Agent Starter Pack
Ship AI Agents to Google Cloud in minutes, not months. Production-ready templates with built-in CI/CD, evaluation, and observability.
Last verified:
What is Agent Starter Pack?
Agent Starter Pack is a Python package that provides production-ready templates for Generative AI agents on Google Cloud. It enables developers to go from idea to deployment in minutes instead of months by offering pre-built agent templates including ReAct, RAG (Retrieval Augmented Generation), multi-agent systems, and Live Multimodal API implementations. The tool handles all the infrastructure concerns so developers can focus on their unique agent logic.
Key features include pre-built agent templates implementing common AI patterns, built-in CI/CD pipelines with Google Cloud Build or GitHub Actions, production-ready infrastructure with Terraform for Cloud Run or Agent Engine deployment, integrated observability with OpenTelemetry GenAI instrumentation, Vertex AI evaluation capabilities, and an interactive playground for local testing with hot-reloading. The package also includes unit and integration tests, Jupyter notebooks for prototyping and evaluation, optional frontend web UI, and automated upgrade functionality.
This tool is designed for AI/ML engineers, Python developers, and teams building GenAI agents who want production-ready deployments on Google Cloud. It's particularly valuable for developers eager to build their first agent but unsure how to handle deployment, operations, evaluation, customization, and observability. The starter pack supports Python 3.10+ and integrates with Google Cloud SDK, making it ideal for organizations already using Google Cloud infrastructure.
Agent Starter Pack pricing
Pricing model: Freemium
Agent Starter Pack itself is open-source and free to use (Python package available via pip or uvx). There are no paid plans for the tool itself. However, deploying agents using the starter pack incurs Google Cloud service costs for Cloud Run, Agent Engine, Vertex AI, BigQuery, Cloud Logging, Cloud Trace, Artifact Registry, and other infrastructure. New Google Cloud customers get $300 in free credits to try Google Cloud products. Google Cloud Free Tier includes 20+ products with free monthly usage limits. The pricing for deployed agents depends on usage of Google Cloud services (CPU, memory, storage, API calls to Vertex AI).
Agent Starter Pack pros
- Pre-built agent templates (ReAct, RAG, multi-agent, Live Multimodal API)
- Go from idea to deployment in minutes instead of months
- Production-ready infrastructure with Terraform
- Built-in CI/CD with Google Cloud Build or GitHub Actions
- Integrated observability with OpenTelemetry GenAI instrumentation
- Vertex AI evaluation included
- Interactive playground with hot-reloading for local testing
- Complete project structure with backend, frontend, and tests
- Jupyter notebooks for prototyping and evaluation
- Automated upgrade command preserves customizations
- Supports Cloud Run and Agent Engine deployment targets
- RAG data ingestion configuration included
- Zero setup options available in Firebase Studio or Cloud Shell
- uvx single-command project creation without permanent install
- enhance command adds production deployment to existing agents
- Intelligent 3-way merge for upgrades auto-updates and preserves code
- Optional frontend web UI for interacting with agents
- GEMINI.md AI assistant context file included
- Python 3.10+ support with uv package management
- Terraform infrastructure-as-code for staging and production
Agent Starter Pack cons
- Requires Python 3.10+ (not compatible with older Python versions)
- Requires Google Cloud SDK installed and configured
- Requires Terraform for infrastructure deployment
- Open-source tool (not a managed service with Google support)
- Infrastructure costs apply for Cloud Run, Agent Engine, Vertex AI
- Learning curve for Terraform and CI/CD configuration
- Primarily designed for Google Cloud (not multi-cloud)
- Go support exists but Python is primary focus
- Manual CI/CD setup requires full control and Git provider knowledge
- Dev environment setup step optional but recommended adds complexity
- Service account and IAM permissions required for deployment
- Cloud Run stores container images in Artifact Registry with potential storage costs
- Large container images incur storage charges
- Not ideal for teams wanting pure no-code solution
- Requires Google Cloud project for deployment
Frequently asked questions about Agent Starter Pack
What is Agent Starter Pack?
