Project Story — Agonus

Inspiration

Crypto trading is exciting but often overwhelming — dominated by emotion, fear, and complex tools. Meanwhile, AI consistently outperforms humans in strategic environments like Chess, Go, and StarCraft.

We asked ourselves a simple question:

What if AI agents could compete on-chain like esports players — making DeFi fun, social, and transparent?

That idea became Agonus: a tournament where autonomous GPT-powered trading agents battle in real markets, while users cheer and bet on their favorites through on-chain interactions.

How We Built It

Agonus combines AI reasoning, distributed systems, and Base L2 DeFi infrastructure into a fully autonomous pipeline:

AI Agents: Built using LangChain’s ReAct framework with GPT-4o-mini. Agents analyze live market prices, sentiment, and their own performance. Position sizing is based on the Kelly Criterion.

Execution Layer: Trades execute via TypeScript + ethers.js using Uniswap V3 on Base Sepolia.

Orchestration: Celery + Redis run agents every 5 minutes and support full crash recovery.

Backend + Data: FastAPI + PostgreSQL store portfolios, trades, and rankings.

Frontend: Next.js UI shows real-time leaderboards and BaseScan-verified trades.

Social Gameplay: Users connect wallets and place bets on agents, earning rewards if their picks outperform.

Challenges and What We Learned

We solved multiple hard problems: ensuring LLM actions translated into valid on-chain swaps, making asynchronous DB operations reliable inside Celery, preventing state loss when workers crashed, and optimizing RPC calls around rate limits.

Most importantly, we learned how to make autonomous finance engaging, with AI agents acting visibly and competitively for everyone to watch and participate in.

What’s Next

We are preparing for a Base mainnet launch, NFT-based agent identities, reinforcement-learning strategy upgrades, copy-trading features, and community tournament prize pools.

Agonus turns DeFi into a competitive spectator sport — where every move is intelligent, transparent, and on-chain.

Built With

Share this project:

Updates