Speed up your development with vibe coding: Build an AI Agent using Google Interactions API

Google interactions api overview

What is Interactions API for Agentic Applications

The Google Interactions API is designed to enable dynamic, agentic applications that communicate with Google’s AI models in an interactive and stateful manner. This API allows developers to build intelligent agents that can handle complex conversations with multiple turns, making it ideal for research, planning, and execution tasks.

By leveraging this API, your agent can interpret and respond to user inputs efficiently while maintaining context across interactions. This sets the foundation for creating powerful AI systems like research planners and executors that automate workflows.

Stateful Interactions with previousInteractionId

A key feature of the Interactions API is stateful conversation management. This is enabled through the use of the previousInteractionId parameter, which links each new API call to a prior interaction.

Maintaining state allows your agent to remember earlier conversation elements, ensuring continuity across phases of a multi-step process. For example, when planning research tasks and executing them, referencing the previous interaction ID lets the system build upon prior context without losing track.

This design enables seamless, layered agent workflows where each step logically follows from the last, greatly enhancing user experience and operational accuracy.

Gemini 3 Pro, flash & deep research agent integration

Combine Gemini 3 Pro Gemini 3 Flash Deep Research Agent

The integration of Gemini 3 Pro, Gemini 3 Flash, and the Deep Research Agent forms a robust AI research ecosystem. Each model plays a specialized role:

• Gemini 3 Flash excels at quick, responsive planning tasks, generating research plans based on user goals.
• Deep Research Agent autonomously conducts extensive web research, gathering detailed information.
• Gemini 3 Pro synthesizes findings into comprehensive executive reports.

By combining these agents using the Interactions API, you create a streamlined pipeline from initial planning through deep research and final report generation. This collaboration empowers your AI system to efficiently accomplish complex research projects.

Supported models in interactions api

The Interactions API currently supports several key agent models that enable effective research workflows:

• gemini-3-flash-preview – for fast and flexible planning tasks
• gemini-3-pro-preview – designed for high-level executive summarization and synthesis
• deep-research-pro-preview-12-2025 – specialist agent for autonomous deep web research

Using these models within the Interactions API framework ensures access to cutting-edge AI capabilities tailored for different stages of research and execution.

ai research agent 3-phase workflow

phase 1: research planner with gemini 3 flash

The first phase of the AI research workflow focuses on creating a clear and structured research plan. Using the gemini-3-flash-preview model, the agent generates a numbered list of research tasks tailored to the user’s goals.

This planning stage is crucial as it sets the direction for the entire research process. By combining Gemini 3 Flash with the GoogleSearch tool, the agent can quickly outline actionable steps in response to a research request, ensuring efficiency and relevance.

phase 2: research executor deep research agent

Once the plan is established, the Deep Research Agent takes over in phase two. This model, accessed via the deep-research-pro-preview-12-2025 agent with background execution enabled, autonomously explores the web to gather extensive information based on the research tasks.

Running as a background process, this agent efficiently scours online sources, compiling data to support comprehensive understanding. The autonomous nature of this phase allows for deep and persistent research without the need for constant user input.

phase 3: synthesis with gemini 3 pro executive report

In the final phase, Gemini 3 Pro synthesizes the collected research into an organized executive report. This step distills the gathered insights into concise summaries and actionable recommendations.

The executive report enables clear communication of research findings, making it easier for users to extract value and make informed decisions. This synthesis phase merges depth of information with readability.

get gemini api key & setup

how to get gemini api key for interactions api

To start using the Interactions API with Gemini models, you’ll need to obtain an API key from Google AI Studio. This key authorizes your application to access the generative AI features offered by Google.

Generating the Gemini API key involves creating credentials in your Google Cloud project linked to AI Studio, which you then integrate into your app to authenticate requests through the Google GenAI client.

googlesearch tool integration

The GoogleSearch tool is a vital component that enhances the AI agent’s research capabilities by enabling real-time search queries within the agent workflow.

Integrating this tool allows the Gemini models, especially Gemini 3 Flash and the Deep Research Agent, to fetch up-to-date information from the web. This integration improves the accuracy and freshness of generated research tasks and findings.

Proper setup involves connecting the GoogleSearch API credentials alongside the Gemini API key, ensuring seamless communication between search operations and AI reasoning.

step-by-step tutorial: build research planner executor

complete open-source code implementation

This tutorial provides a full open-source codebase to help you build your AI research planner and executor agent using the Google Interactions API. The implementation covers all phases of the workflow—from task generation with Gemini 3 Flash, through autonomous research using the Deep Research Agent, to final report synthesis via Gemini 3 Pro.

Following the code examples step-by-step allows you to customize and extend the agent’s capabilities as needed. The modular structure ensures clear separation of planning, execution, and synthesis functions, making your project maintainable and scalable.

background execution & progress tracking

To handle lengthy research jobs effectively, the Deep Research Agent runs in the background using the background=true parameter. This approach frees up resources while the agent performs continuous, autonomous browsing and data gathering.

Progress tracking is built into the system, enabling you to monitor execution status and results in real time. This feature improves transparency and control over the research lifecycle without interrupting the workflow.

interactive task selection features

The user experience is enhanced with interactive task selection, allowing you to pick which research tasks to execute or prioritize. This interactivity offers flexible control over the research process, letting you adapt dynamically as new insights emerge.

You can implement interfaces that present available tasks, accept user input for selection, and update the agent’s focus accordingly. This empowers customized research flows and better resource management.

deploy your ai research agent

streamlit app setup & running

Deployment is streamlined with a sample Streamlit app included in the open-source repository. This user-friendly interface enables easy launching and controlling of the AI research planner and executor.

Setting up the Streamlit app requires installing dependencies, configuring API keys, and running the app locally or on a server. The app provides interactive components to input research goals, track progress, and view outputs seamlessly.

export executive reports & infographics

After completion of research workflows, the system lets you export executive reports generated by Gemini 3 Pro, including visual infographics summarizing the key findings.

These exports facilitate sharing insights in accessible formats suitable for presentations or decision-making. The ability to generate polished reports enhances the professional utility of your AI research agent.

faq

Ready to harness the power of vibecoding with this cutting-edge AI research planner and executor? Join the community of enthusiasts and developers pushing the boundaries of dynamic AI agents. Connect, share, and grow your skills by joining the Vibecoding Hub today.