Deploy and Host a WhatsApp Voice Agent (Kapso) on Railway

Answer WhatsApp calls with your own AI front desk. Riley, the sample agent, says it is an AI, answers questions about your business, and books, moves or cancels appointments after a clear yes. Your number and messages stay in Kapso. This server carries the call audio and runs the tools; Grok Voice listens, thinks and speaks.

About Hosting the WhatsApp Voice Agent

You fill in three values: your xAI API key, your Kapso project API key and the Meta phone number ID of your Kapso number. Railway generates the server's own secrets.

On its first start, the server registers its Railway URL as your number's webhook in Kapso, if the number has none, and never replaces an existing one. Then call the number from WhatsApp, and Riley answers. Open https:///operator/ and sign in with the OPERATOR_TOKEN variable to see recent calls, the setup and an event log.

Before you deploy, in the Kapso dashboard:

Enable Calling on a dedicated number.
Set the number's Incoming call answering to Agent or integration (no Inbox fallback), and do not assign a dashboard voice agent to it.

Test this server with a WhatsApp call. Browser test calls from the console need the server on your computer or on a server with a public IP.

Common Use Cases

An AI receptionist that answers WhatsApp calls and books appointments
A starting point for building calling on Kapso with your own coding agent
A test bench for voice agent prompts and tools on real WhatsApp calls

Dependencies for WhatsApp Voice Agent Hosting

A Kapso project with a WhatsApp number that has Calling enabled
An xAI API key for Grok Voice

Deployment Dependencies

Source: gokapso/whatsapp-voice-agent-starter
Setup, customization and security: README.md, docs/deploy.md, docs/customize.md

Implementation Details

One service, built from the repository's Dockerfile: Python, Pipecat and WebRTC. It keeps a 1 GB volume at /app/data for the map of which caller owns which booking.
Call audio leaves the server over UDP and reaches Meta's call servers directly. A public STUN server is set in ICE_SERVERS_JSON; no TURN relay is needed.
CALENDAR=local is a sample calendar for a first test. For real bookings, set CAL_API_KEY and CALENDAR=calcom, and describe your business in agent/business.json.
To change the agent, eject the template into your own repository and edit agent/prompt.md, agent/agent.toml and agent/business.json.
Each call is a Grok Voice session that xAI bills per minute. WhatsApp calling and Railway usage are billed separately.
One server answers one number. Call state is in memory, so keep one replica.

Why Deploy the WhatsApp Voice Agent on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying the WhatsApp Voice Agent on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.