Gemini Enterprise Agent Platform, powering the next wave of agents

双子座企业代理平台，为下一代代理提供动力

https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform

---

Google推出Gemini Enterprise Agent Platform，为企业提供下一代AI代理开发平台。该平台整合了Vertex AI的模型选择和构建能力，新增代理集成、DevOps编排和安全功能，支持200多个领先模型（包括Gemini 3.1系列和开源模型Gemma 4）。平台特点包括：低代码开发环境Agent Studio、支持多日工作流的Agent Runtime、安全沙箱环境以及实时异常检测功能。企业可通过该平台快速构建、部署和管理生产级AI代理，实现从概念验证到实际应用的平滑过渡，同时确保安全性和治理合规。该平台现已上线Google Cloud控制台。

---

In the early days of generative AI, building safe and reliable business tools took massive engineering effort and a high tolerance for trial and error. We helped solve that with Vertex AI, our trusted AI development platform. But today, we’re managing a different level of complexity, with agents interacting across multiple systems — and often without security and governance guardrails. 

To move toward a truly autonomous enterprise, one where agents can act with the same independence and reliability as a member of your team, you need a foundation that can sustain that level of trust. 

What’s new: Today, we’re launching Gemini Enterprise Agent Platform — our new, comprehensive platform to build, scale, govern, and optimize agents. It’s the evolution of Vertex AI, bringing the model selection, model building, and agent building capabilities that customers love, together with new features for agent integration, DevOps, orchestration, and security. 

Agent Platform provides a single destination for your technical teams to build agents that can transform your products, services, and operations. These agents can be seamlessly delivered to your employees through the Gemini Enterprise app, all while remaining tightly integrated with your IT operations to help ensure control, governance, and security as you scale.

The platform also provides first-class access to more than 200 of the world’s leading models through Model Garden. This includes our latest first-party breakthroughs like Gemini 3.1 Pro, Gemini 3.1 Flash Image, and Lyria 3, alongside our open models like Gemma 4. And, of course, customers have full flexibility to use the best model for the job with support for third-party models like Anthropic’s Claude Opus, Sonnet and Haiku. 

Moving forward, all Vertex AI services and roadmap evolutions will be delivered exclusively through the Agent Platform, rather than as a standalone service, to power the next generation of agent development.

Why Agent Platform matters for your business: Agent Platform helps you move from managing individual AI tasks to delegating business outcomes with total confidence. You can: 

• Build: Choose the right environment for the job — from the low-code, visual interface of the new Agent Studio, to the code-first logic of the upgraded Agent Development Kit (ADK). We’ve simplified the entire lifecycle with AI-native coding capabilities to help you ship production-grade agents faster.

• Scale: Clear the path to production with the re-engineered Agent Runtime. This supports long-running agents that maintain state for days at a time and are backed by Memory Bank for persistent, long-term context.

• Govern: Establish centralized control with Agent Identity, Agent Registry, and Agent Gateway. These capabilities help ensure every agent — whether built on Agent Platform or sourced from our partner ecosystem — has a trackable identity and operates within enterprise-grade guardrails. 

• Optimize: Guarantee quality with Agent Simulation, Agent Evaluation, and Agent Observability. These tools provide full execution traces and a real-time lens into agent reasoning to help ensure your agents always hit their goals.

Get started with Agent Platform: Visit Agent Platform in the Google Cloud console to explore new features and start building today. 

Keep reading for a deeper look at our latest releases and how Agent Platform helps you deliver the production-ready agents you can trust at every stage of the journey.

Build AI agents

Build agents quickly and easily by empowering your developers, business users and everyone in between to build and deploy agents at scale.

Build smarter agents, faster

• A major upgrade to ADK: More than six trillion tokens are processed monthly on Gemini models through ADK. Unlock more powerful reasoning by organizing agents into a network of sub-agents. This new, graph-based framework allows you to define clear, reliable logic for how agents work together to solve complex problems.

• Workspaces are secure-by-design: Give agents a hardened, sandboxed environment to run bash commands and manage files safely, isolated from your core systems.

• Multimodal streaming: Bring human-like stability to real-time interactions with multimodal support for live audio and video cues.

Connect your agents to the enterprise

• Securely access any system: Use plug-and-play architecture with Native Ecosystem Integrations to connect agents to your internal data and tools without custom coding.

• Automate background operations: Activate your data in BigQuery and Pub/Sub with Batch & Event-driven agents. This way, you can run massive, asynchronous tasks like content evaluation or data analysis in the background.

