In the rapidly evolving world of AI, developers constantly seek more robust and flexible tools to build sophisticated applications. Today, we’re thrilled to announce a significant expansion of capabilities for Managed Agents within the Gemini API. These updates are a direct response to your feedback, designed to empower you to create truly reliable and production-ready AI agents.
The Gemini Interactions API streamlines agent development by handling complex backend operations for you. With just a single endpoint call, Gemini takes care of reasoning, code execution, package management, file operations, and web information retrieval, all within a secure, isolated cloud sandbox. This allows you to focus purely on your agent’s core logic and functionality.
Powering Production-Ready AI Agents with Gemini
Managed Agents are transforming how developers integrate AI into their applications, offering a robust foundation for building intelligent systems. These new features – background execution, remote MCP server integration, custom function calling, and seamless credential refreshing – are crucial for creating powerful and resilient agents. They ensure your AI can operate efficiently and securely in real-world development environments, directly addressing key developer needs.
For those eager to dive in, installing the Interactions API skill is straightforward: use npx skills add google-gemini/gemini-skills --skill gemini-interactions-api. Our examples typically leverage the @google/genai JavaScript SDK, but comprehensive documentation for Python and cURL is readily available in the Antigravity agent documentation.
Unleash Asynchronous Workflows and Remote Connectivity
Long-running tasks can often be a bottleneck, making applications fragile when HTTP connections must be held open. We’re solving this with background execution, allowing you to pass background: true to run interactions asynchronously on the server. The API immediately returns an ID, which your client applications can use to poll for status, stream progress, or simply reconnect later while the agent completes its work remotely.
This capability transforms Managed Agents into powerful, asynchronous workers that won’t block your application’s flow, significantly improving user experience and system reliability. To learn more about implementing this, delve into our detailed background execution guide.
Accessing private databases or internal APIs has traditionally required developers to build custom proxy middleware, adding layers of complexity. Now, Managed Agents can connect directly to your remote Model Context Protocol (MCP) servers. This powerful integration eliminates the need for bespoke proxy solutions, simplifying your architecture considerably.
You can seamlessly mix and match these remote tools with Gemini’s built-in sandbox capabilities. By passing an mcp_server tool at interaction time, alongside features like Google Search or code execution, your agent can securely communicate with your private endpoints directly from its isolated sandbox. Remember to follow best practices to maintain security as you extend your agent with external tools and APIs.
Elevate Agent Capabilities with Custom Logic and Seamless Credentials
While Gemini’s sandbox provides a rich set of built-in tools, your applications often require custom business logic executed locally. With custom function calling, you can define and integrate your own tools alongside the sandbox’s native capabilities. The API intelligently uses step matching, ensuring built-in tools run automatically on the server, while custom functions transition the interaction to requires_action.
This seamless hand-off means your client application can then execute its specific local business logic, giving you ultimate control and flexibility over your agent’s behavior. This hybrid approach allows you to leverage the best of both worlds: Gemini’s powerful server-side capabilities and your application’s unique functionalities.
Managing expiring access tokens and short-lived API keys is a common challenge for developers, often leading to interrupted workflows. Our new credential refresh capability addresses this head-on, allowing you to effortlessly update your agent’s network configuration. Simply pass your existing environment_id with a new network configuration on your next interaction.
The updated rules immediately replace the old ones, ensuring your agent always operates with valid credentials. Crucially, your sandbox maintains its complete state, including its filesystem, installed packages, and cloned repositories, eliminating the need for tedious re-setup after a refresh. This significantly enhances the operational stability and ease of managing your AI agents.
These latest advancements collectively transform Managed Agents into even more versatile and robust workers, ready for the demands of production environments. They free your applications from blocking operations and simplify complex integrations, allowing you to innovate faster and more reliably.
We encourage you to explore these exciting new features further to unlock their full potential. Dive into the Gemini Interactions API overview and the managed agents quickstart to learn more about defining custom agents, configuring environments, setting network rules, and mastering advanced streaming patterns.
Source: Google Blog (The Keyword)
