Article

    Connect Anthropic Claude through Google Vertex AI

    6 min read
    Last updated 10 hours ago

    Connect Unthread to Anthropic Claude models in your Google Cloud project through Google Vertex AI. Your Google Cloud account manages model access, quota, and inference charges.

    Before you start

    • You need permission to manage your Unthread workspace settings.
    • A Google Cloud administrator must be able to enable APIs and Claude model access, grant permissions, and create a service-account JSON key.
    • Your Google Cloud project needs billing enabled and available quota for your chosen Claude model.

    This guide covers the Google-managed Claude offering accessed through Vertex AI. Google may refer to Vertex AI as Agent Platform in its console and documentation.

    1. Enable Claude in Google Cloud

    1. Open the Google Cloud console and select the project that will run your model requests. Copy its project ID.
    2. Enable the Vertex AI API (aiplatform.googleapis.com) for that project.
    3. Open Model Garden, find the Claude model you want to use, and click Enable. Have an authorized administrator complete any required provider terms or access approvals.
    4. Check the model's supported locations and copy its exact Vertex model ID. Enable each model you intend to use.

    For example, Google's Claude Sonnet 4.6 model page lists claude-sonnet-4-6 and the us-east5 location. Confirm current availability for your selected model before configuring it.

    See Google's Claude setup documentation for model enablement requirements.

    2. Create a service-account JSON key

    1. In Google Cloud, open IAM & Admin → Service Accounts → Create service account.
    2. Give it a descriptive name, such as unthread-vertex-ai, and click Create and continue.
    3. Grant the account the Vertex AI User / Agent Platform User role (roles/aiplatform.user) on the project that will run inference, then finish creating the account. A custom role must include aiplatform.endpoints.predict. See Anthropic's Google-managed Claude requirements.
    4. Open the service account, select Keys → Add key → Create new key, choose JSON, and click Create.
    5. Store the downloaded file securely. You will paste its complete contents into Unthread in the next step.

    The JSON contains a private key. Enter it only in Unthread's credential field; do not include it in tickets, screenshots, or email. If your organization blocks service-account key creation, consult your Google Cloud administrator and Unthread support before proceeding.

    Google documents service-account creation and JSON key creation in more detail.

    3. Configure the model in Unthread

    1. Open Workspace Settings → Artificial Intelligence → AI Model.
    2. Click Use a custom model, or edit your existing custom configuration.
    3. Choose Google Vertex as the provider.
    4. Fill in the fields below.
    5. Click Save custom model, or Save model settings when editing.
    FieldWhat to enterExample
    ProviderGoogle VertexGoogle Vertex
    Modelanthropic/ followed by the exact Vertex Claude model IDanthropic/claude-sonnet-4-6
    Google Cloud ProjectYour project ID, rather than its display name or numeric project numbermy-google-cloud-project
    Google Cloud LocationA location supported by your selected modelus-east5
    Service account JSONThe complete contents of the downloaded JSON key filePaste the JSON object, including its opening and closing braces

    Keep the anthropic/ prefix in Unthread's Model field. It tells Unthread to use Claude through Vertex. Preserve any version suffix included in the Vertex model ID. The Anthropic provider option connects to Anthropic's direct API; this setup uses Google Vertex.

    The credential must be the service-account JSON, rather than a Google AI Studio key, an Anthropic API key, a file path, or only the private-key value. When editing the same provider, you can leave the credential field blank to reuse the saved credential.

    4. Test the connection

    Unthread checks model access before saving, which can make a small model request. After saving, run an AI action that uses your workspace's configured model, such as generating a suggested response, and confirm that it returns a response. Test any AI features with separate model overrides as well.

    Troubleshooting

    • Permission denied / 403: Confirm that the service account has the required inference permission on the project entered in Unthread, the API is enabled, and the selected Claude model is enabled. Check organization policies that may restrict access.
    • Model not found / 404: Check the exact Vertex model ID, its version suffix if applicable, the anthropic/ prefix, and the selected location's support for that model.
    • Invalid credentials or JSON: Paste the complete, unmodified JSON file. It should include type: service_account, project_id, client_email, and private_key. Confirm that the account and key are active.
    • Quota exceeded / 429: Review the selected model's quota and available capacity in Google Cloud, confirm billing is active, and retry or request additional quota as appropriate.
    • Google Vertex is unavailable in your workspace: Contact Unthread support.

    When contacting support, share the error message, model ID, location, and approximate request time. Keep the JSON key and private key out of your message.

    Replace credentials or return to the default model

    To rotate credentials, create a replacement JSON key, save it in the same Unthread configuration, and verify an AI request succeeds before deleting the old key in Google Cloud. To switch back to Unthread-managed model settings, choose Use Unthread default on the AI Model page.