Agent Starter Pack is a Python package that provides production-ready Generative AI Agent templates built for Google Cloud. It accelerates development by providing a holistic, production-ready solution addressing common challenges in building and deploying GenAI agents including Deployment & Operations, Evaluation, Customization, and Observability. The package gives you pre-built templates, infrastructure, CI/CD, observability, and security so you can focus on your agent logic.
How do I create a new agent project?
You can create a new agent project using either pip or uvx. With pip: create a virtual environment with python -m venv .venv, activate it with source .venv/bin/activate, install with pip install agent-starter-pack, then run agent-starter-pack create my-awesome-agent. With uvx: simply run uvx agent-starter-pack create which downloads and runs the latest version without permanent install. The create command lets you choose an agent template (adk_base, agentic_rag) and deployment target (cloud_run, agent_engine), then generates a complete project structure.
What agent templates are available?
Agent Starter Pack provides pre-built agent templates implementing common patterns including ReAct (Reasoning and Acting), RAG (Retrieval Augmented Generation), multi-agent systems, and Live Multimodal API. Specific templates include adk_base for base Agent Development Kit agents and langgraph_base_react for LangGraph-based ReAct agents. You can select templates during project creation using the -a flag.
Where can I deploy my agent?
You can deploy your agent to Cloud Run or Agent Engine on Google Cloud. Agent Engine is a new fully managed solution for deploying agents. The deployment target is selected during project creation using the -d flag (e.g., agent-starter-pack create my-agent -d agent_engine -a langgraph_base_react). Both options include production-ready infrastructure with monitoring, observability, and CI/CD.
How does CI/CD work with Agent Starter Pack?
Agent Starter Pack provides built-in CI/CD with two options: Google Cloud Build or GitHub Actions. Run agent-starter-pack setup-cicd from your project root to automate the entire process: it creates a GitHub repository, connects it to your chosen CI/CD provider, provisions infrastructure for staging and production environments using Terraform, and configures deployment triggers. After setup, commit and push code to trigger the pipeline which deploys to staging first.
What observability features are included?
OpenTelemetry GenAI instrumentation automatically captures telemetry data and exports it to Google Cloud services. Features include: BigQuery for querying telemetry data including token usage, model interactions, and performance metrics with external tables and linked datasets; Cloud Logging for GenAI operation logs and user feedback with 10-year retention in dedicated buckets; Cloud Trace for inspecting request flows and analyzing latencies; and optional visualization by connecting BigQuery data to BI tools for custom dashboards.
How do I run my agent locally?
After creating your project, navigate into it with cd my-awesome-agent and run make install && make playground. This installs dependencies and launches the interactive playground with hot-reloading. The specific UI playground depends on the agent template you selected. You can prototype using notebooks in the notebooks/ folder, edit app/agent.py to incorporate your logic, and test changes in the playground during your development loop.
Can I upgrade an existing project?
Yes, as agent-starter-pack evolves with new features, security fixes, and best practices, you can upgrade existing projects using the upgrade command. Run uvx agent-starter-pack upgrade --dry-run to preview changes, then uvx agent-starter-pack upgrade to apply. The upgrade uses intelligent 3-way merge: auto-updates scaffolding files you haven't modified, preserves your customizations when ASP hasn't changed those files, and prompts you to resolve conflicts when both have changed.
How do I add RAG data ingestion?
For Retrieval Augmented Generation (RAG) agents, you can configure data pipelines to process your information and load embeddings into Vertex AI Search or Vector Search. Use the agentic_rag template when creating your project. See the Data Ingestion Guide for detailed instructions on configuring RAG data ingestion for knowledge-based agents. This is one of the next steps after creating and running your agent locally.
What files are in the generated project structure?
The generated project includes: app/ (backend agent code with prompts, tools, business logic) or agent/ for Go projects, .cloudbuild/ (CI/CD for Google Cloud Build if selected), .github/ (CI/CD for GitHub Actions if selected), deployment/ (Terraform infrastructure-as-code files), tests/ (unit, integration, and load tests), notebooks/ (Jupyter notebooks for prototyping and evaluation), frontend/ (optional web UI for interacting with your agent), README.md (project-specific instructions), and GEMINI.md (AI assistant context file).