This is a project to demonstrate the capabilities of Google Deepmind's Gemini 3.1 Flash Live model, integrated into the Vision Agents SDK.
An agent takes the user through creating a product page for selling a used item, giving you a polished product image, a clear description and title, together with a price suggestion.
The user is guided through the entire process without interruption, which is all made possible by:
- the instruction following capabilities of the Gemini 3.1 Flash Live model
- Stream's low-latency edge infrastructure
- seamless function calling integration of the Gemini model
- the natural and low-latency dialogue functionality
The accompanying backend repository (required to run the sample) can be found here.
Get started today by setting up an account for Stream and creating your first Vision Agent application today!
- Node.js 20+
- npm
- A Stream app with valid API credentials
- The companion backend running locally or reachable over HTTP
- Install dependencies:
npm install- Create your local environment file from
./.env.example:
cp .env.example .env- Fill in the required values in
.env:
NEXT_PUBLIC_STREAM_API_KEY: Stream API key used by the frontendSTREAM_API_SECRET: Stream API secret used by related local tooling/backend flowsNEXT_PUBLIC_STREAM_USER_ID: Stream user id for the local demo userNEXT_PUBLIC_STREAM_USER_TOKEN: Stream user token for that userNEXT_PUBLIC_STREAM_CALL_ID: Call id the app should join
- Review the optional values if you need to customize local development:
NEXT_PUBLIC_BASE_URL: backend base URL. Defaults tohttp://localhost:8000NEXT_PUBLIC_STREAM_USER_NAME: display name for the local userNEXT_PUBLIC_STREAM_CALL_TYPE: Stream call type. Defaults todefaultNEXT_PUBLIC_STREAM_WORKFLOW_CHANNEL_TYPE: Stream chat channel type. Defaults tomessaging
- Start the companion backend so the frontend can create and delete agent sessions through
NEXT_PUBLIC_BASE_URL. - Start the Next.js dev server:
npm run dev- Open http://localhost:3000.
If the backend is not running, the UI may load but agent session creation will fail when the app calls the session endpoints.
npm run dev: start the local development servernpm run build: build the app for productionnpm run start: run the production buildnpm run lint: run ESLintnpm run test: run the Vitest test suite