agoraAI

Inspiration

Most online marketplaces still put too much work on the user. You search through listings, open dozens of tabs, compare details that are not always written clearly, message sellers one by one, wait for replies, and then try to decide which option is actually best. In practice, the hard part is often not only finding something. It is understanding tradeoffs and negotiating under real constraints.

That is where agoraAI started. We wanted to build something closer to how a good human assistant would behave: understand what the buyer wants, search the market, contact relevant sellers, negotiate when possible, compare the final options, and explain the recommendation clearly.

The idea evolved through several hackathons, but we kept building it beyond them. At the Mistral AI Worldwide Hackathon, we first worked on SpendTrack, which got us thinking about agents that act on behalf of users rather than just answer questions. At the Gemini Paris Hackathon, we built the first version of agoraAI as an agent-to-agent marketplace, with buyer and seller agents negotiating with each other. Later, at Paris Hack, we pushed the product closer to a real-world setting. In a housing use case, a buyer agent could search Greek property listings such as Spitogatos, turn the user’s intent into outreach, and help move from browsing listings to contacting a real person on the other side.

In parallel, we started looking more seriously at automated negotiation through ANAC/IJCAI, including the Human-Agent Negotiation setting. That helped clarify what makes the problem interesting. The difficult part is not generating polite messages. Agents need preferences, strategy, structured offers, negotiation state, acceptance decisions, and a way to explain why one outcome is better than another.

So agoraAI became more than a marketplace interface for us. It is our attempt to build an agentic transaction layer: a place where buyer and seller agents can discover each other, negotiate, and help users reach better deals without hiding the process. Hackathons helped us move fast and test the idea, but we are continuing to build agoraAI independently as a real product.

Built With

• aiosqlite
• autonomous-negotiation
• beautiful-soup
• browser-automation
• docker
• fastapi
• github
• google-gemini
• httpx
• imap
• multi-agent-systems
• multi-agent-workflows
• next.js
• ngrok
• offer-scoring
• openai-compatible-apis
• openrouter
• playwright
• pydantic
• python
• react
• resend
• rest-apis
• server-sent-events
• smtp
• spitogatos-scraper
• sqlalchemy
• sqlite
• sse-starlette
• structured-llm-outputs
• svix
• tailwind-css
• typescript
• uvicorn