AFlow: Automating Agentic Workflow Generation

Free

Automating Agentic Workflow Generation

FreeFree tier

Type

Open Source

Visit Website GitHub (40 stars)

About AFlow: Automating Agentic Workflow Generation

AFlow is an automated framework for generating and optimizing agentic workflows used by large language models (LLMs). It reformulates workflow optimization as a search problem over code-represented workflows, where LLM-invoking nodes are connected by edges. The system employs Monte Carlo Tree Search to iteratively explore and refine workflows through code modification, tree-structured experience, and execution feedback. Empirical evaluations across six benchmark datasets show a 5.7% average improvement over state-of-the-art baselines. Notably, AFlow enables smaller models to outperform GPT-4o on specific tasks at only 4.55% of its inference cost. The project is open source and the code is available on GitHub.

Key Features

Reformulates workflow generation as a search problem over code-represented workflows

Uses Monte Carlo Tree Search for iterative refinement

Incorporates tree-structured experience and execution feedback

Code-driven workflow modification and optimization

Achieves 5.7% average improvement over state-of-the-art baselines on six benchmarks

Enables smaller models to outperform GPT-4o at 4.55% of its inference cost

Pros & Cons

Pros

Fully automated workflow generation without manual initial setup
Significant performance improvement (5.7%) over existing methods
Substantial cost reduction (up to 95.45% vs GPT-4o) for specific tasks
Allows smaller models to achieve high performance on complex tasks

Cons

Limited to code-represented workflows, may not cover all agent architectures
Performance improvements are benchmark-specific and may not generalize
Requires execution feedback, which may be computationally intensive

Best For

Automated construction of agentic workflows for complex LLM tasksCost-efficient deployment of smaller LLMs with competitive performanceResearch in automated workflow optimization and LLM agent systems

FAQ

What is AFlow?

AFlow is an automated framework that generates and optimizes agentic workflows for large language models using Monte Carlo Tree Search.

How does AFlow work?

It reformulates workflow optimization as a search over code-represented workflows, iteratively refining them through code modification, tree-structured experience, and execution feedback.

What are the main benefits of AFlow?

AFlow achieves a 5.7% average improvement over state-of-the-art baselines and enables smaller models to outperform GPT-4o on specific tasks at 4.55% of its inference cost.

Is AFlow open source?

Yes, the code for AFlow is publicly available on GitHub.