agents-cli

Product Introduction

1. Definition: The Google Agents CLI (agents-cli) is a command-line interface tool and associated skill suite designed for the AI agent development lifecycle. Technically, it is a scaffolding, evaluation, and deployment automation framework built specifically for the Google Cloud AI agent ecosystem, including the Agent Development Kit (ADK) and Gemini Enterprise Agent Platform.
2. Core Value Proposition: It exists to bridge the gap between AI coding assistants (like Antigravity CLI, Claude Code, or Codex) and the production-ready deployment of AI agents on Google Cloud. Its primary value is enabling developers and coding agents to rapidly build, rigorously evaluate, and reliably ship enterprise-grade AI agents without deep, manual expertise in every underlying Google Cloud service and CLI tool.

Main Features

1. Agent Project Scaffolding: The agents-cli scaffold command generates a complete, production-ready agent project structure from a single command. It sets up the necessary directory layout, configuration files (like agent.yaml), dependency management (using uv), and foundational code adhering to ADK best practices. This eliminates manual boilerplate setup and ensures consistency across projects.
2. Integrated Evaluation Framework: The tool provides a robust, multi-stage evaluation pipeline via agents-cli eval. The generate phase runs your agent against a dataset of test cases, producing execution traces. The grade phase then analyzes these traces against configurable metrics (including LLM-as-a-judge). Advanced features include synthesize for creating multi-turn test scenarios, analyze for clustering failure modes, and optimize for auto-tuning prompts based on eval results, enabling data-driven agent improvement.
3. Multi-target Deployment Automation: The agents-cli deploy command abstracts the complexity of deploying AI agents to various Google Cloud runtimes. It supports deploying to fully-managed Agent Runtime, serverless Cloud Run, or Kubernetes-based GKE, handling infrastructure provisioning, containerization, and service configuration. This feature ensures a smooth path from local development to scalable, secure cloud hosting.
4. Coding Agent Skill Integration: Beyond the standalone CLI, agents-cli provides a suite of installable "skills" (via npx skills add). These skills embed deep knowledge of the Google Cloud agent stack—including ADK APIs, evaluation methodologies, and deployment patterns—directly into compatible coding agents. This transforms a general-purpose coding assistant into a specialized expert for building on Google's AI platform.

Problems Solved

1. Pain Point: The high complexity and fragmentation of tools required to build, test, and deploy production AI agents. Developers traditionally need to learn the ADK framework, evaluation tooling, cloud deployment CLIs (gcloud, Terraform), and CI/CD pipelines separately, leading to slow development cycles and inconsistency.
2. Target Audience: The primary personas are AI/ML Engineers and Full-stack Developers building conversational AI and autonomous agents; DevOps/SRE Engineers tasked with operationalizing agent deployments; and users of AI-powered coding assistants who want those assistants to efficiently build agents for Google Cloud.
3. Use Cases: Rapid prototyping of a new customer support agent with built-in evaluation; adding a Retrieval-Augmented Generation (RAG) pipeline and CI/CD to an existing experimental agent; migrating a locally-tested agent to a scalable, monitored production environment on Agent Runtime; using an AI coding assistant to generate correct ADK code patterns and deployment configurations.

Unique Advantages

1. Differentiation: Unlike generic AI frameworks (e.g., LangChain, LlamaIndex) or cloud CLIs, agents-cli is a vertically integrated toolchain specifically for the Google Cloud AI agent stack. It is not a competing coding agent but a force multiplier for them. It differs from using the ADK alone by providing the surrounding production-grade tooling (scaffolding, eval, deploy) that the framework does not.
2. Key Innovation: Its deep integration with AI coding agents via the "skills" system is a key innovation. It operationalizes the knowledge of building on Google Cloud's AI platform, allowing the coding agent to execute complex, correct workflows end-to-end. This shifts the paradigm from developers manually using a CLI to AI agents being directed to use the correct CLI commands and patterns autonomously.

Frequently Asked Questions (FAQ)

1. Do I need a coding agent like Claude Code to use Google Agents CLI? No, the agents-cli tool is a fully functional standalone CLI that you can run directly from your terminal for scaffolding, evaluation, and deployment. The coding agent skills are an optional layer that enhances productivity by allowing your AI assistant to use the CLI expertly on your behalf.
2. Can I use Agents CLI with an existing AI agent project I built with ADK? Yes. The agents-cli scaffold enhance command is designed specifically for this use case. It can add missing production components like deployment configurations, CI/CD pipelines, or RAG infrastructure to an existing ADK-based project, helping to mature it.
3. Is a Google Cloud account mandatory to start developing with agents-cli? For local development, testing, and evaluation, you do not need Google Cloud. You can use a Gemini API key from AI Studio to run ADK agents locally. A Google Cloud project is only required when you reach the stage of deploying your agent to Cloud Run, GKE, or Agent Runtime.
4. How does the evaluation feature work and is it automated? The evaluation is a two-phase process. First, eval generate runs your agent against a dataset, producing detailed trace files. Second, eval grade analyzes these traces using predefined metrics, which can include automated LLM-as-a-judge grading based on custom rubrics. This provides an automated, repeatable performance assessment.
5. What is the difference between deploying to Agent Runtime vs. Cloud Run with agents-cli? Agent Runtime is Google's fully managed, purpose-built service for hosting ADK agents with built-in grounding, safety, and monitoring. Cloud Run is a general-purpose serverless container platform. agents-cli deploy abstracts the configuration for both, allowing you to choose based on your need for managed agent features versus a more generic, customizable container environment.