AFLOW: MetaGPT's Open-Source Agent Workflow at 4.55% of GPT-4o Cost

AFLOW by MetaGPT automates Agentic Workflow optimization, reducing LLM costs by 95%, with open-source MCTS-driven solutions for seamless AI deployment.

For LLM practitioners, implementing and making LLM applications work requires manually building and repeatedly debugging Agentic Workflows.

This is undoubtedly a tedious process, involving modifying similar code over and over, debugging prompts, manually running tests, and observing results.

Moreover, switching to a different LLM might render the workflow ineffective, leading to high labor costs. Many companies even hire dedicated Prompt Engineers to handle this work.

Now, Agentic Workflow has its own automation optimization tool.

MetaGPT has open-sourced AFLOW, which uses MCTS to automatically search and optimize Agentic Workflows.

It allows for the complete automation of building and optimizing Agentic Workflow problems, eliminating the need for hand-coding and prompt debugging.

This is a further exploration of automated prompt optimization.

Through Monte Carlo Tree Search, AFLOW fully takes over the generation and optimization process of Agentic Workflows, outperforming other workflow automation solutions and even surpassing all hand-crafted workflow baselines in comparisons.