Skip to content

AFlow

A framework that generates and optimizes agent workflows

Visit official site

What AFlow is

The framework is described as automatically generating and optimizing agent workflows with Monte Carlo tree search over a code-represented search space. It is presented as an alternative to hand-developed workflows.

What you can do with it

  • Generate and tune workflows on HumanEval
  • Run math problem experiments on MATH
  • Measure task performance on GSM8K

How it is built

  • Nodes expose LLM, temperature, format and prompt controls
  • Operators bundle predefined node combinations
  • An optimizer explores and an evaluator scores workflows

Pros & cons

✓ What we like

  • Custom operators and workflows are supported
  • An evaluator feeds performance back into optimization

! What to watch out for

  • Optimization parameters must be configured first
  • Workflows are represented as code, graphs or networks

FAQ

What search method does it use?

Monte Carlo tree search over a code-represented workflow space is described.

Can I define my own operators?

Yes, custom operators and workflows are listed as supported.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review