AI Providers & Models3 min read

GCP Vertex AI

Gemini and Claude via Google Cloud

Vertex AI is Google Cloud’s managed model platform. Through the GCP Vertex AI provider entry, Nexgile Code can run both Gemini and Anthropic Claude models (plus several partner-hosted open-weight models) entirely inside your Google Cloud project — with your IAM, your billing account, and your data-residency controls.

Prerequisites

The settings panel itself lists the three setup steps, with links to the Google Cloud docs:

Create a Google Cloud account, enable the Vertex AI API, and enable the specific models you want (Claude models require enabling each partner model in the console).
Install the Google Cloud CLI and configure application default credentials (ADC) — or
Create a service account with Vertex AI permissions and download its JSON key.

Grant the identity a role that can call Vertex predictions (for example the Vertex AI User role).

Configure in Nx IDE

Open Settings → Providers and choose GCP Vertex AI. Fill in:

Field	What it takes
Google Cloud Credentials	The service-account key JSON pasted inline (optional)
Google Cloud Key File Path	Path to a service-account key file on disk (optional)
Google Cloud Project ID	The project that has Vertex AI enabled
Google Cloud Region	Region dropdown — includes `global` plus specific regions

Credential resolution order: inline JSON first, then the key file path, and if both are empty, application default credentials — so on a workstation where you’ve run gcloud auth application-default login, you can leave both credential fields blank.

Models through Vertex

Both major families, one auth path:

Gemini — gemini-3.1-pro-preview, gemini-3-pro-preview, gemini-3-flash-preview, the 2.5 Pro/Flash/Flash-Lite line, and older 2.x/1.5 models. Same behavior as the direct Google Gemini provider.
Anthropic Claude — claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-sonnet-4-6, claude-sonnet-4-5@20250929, claude-haiku-4-5@20251001, plus earlier 4-series, 3.7 (including the :thinking variant), 3.5, and 3.x models. Vertex model ids use an @ before the date stamp. Same thinking controls and caching behavior as the direct Anthropic provider.
Partner MaaS models — models-as-a-service entries: llama-4-maverick-17b-128e-instruct-maas, gpt-oss-120b-maas, and gpt-oss-20b-maas. Vertex publishes more partner models than these; Nx IDE is the TAA-compliance edition, so Chinese-origin families are filtered out of the dropdown even where your project has access — see Providers Overview.

For Claude Sonnet 4/4.5/4.6 and Opus 4.6, an Enable 1M context window (Beta) checkbox appears, extending context from 200K to 1M tokens at tiered pricing.

Why enterprises pick Vertex

One vendor relationship. Frontier models from two labs under an existing Google Cloud agreement — no new procurement cycle, no separate API invoices.
IAM-native access control. Who may call which model is a cloud IAM question, not an API-key-sharing question; keys can be short-lived service-account credentials or no static keys at all (ADC).
Data governance. Requests stay within your chosen project and region, subject to Google Cloud’s enterprise data-use terms.
Unified billing and quotas. Model spend appears alongside the rest of your GCP usage, with project-level quota management.

If your organization is AWS-based instead, AWS Bedrock plays the identical role there.

Troubleshooting

Permission errors: confirm the Vertex AI API is enabled and the identity has a Vertex user role in that project.
Model not found: Claude and partner models must be individually enabled in the Vertex console, and not every model exists in every region — try global or the model’s documented regions.
Works in gcloud but not in Nx IDE: clear one of the credential fields; an empty inline JSON field beats a valid key file only when it is truly empty, and a stale pasted key overrides your ADC login.