Gemini Interactions API: Stateful Agent and Image Editing Demo in Anti-gravity

Gemini Interactions API is introduced as a unified and stateful interface for Gemini models and agents, covering synchronous calls, managed agents, background tools, deep research, and media generation. The demonstration installs the Gemini Interactions API skill in the Anti-gravity IDE and builds a small image-editing application that accepts a portrait plus text or voice input. It then walks through iterative edits that reuse previous interaction IDs.

■ Unified API and agent features
- One stateful Interactions API for Gemini models and agents
- Supports synchronous calls, managed agents, background tools, deep research, and media generation
■ Setup with Anti-gravity IDE
- Installs the Gemini Interactions API skill and configures a Gemini API key from ai.dev
- Uses npm and ServerJS with Gemini 3.1 Flash Image and Gemini 3.5 Flash
■ Stateful image-editing demo
- Builds an app for portrait input, text or voice prompts, and image output
- Transcribes audio first because Nano Banana does not accept audio input
- Reuses previous interaction IDs to make iterative edits without re-uploading the image

Viewers interested in Gemini API development, agent-based workflows, and stateful application patterns can see how to start with the Interactions API skill and a coding agent. The demo offers an overview of setup, capabilities, and an iterative image-editing use case rather than a full production guide.