Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.pullfrog.com/llms.txt

Use this file to discover all available pages before exploring further.

Pullfrog can route LLM calls through your Google Cloud project using Vertex AI instead of going to the model vendor directly. This is useful when your team already buys through Google Cloud, has Vertex model-access approvals, or wants Google Cloud audit and billing controls around inference.
Vertex AI support is BYOK only. Inference is billed by Google Cloud to your account; Pullfrog never proxies Vertex traffic.

​
Setup

Four things, in order: pick the model in the dropdown, add your service-account JSON as a secret, put project/location/model ID in your workflow, and enable the model in Model Garden.

​
1. Select Google Vertex AI in the model dropdown

In pullfrog.com/console/<account>/<repo>, open the model dropdown and pick Google Vertex AI. The console will then show the required env vars. Unlike normal catalog entries, the Vertex entry doesn’t have per-model variants in the dropdown. The actual model — Claude Opus 4.1, Gemini 2.5 Pro, or a partner model — is supplied separately via VERTEX_MODEL_ID (see step 3). Pullfrog never resolves or upgrades that value.

​
2. Store your service-account JSON as a secret

Create a Google Cloud service account with roles/aiplatform.user, download a JSON key, and store the entire JSON blob as a single secret named VERTEX_SERVICE_ACCOUNT_JSON. You can store it as either a Pullfrog secret or a GitHub Actions secret. Pullfrog secrets are auto-injected into every run with no workflow changes; GitHub secrets keep the value in your existing org-wide secret tooling but require an env: mapping in pullfrog.yml.

​
3. Hardcode project, location, and model ID in pullfrog.yml

These values are not sensitive. Put them directly in your workflow’s env: block: 
- name: Run agent
  uses: pullfrog/pullfrog@v0
  with:
    prompt: ${{ inputs.prompt }}
  env:
    # ... your other provider keys ...
    VERTEX_SERVICE_ACCOUNT_JSON: ${{ secrets.VERTEX_SERVICE_ACCOUNT_JSON }}
    GOOGLE_CLOUD_PROJECT: my-project
    VERTEX_LOCATION: global
    VERTEX_MODEL_ID: <vertex-model-id>
(If you stored the service-account JSON as a Pullfrog secret instead of a GitHub Actions secret, drop the VERTEX_SERVICE_ACCOUNT_JSON line — Pullfrog auto-injects it.) GOOGLE_CLOUD_PROJECT can be omitted when VERTEX_SERVICE_ACCOUNT_JSON contains a project_id; Pullfrog will default it from the JSON. Keeping it explicit in pullfrog.yml is usually clearer.

​
4. Enable the model in Model Garden

Enable the Vertex AI API (aiplatform.googleapis.com) for your Google Cloud project, then enable any gated models in Model Garden. Prefer global unless your compliance or data residency requirements need a specific regional endpoint.

​
Picking a model ID

VERTEX_MODEL_ID should be the exact Vertex AI model ID you want to run:
  • a Claude model ID from Model Garden, for example the latest enabled Opus, Sonnet, or Haiku ID
  • gemini-2.5-pro — Gemini 2.5 Pro
  • other Gemini and partner model IDs accepted by Vertex AI
To switch model versions later, edit VERTEX_MODEL_ID directly and commit. Pullfrog never resolves or upgrades the value.

​
How routing works

Pullfrog detects Anthropic Vertex IDs with an anchored claude- prefix check:
  • Claude IDs (for example claude-opus-4-1@20250805) run through the Claude Code Vertex AI integration. Pullfrog sets CLAUDE_CODE_USE_VERTEX=1 automatically, and aliases GOOGLE_CLOUD_PROJECT / VERTEX_LOCATION into Claude Code’s expected env var names.
  • Everything else (Gemini and partner models) runs through OpenCode’s google-vertex provider. Pullfrog prepends the google-vertex/ prefix automatically.

​
Troubleshooting

“VERTEX_MODEL_ID env var is required when the model is set to vertex/byok” You picked Google Vertex AI as the model but didn’t set VERTEX_MODEL_ID in your workflow’s env: block. Add it (see step 3 above). “Google Vertex AI model selected but required configuration is missing” One or more required values isn’t set. The error message lists exactly what’s missing: VERTEX_SERVICE_ACCOUNT_JSON, GOOGLE_CLOUD_PROJECT (unless derivable from the JSON), VERTEX_LOCATION, or VERTEX_MODEL_ID. Run starts but the LLM call returns a Google authentication error Make sure VERTEX_SERVICE_ACCOUNT_JSON is the whole JSON object string, not a path to a local file. Pullfrog writes it to a harness-owned credentials file and sets GOOGLE_APPLICATION_CREDENTIALS internally for the agent process. Run starts but the LLM call returns a permission or model-access error Make sure the service account has roles/aiplatform.user, the Vertex AI API is enabled, and the model has been enabled in Model Garden for the selected region.