Google Cloud's managed platform for running, tuning and serving models, and the enterprise route to the Gemini API, with grounding, batch prediction, tuning and a model garden.
Vertex AI is Google Cloud's managed platform for running, tuning and serving models, and the enterprise route to the Gemini API. It exposes the same models as Google AI Studio with the controls a production deployment needs: project-level access management, regional data residency, private networking, quotas and billing attached to a Cloud project.
Grounding against Google Search and against private data stores, so a request is answered from retrieved sources rather than from parametric memory. Batch prediction for large offline jobs at reduced cost. Supervised fine-tuning and distillation pipelines. A model garden carrying Google's own models alongside third-party and open-weight ones. Embedding endpoints feeding a managed index for vector search.
Vertex AI is where grounded generation can be run and measured under controlled conditions. The same prompt can be issued with search grounding on and off, with token probabilities enabled, and the two runs compared. That is the setup behind probing what a model recalls from training against what it retrieves live, which is how Tree Walker works.
