Skip to content
faros Docs
Open console

Enter a topic to search documentation.

    Connect a model

    Create and test a workspace model credential before assigning it to an agent.

    Prerequisites

    Enable AI agents in your selected workspace. You need permission to manage its model credentials, an OpenAI-compatible endpoint, an exact model identifier served by that endpoint, and an API key permitted to invoke it. Model calls can incur charges at your model provider.

    Add and test the credential

    1. Open AI agents → Models and select Add model credential.
    2. Enter a Name, such as notes-assistant-model, using lowercase letters, digits, and hyphens.
    3. Choose Provider, then verify Base URL against your endpoint. The preset is a starting value, not proof that your account can use it.
    4. Enter the exact Model identifier and API key, then choose Add credential. Faros stores the key as a workspace Secret.
    5. Find the new credential and select Test. Expect healthy with a latency value. untested means no successful check has been recorded; failed requires investigation.

    If testing fails, check the endpoint URL, key validity, model availability, and the provider runtime’s outbound network access. Correct the credential with Rotate / model, save it, and test again. Never paste the key into an agent prompt or diagnostic report.

    Expected result: the credential reports healthy with latency, and an agent can use it for a short text response. A healthy credential test does not prove tool access or every model feature.

    Assign the model

    Create an agent and select this credential as its model. Run a short conversation to verify generation. A healthy credential test does not validate tool permissions or every model feature.

    Rotate or remove

    Use Rotate / model to change the model, endpoint, or key. Leaving the new key blank keeps the existing key. After saving, test the credential and an assigned agent again.

    Review the credential’s primary/fallback agent assignments before deleting it. Reassign those agents first; deletion does not revoke the upstream API key. Revoke an exposed key at its issuer separately.

    Next: create an AI agent . If the credential remains failed after one corrected test, use AI agents troubleshooting and provide the status, endpoint class, and request/run ID without the key.