AI & Agentic Systems
Voyagent — AI Multi-Agent Travel Planner
Four AI agents negotiate your trip, and you watch them do it.
- agents
- 4agents
- negotiation rounds max
- 2negotiation rounds max
- days per trip
- 5days per trip
- Docker services
- 3Docker services
// at a glance
The problem and the idea
The problem
Planning a trip means balancing interests, a budget and a realistic daily schedule at once. Single-prompt AI planners often ignore the budget or pack a day with places that are too far apart.
The idea
Give each concern its own agent. Interest proposes places, Budget and Logistics object when the plan breaks the budget or the schedule, and Planner turns the approved plan into a timed route. The whole negotiation streams live to the user, and the result is drawn on an interactive map.
// who it's for
Who it's for
- 01
Travelers planning a short city trip on a fixed budget
- 02
Groups that need a plan that fits everyone's interests and wallets
- 03
Anyone curious how multi-agent systems resolve conflicts
// how it works
How it works
1/5 Trip request
Swipe sideways to see the whole diagram →// key features
Key features
- Live agent chat with proposals, objections and approvals as they happen
- Each day's route on the map in its own colour, with a route animation
- Budget simulator with an A/B comparison of the current and a cheaper plan
- Drag-and-drop itinerary editing with automatic re-timing
- Read-only share links
- EN / AZ interface; agents reply in the selected language
// design decisions
Design decisions
- 01
Cost and distance checks run in code; the LLM only writes proposals and short messages, which keeps token cost low.
- 02
If OpenAI fails, the system switches to OpenRouter and says so openly in the agent chat.
- 03
The budget simulator only trims approved stops and never invents new places.
// built with
Built with
- FastAPI
- React
- PostgreSQL
- SSE
- Leaflet
- OpenStreetMap
- Docker
Run it locally
$ docker compose up --build