Build a Practical AI Voice Agent with Fish Audio

Want to create a helpful voice-powered assistant without heavy costs or complex infrastructure? This walkthrough shows how to build a real-world AI voice agent using Fish Audio’s text-to-speech API—perfect for scenarios like customer service or specialized support.

The example focuses on an animal shelter assistant. Users see clear cards for common needs: rescue process, donations, dropping off food and supplies, or surrendering a stray. Clicking a card instantly plays a natural pre-recorded voice response. For anything else, users type a free-form question; an LLM generates the answer and Fish Audio converts it into speech on the fly.

The stack keeps things efficient and affordable: Vercel’s AI SDK for the clean conversational interface, free models via OpenRouter (such as Poolside Laguna), and Fish Audio’s free S2.1 Pro model for high-quality TTS. After signing up at Fish Audio, you generate an API key in seconds. A detailed prompt that includes the relevant documentation helps an AI coding assistant assemble the full app—complete with mock data, automated tests, and a persistent audio player.

The result is a polished landing page where pre-filled responses stay cheap and fast, while open questions still feel intelligent and spoken. You control token usage by limiting full voice conversations, making the solution practical for real deployments.