AgentEats
A small team with a track record of shipping useful tools
YouTube Video
Project Description
AgentEats is an AI-powered phone-ordering system that answers restaurant calls 24/7, takes orders through a natural voice conversation, and sends the final order directly to the kitchen or POS with zero staff involvement and zero commission fees. It solves the real problem of missed phone orders, overworked staff, and expensive delivery-app margins.
End-to-end: A customer calls a restaurant’s phone number → our agent engages in a fully natural conversation → retrieves the real menu via live API → collects items, modifiers, and customer details → confirms the order → writes it into the kitchen system or POS → logs the interaction. The demo shows the complete flow live from a browser-based phone interface using ElevenLabs voice, OpenAI reasoning, and real menu integrations.
Judging Criteria
- Working Prototype
AgentEats is fully functional: users can place real orders via a live voice call in the browser. The agent handles multi-turn dialog, menu lookups, item confirmation, error recovery, and order submission to a real Postgres-backed kitchen system.
- Technical Complexity & Integration
The project integrates multiple systems in real time:
- Next.js frontend with a live phone-call simulation UI
- Clerk for authentication
- ElevenLabs for low-latency voice input/output
- Prisma + Postgres for menu, order, and restaurant data models
- Railway for hosting, deployments, and database infrastructure
The hardest part: robust multi-turn conversational order-taking that stays grounded in the restaurant’s real menu, supports modifiers, handles mistakes, and maintains accuracy without hallucinations.
- Innovation & Creativity
AgentEats replaces a traditionally manual, error-prone workflow with a fully autonomous AI phone agent. Instead of another chatbot or web-ordering widget, it uses voice-first natural conversation, grounded menu retrieval, and autonomous ordering logic to behave like a trained staff member that never sleeps. It’s designed to be drop-in simple for any small restaurant, so no training, no configuration, no extra hardware.
- Real-World Impact
Restaurants lose thousands annually to missed calls and 20–30% commission fees. This solves both problems instantly by:
- Capturing every order
- Operating 24/7
- Eliminating commission fees
- Reducing staff load
- Increasing revenue without hiring
- Theme Alignment (Agentic Workflows)
- AgentEats is built around a fully autonomous agent that handles an entire workflow from start to finish:
- Receives an incoming call
- Conducts the ordering conversation
- Performs menu lookups
- Executes confirmations and corrections
- Finalizes and writes the order to the kitchen
- All without human supervision or manual triggers
It’s a complete agentic loop that continuously acts, reasons, and executes tasks until the order is delivered to the kitchen.
Technologies Used
Frontend & UX
- Next.js
- React
- Clerk (auth)
- Elevenlabs ui kit
- shadcn, tailwind
- figma
AI & Voice
- OpenAI Assistants (reasoning + structured order output)
- ElevenLabs (bidirectional voice)
Backend & Database
- Prisma ORM
- Postgres
- Railway (hosting + DB)
Setup / Demo Steps (Reproducible)
To see call workflow
- Go to https://agenteats.online and tap “Start Voice Call.”
- Speak a real order e.g., “Hi, can I get a prawn fried rice”
- Agent responds via ElevenLabs voice and walks through the entire ordering flow.
To see back of house admin management
- Visit the live demo URL https://agenteats.online
- Sign in via Clerk (demo mode).
Prior Work
We started with a railway/next starter repo