Support GCP credentials (Vertex AI) as a BYOK authentication method

[etirelli]

Pre-submit Checks

• I have searched Warp feature requests and there are no duplicates
• I have searched Warp docs and my feature is not there

Describe the solution you'd like?

Add GCP credentials as a supported BYOK authentication method for accessing AI models via Google Cloud's Vertex AI platform.

Concretely, this means:

• A new GCP / Vertex AI option in BYOK settings, alongside the existing Anthropic and Gemini API key options.
• Support for standard GCP auth methods: Application Default Credentials (gcloud auth application-default login), service account JSON key file, or GOOGLE_APPLICATION_CREDENTIALS environment variable.
• A GCP project ID and region field (required by Vertex AI's API endpoint, unlike direct provider APIs).
• Model routing through Vertex AI endpoints for any model available there — at minimum Claude (Anthropic on Vertex AI) and Gemini (via Vertex AI, distinct from the existing Gemini API key path).

The auth layer is the same for all Vertex AI models, so a single implementation covers the full Vertex AI model catalog — current and future.

Is your feature request related to a problem? Please describe.

Enterprise and GCP-centric teams commonly operate under policies that:

• Require all cloud access to go through GCP IAM and service accounts rather than raw API keys.
• Mandate data residency within GCP's infrastructure (relevant for regulated industries).
• Consolidate billing and auditability through a single GCP project.

Warp's current BYOK options require direct API keys from Anthropic (api.anthropic.com) or Google AI Studio. These don't satisfy GCP-based access controls. The sharpest gap today is Claude on Vertex AI — Anthropic makes its models available through Vertex AI for enterprise customers, but there is no way to use that access in Warp. The same applies to Gemini accessed via Vertex AI (which has a different endpoint and auth model than the Gemini API).

Additional context

• Anthropic's Claude models are available on Vertex AI for enterprise customers (docs).
• Gemini on Vertex AI uses a different endpoint and auth model than the Google AI Studio API key — both paths would benefit from this.
• GCP auth is standardized through the ADC ecosystem, so one auth implementation covers all current and future Vertex AI models.
• This is complementary to, not a replacement for, the existing Gemini API key BYOK — users on AI Studio continue to use API keys; only Vertex AI users need GCP credentials.

Operating system (OS)

macOS, Linux, Windows

How important is this feature to you?

5 (Can't work without it — enterprise GCP users have no path to use Claude or Vertex AI models in Warp today)

Warp Internal (ignore) - linear-label:39cc6478-1249-4ee7-950b-c428edfeecd1

Ignore

[oz-for-oss]

@etirelli

You can view the triage session on Warp.

Here's what I found while triaging this issue:

Current BYOK supports direct provider API keys and AWS Bedrock credentials, but not GCP/Vertex AI credential or routing flows. #9303 is related BYO-provider work, but it does not cover Vertex AI IAM/ADC auth.

Maintainer details

I concluded that feature request to add GCP/Vertex AI credentials and model routing as a BYOK option for Warp AI..

Triage summary

This is an enhancement request to add GCP / Vertex AI as a BYOK authentication and routing option for Warp AI, covering ADC/service-account-style GCP auth plus project and region configuration for Vertex-hosted Claude and Gemini models.

Classification

• Type: enhancement
• Areas: Agent/model-provider routing and AI settings/BYOK configuration
• Reproducibility: high for the feature gap. The current client surfaces inspected only model direct provider API keys and AWS Bedrock credentials; there is no GCP / Vertex AI credential or host path modeled today.

Code context

• crates/ai/src/api_keys.rs stores direct OpenAI, Anthropic, Google, and OpenRouter keys and can attach AWS credentials to AI requests.
• app/src/settings_view/ai_page.rs renders the existing API key inputs and AWS Bedrock credential settings, but no GCP / Vertex AI settings.
• app/src/ai/llms.rs, crates/graphql/src/api/queries/get_feature_model_choices.rs, and crates/graphql/src/api/workspace.rs currently model LLM hosts as direct API or AWS Bedrock, so Vertex AI would require schema/client/server host-routing additions.

Likely implementation areas

Adding this likely needs a Vertex/GCP host and credential model, desktop credential loading for ADC/env/service-account-file flows, project/region settings, request serialization, and routing support for Vertex-hosted Anthropic and Gemini models.

Duplicate check

No duplicate should be marked from the provided recent/open issue list. #9303 is adjacent BYO-provider work, but it asks for custom OpenAI-compatible endpoints rather than GCP IAM/ADC credentials and Vertex AI routing semantics.

This is my automated analysis and may be incorrect. A maintainer will verify the details.

Powered by Oz

[etirelli]

If the ticket is accepted, I am happy to contribute the implementation.