Gcp Vertex Ai

Vertex AI is GCP's ML platform — Gemini model access (text, multimodal, embedding), custom model training (managed GPUs, distributed training), model deployment to prediction endpoints, AutoML for no-code training, and Vertex AI Pipelines for orchestrated ML workflows. For agents: use the Generative AI on Vertex API for Gemini inference and embeddings (text-embedding-004 for RAG), Vertex AI Vector Search for managed vector retrieval, and custom prediction endpoints for deployed models. Confidence is docs-derived.

Python SDK: from vertexai.generative_models import GenerativeModel. vertexai.init(project=project_id, location="us-central1"). model = GenerativeModel("gemini-1.5-pro"). response = model.generate_content("Hello!"). response.text. Embeddings: from vertexai.language_models import TextEmbeddingModel. model = TextEmbeddingModel.from_pretrained("text-embedding-004"). embeddings = model.get_embeddings(["text1", "text2"]). Prediction endpoint: from google.cloud import aiplatform. endpoint = aiplatform.Endpoint(endpoint_name). endpoint.predict(instances=[...]).

Service account or Application Default Credentials (ADC). Roles: roles/aiplatform.user (inference), roles/aiplatform.developer (training + deployment). ADC: gcloud auth application-default login. HTTPS enforced. VPC Service Controls for data isolation. Managed identity for GKE/Cloud Run workloads.

ResourceExhausted (RESOURCE_EXHAUSTED) for quota limits — use quotas page to increase. FAILED_PRECONDITION for model not enabled in project. Vertex AI SLA: 99.9% for prediction endpoints. Training job failures: check job logs in Cloud Logging. Gemini API rate limits: per-minute and per-day limits by model tier. Streaming: GenerativeModel.generate_content_async with stream=True.

cloud.google.com/vertex-ai/docs covers Gemini API, custom training, prediction endpoints, AutoML, pipelines, and vector search. Getting started: gcloud auth, enable Vertex AI API, first Gemini inference in under 5 minutes. Free tier: Gemini 1.0 Pro has free quota. SDK for Python, Java, Go, Node.js. Google Cloud Skills Boost for tutorials. Community via Google Cloud Community.