or point the product at an OpenAI-compatible endpoint you run yourself. Provider management is limited to people with the admin ability. Anyone else who opens the page sees "Managing providers requires the admin role." and the provider list without any buttons.

Open the providers page#

  1. Select Settings, then the Administration tab (it appears only for people with the admin ability).
  2. On the Model providers card, select Manage.

The page lists every provider in a table with the columns Provider, Key, Models and Status. Below the table are the default-model panels, including Image reading model, described in Default models.

Add a provider#

  1. Select Add a provider.
  2. Enter a Name. This is the label people see in model pickers, for example Anthropic or Ollama (local). Each provider needs a name that is distinct from the others, ignoring capitalization and punctuation, and names are limited to 80 characters.
  3. Choose a Type (see the table below).
  4. Fill in the fields the type asks for.
  5. Select Add provider.

Add provider stays unavailable until the name is filled in and the type's requirements are met. Select Cancel to close the form without saving.

Provider types#

Type Key Base URL
Anthropic Required Not shown
OpenAI / OpenAI-compatible Required unless you give a base URL Optional
Google Gemini Required Not shown
xAI Required Not shown
OpenRouter Required Not shown
DashScope (Alibaba) Required Not shown
Groq Required Not shown
Mistral AI Required Not shown
DeepSeek Required Not shown
Together AI Required Not shown
Fireworks AI Required Not shown
DeepInfra Required Not shown
Perplexity Required Not shown
Cerebras Required Not shown
Amazon Bedrock None for the server's IAM role; otherwise access keys or a Bedrock API key Not shown (choose a region instead)
Azure OpenAI Required Not shown (enter the resource's endpoint instead)
Google Vertex AI A service account key file Not shown (choose a location instead)
Ollama Optional Required
  • Amazon Bedrock replaces the key and base URL fields with Authentication, Region and an optional Custom endpoint. Add provider stays unavailable until you choose a region and give the credentials the authentication option needs. See Use AWS Bedrock.
  • Azure OpenAI and Google Vertex AI replace the key and base URL fields with their own fields. See Use Azure OpenAI and Use Google Vertex AI.
  • Keyed providers need an API key of at least 8 characters.
  • Groq, Mistral AI, DeepSeek, Together AI, Fireworks AI, DeepInfra, Perplexity and Cerebras need only a key. Each is always reached at its own address, so there is no base URL field, and the key is only ever sent to that provider. The model list comes from the provider, except for Perplexity, which has no model-list service: it offers its Sonar chat models, so Test on a Perplexity provider cannot check the key (a wrong key shows up on the first Coworker request).
  • OpenAI / OpenAI-compatible works against api.openai.com when you leave Base URL (optional) blank and give a key. Enter a base URL to reach any OpenAI-compatible host, such as vLLM or LM Studio. For a host that needs no key, leave API key (optional for local endpoints) blank and give a base URL instead.
  • Ollama needs the address of your Ollama server in Base URL, with or without a trailing /v1. The key is optional. The form adds a checkbox, Disable extended thinking for this provider's models by default, which you can tick if you do not want these models to use extended thinking unless asked.

A base URL must start with http:// or https://. The server connects to the address you enter, so use only addresses you trust.

What happens after you add one#

The provider becomes active right away and the product fetches its list of models.

If the model list cannot be fetched, the provider is still saved, and a red message under Models reads "Provider added, but fetching its models failed:" followed by the reason. Check the key or address, then select Test as described in Testing, activating and removing a provider.

Read the table#

  • Provider shows the name, with a green dot when it is active, then the type and base URL.
  • Key shows a masked label of the stored key, or "no key". The full key is never shown.
  • Models shows up to five model chips and "+N more" for the rest. A model marked (embed) is an embedding model. A model marked (unverified) is one whose context window cannot be confirmed.
  • Status shows active or inactive.

Edit a provider#

  1. Select Edit on the provider's row.
  2. Change the Name, Type, Base URL, the thinking checkbox (Ollama) or the key. Leave the key field blank to keep the current key, unless you change the Type: a stored key is never sent to a different provider type, so you enter a key for the new type (a type that needs no key, such as Ollama, simply drops the old one). For Amazon Bedrock, Azure OpenAI and Google Vertex AI, see Use AWS Bedrock, Use Azure OpenAI and Use Google Vertex AI for the fields.
  3. Select Save changes.

Saving fetches the model list again. If a new key is entered, the provider becomes active.