Go from idea to production in hours

• Enable AI-driven development: A programmatic interface for coding agents to access Google’s complete suite of agentic capabilities, allowing them to build, evaluate, and deploy production-ready agents on your behalf. 

• Bringing agent building directly to Agent Studio: Now, you can move seamlessly from building simple prompts to deploying complex agents in Agent Studio. Once you're ready for deep customization, export your logic directly into ADK to continue development in a full-code environment.

• Get a head start with pre-built agents: Access a curated set of agent templates in Agent Garden — including code modernization, financial analysis, economic research, invoice processing, and more — that serve as immediate building blocks for your multi-agent systems.

Scale AI agents

To move from a proof-of-concept to a live environment, you need a platform that can handle the performance, state, and security requirements of real-world work. 

Powering high-performance agent execution

• The latest Agent Runtime: Our revamped Agent Runtime delivers sub-second cold starts and allows you to provision new agents in seconds.

• Support for multi-day workflows: You can now deploy long-running agents that run autonomously for days at a time. This allows your agents to manage complex, multi-step workflows and deep reasoning tasks that require extended persistence, like managing a sales prospecting sequence. 

• Autonomous action with security-by-design environments:Agent Sandbox provides a hardened environment to safely execute model-generated code and perform computer use tasks like browser-based automation without risk to your host systems.

• Agent-to-agent orchestration: Enables agents to seamlessly delegate tasks to one another, including support for complex, generative, and deterministic orchestration patterns. This ensures that for critical flows such as compliance, your agents follow well-specified paths every time.

Move beyond temporary session data to high-accuracy context

• Personalize interactions: Agent Memory Bank dynamically generates and curates long-term memories from conversations. Using new Memory Profiles, agents can recall high-accuracy details with low latency, ensuring context is never lost.
• Link AI interactions to your existing records: Store and manage history using Agent Sessions. With Custom Session IDs, you can use your own unique identifiers to track sessions and map them directly to your internal database and CRM records.

• Enable real-time, human-like interactions: Using the WebSocket protocol for Bidirectional Streaming, you can help ensure your agents are highly responsive during live customer or employee interactions, processing audio and video without lag.

Govern AI agents

Govern with a secure-by-design architecture that applies enterprise rigor to every agent in your fleet – from the ones you build on Agent Platform to the ones you source from our partner ecosystem.

Manage all of your agents through a single source of truth for identity and access.

• Assign every agent a verifiable identity: Agent Identity improves the security posture of your agents by ensuring every agent receives a unique cryptographic ID. This creates a clear, auditable trail for every action an agent takes, mapped back to defined authorization policies.

• Maintain a central library of approved tools: Our new Agent Registry provides a single source of truth for your enterprise. It indexes every internal agent, tool, and skill, simplifying discovery and ensuring only governed, approved assets are available to your users.

• Manage your agent fleet from one control point: Agent Gateway acts as the air traffic control for your agent ecosystem. It provides secure, unified connectivity between agents and tools across any environment, while enforcing consistent security policies and Model Armor protections to safeguard against prompt injection and data leakage.

Use AI-powered insights to detect hidden risks and suspicious behavior before they impact your business.

• Detect suspicious behavior in real-time: Agent Anomaly Detection uses statistical models and an LLM-as-a-judge framework to flag unusual reasoning. This works alongside Agent Threat Detection to provide visibility into malicious activity, such as reverse shells or connections to known bad IP addresses.
• Uncover vulnerabilities automatically: A new Agent Security dashboard, powered by Security Command Center, unifies threat detection and risk analysis. It allows your teams to map relationships between agents and models, automate asset discovery, and scan for vulnerabilities in the underlying operating system and language packages.

Optimize AI agents 

Agent Platform gives you the visibility needed to understand how your AI is performing, making it easy to refine their logic and get smarter over time.

Test your agents before they ship

• Simulate realistic conversations: Use Agent Simulation to test agents against human-like synthetic user interactions and virtualized tools in a controlled environment. Agents are automatically scored based on task success and safety across multi-step conversations.

Monitor and improve in production

• Track live performance: Use Agent Evaluation to continuously score agents against live traffic using multi-turn autoraters that can evaluate the logic of an entire conversation, not just a single response. With turnkey dashboards and Agent Observability, you can visually trace complex reasoning to debug issues as they happen.
• Automate agent refinement: Instead of manually digging through logs, Agent Optimizer automatically clusters real-world failures and suggests refined system instructions to improve accuracy.

Detailed technical guides and a full list of updates are available in our updated documentation and release notes. Agent Platform is the new standard for enterprise agent development, built to help you move from experimentation to production-scale impact, starting today.