Template name
Talk to an eve agent in the browser, with voice replies, tool execution, approvals, and durable conversations.
Build a browser-based voice assistant that can run tools and ask for approval while you talk. This Next.js template connects OpenAI's GPT-Live-1 through Edgenivo AI Gateway to an eve agent, with a live transcript and tool activity on screen.
The voice model handles listening, speaking, and turn-taking. It passes requests to eve, which runs tools and keeps the call's conversation state. Progress appears on screen, and the voice model speaks the agent's answer.
The included weather and reminder tools demonstrate the flow with sample data; they do not fetch live weather or schedule real reminders. Replace them with tools for your own services.
When a tool needs approval, say an option such as "approve" or select it on screen. Free-text questions and connection sign-ins use the on-screen interface. Hanging up cancels the active agent turn.
Each call starts a new eve session. Calls are not saved for later, and reloading ends the conversation.
Set AI_GATEWAY_API_KEY and a strong EVE_VOICE_PASSWORD during deployment, then open the app, enter the password, and allow microphone access. Everyone with the password shares one agent identity.
Voice sessions are billed per second, including silence and time waiting for tools; agent model calls are billed separately. The browser ends calls after 10 minutes, but this is not a server-enforced spending limit.
Edit the agent's instructions and tools to change what it can do. See the repository's setup guide for local development, voice configuration, and multi-user authentication.
