LLM-Powered Crypto Intelligence
Recurrent-Depth Transformer (RDT) implementation for autonomous flash-arbitrage trading on DEXs (Uniswap V3). Includes smart contracts in Solidity (~50 000 lines), a FastAPI dashboard, Telegram bot, RL agent loop, and multiple LLM integrations (Claude, OpenRouter, DeepSeek).
- Language: Python 3.13 / Solidity 0.8.20
- Blockchain: Uniswap V3, Flash Arbitrage, Base/Arbitrum
- AI: RDT (Recurrent-Depth Transformer), MoE, MLA, GQA
- Infra: FastAPI, Telegram Bot, RL Agents, Foundry, vLLM
src/
├── open_mythos/ # RDT model (1B to 1T)
├── contracts/ # Solidity flash-arb contracts
├── agents/ # RL agents, news, macro
├── connectors/ # Telegram, blockchain RPC
├── api/ # FastAPI dashboard
├── tests/ # Benchmarks and unit tests
└── training/ # Fine-tuning scripts (3B)
After running start.sh:
http://<host>:8080
GET /api/status returns JSON status for inference, contracts, agents, network.
# Clone
git clone https://github.com/lionj4/openmythos-bot.git
# Python
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
# Contracts (Foundry)
cd contracts && forge script src/OpenMythosMicroArb.sol:OpenMythosMicroArb
# Run full bot
./start.sh| Action | Command |
|---|---|
| API only | python run.py --no-agent --no-telegram |
| Full start | ./start.sh |
| Simulate profit | python simulate_profit.py |
| Health check | curl http://localhost:8080/health |
Lionel Jores · @lionelj4
MIT