Deploy and Host Vercel EVE Agent Starter on Railway

Vercel EVE Agent Starter is a production-ready template for building conversational AI agents with Vercel Eve. It ships with a working example — a support agent that answers from a FAQ knowledge base and escalates to a ticket when it can't find an answer — so you can see it running before building your own.

About Hosting Vercel EVE Agent Starter

Hosting this template on Railway means running a standard Node.js 24 service built with Railpack from a pnpm-managed repository — no Dockerfile needed. The build compiles the agent with eve build and serves it with eve start, listening on Railway's injected PORT. Because the container has no Docker-in-Docker or KVM access, the template ships just-bash so Eve's execution sandbox can initialize on a plain Linux host. The OpenAI model, reasoning effort, and HTTP Basic credentials that gate /eve/v1/session in production are read from environment variables, with /eve/v1/health left public for Railway's healthcheck.

Common Use Cases

• Customer support chatbot that answers FAQs and escalates unresolved questions to a ticket
• Internal helpdesk or knowledge-base assistant for a team or product
• Starting point for any conversational agent built on Vercel Eve — swap the FAQ example for your own domain
• Proof-of-concept environment for evaluating the Vercel Eve framework

Dependencies for Vercel EVE Agent Starter Hosting

• An OpenAI API key
• Node.js 24+ and pnpm (already pinned via packageManager and RAILPACK_NODE_VERSION)

Deployment Dependencies

• Vercel Eve Documentation
• OpenAI API Documentation
• Railway Railpack Builder

Implementation Details

Configuration is centralized and typed in src/config/ — agent/agent.ts and agent/channels/eve.ts read from it instead of touching process.env directly, so every setting is validated once at boot:

// agent/agent.ts
export default defineAgent({
  model: openai(config.openai.model),
  ...(config.openai.reasoningEffort ? { reasoning: config.openai.reasoningEffort } : {}),
});

The knowledge base sits behind a small repository interface (FaqRepository), so swapping the bundled JSON file for a real database or CMS later only means writing a new implementation and pointing getRepository() at it — the agent's tools don't change.

Getting Started After Deploy

1. Generate a public domain for the service from the Railway dashboard (or railway domain).

2. Check GET /eve/v1/health — it's public and returns {"ok":true,"status":"ready"} once the service is up.

3. ROUTE_AUTH_BASIC_PASSWORD is generated with Railway's secret() function and regenerates on every redeploy — read the current value from the running service with railway ssh -- printenv ROUTE_AUTH_BASIC_PASSWORD rather than hardcoding it anywhere.

4. Send a test message with HTTP Basic auth:

   curl -X POST https:///eve/v1/session \
     -u "eve-agent:" \
     -H "Content-Type: application/json" \
     -d '{"message":"how do I update my payment method?"}'
   

5. Replace agent/instructions.md, the FAQ data in knowledge-base/faqs.json, and the tools in agent/tools/ with your own — the FAQ support agent is a working example, not the end state.

Why Deploy Vercel EVE Agent Starter 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 Vercel EVE Agent Starter 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.