gemini-interactions-api

Build stateful chatbots with Gemini models by using a unified Interactions API that maintains server-side conversation history via the previous_interaction_id parameter. This allows your AI agents to retain context across multiple user turns without managing history client-side.

Implement streaming responses for Gemini AI agents using Server-Sent Events (SSE) through the Interactions API. SSE enables real-time token streaming directly to your application, providing immediate feedback to users during long generations.

The Gemini Interactions API supports tool orchestration including function calling, Google Search, code execution, URL context, file search, and remote MCP. This enables AI agents to autonomously select and execute tools to complete complex automation pipelines.

Gemini interactions support multimodal understanding and generation alongside end-to-end workflows with asynchronous background tasks. This allows your applications to process images and text while running long-running research or automation jobs without blocking the main thread.

You need either Python or TypeScript SDKs to instantiate a Gemini Interactions client and start an interaction. You must also provide valid model or agent IDs to route your requests correctly to the underlying Gemini architecture.

Access built-in agents like Deep Research through the unified Gemini Interactions API by specifying the appropriate agent ID when instantiating your client. This provides immediate entry to advanced research and automation capabilities without custom agent configuration.