agent-starter-pack

Why we included this project

Getting from a working agent prototype to something you can actually deploy is the messy part, and this pack removes much of the hand-wiring. It generates a complete project in one command: backend code, a frontend, and deployable infrastructure up front. The templates cover common patterns like ReAct, RAG over Vertex AI Search or Vector Search, multi-agent setups, and a real-time multimodal agent, so you start from a sensible baseline rather than an empty directory. On top of scaffolding, it adds the operational pieces that tend to get postponed: CI/CD, Vertex AI evaluation runs, an interactive playground, and observability wired into Cloud Run or Agent Engine. One caveat: the project is in maintenance mode and Google is steering new development toward agents-cli, so treat it as a proven starting point and plan to migrate if you build something long-lived.