Agora-Opt: Decentralized LLM Agents for Optimization Modeling
Agora-Opt is a versatile framework for optimization modeling that integrates decentralized discussions with a read-write memory system. This setup enables various agent teams to autonomously generate comprehensive solutions, which they can reconcile via an outcome-focused debate protocol. The memory retains artifacts verified by solvers and records past resolutions of disagreements, facilitating improvement without the need for training. Its design is adaptable to various backbones and methods, minimizing reliance on specific base models and allowing transferability among different LLM families. In public benchmarks, Agora-Opt demonstrates the highest overall performance.
Key facts
- Agora-Opt is proposed in arXiv:2604.25847
- It combines decentralized debate with a read-write memory bank
- Multiple agent teams produce end-to-end solutions independently
- Solutions are reconciled through an outcome-grounded debate protocol
- Memory stores solver-verified artifacts and past disagreement resolutions
- Supports training-free improvement over time
- Flexible across backbones and methods
- Achieves strongest overall performance on public benchmarks
Entities
Institutions
- arXiv