ianderrington / reflexion

Reflexion: an autonomous agent with dynamic memory and self-reflection

Reflexion Implementation

Code for the approach proposed in Reflexion: an autonomous agent with dynamic memory and self-reflection

Note

Reflexion is not a replacement for ReAct or any other decision-guiding approach! It is a simple retry technique that can be used enhance other approaches.

Check out an implementation using Reflexion to master HumanEval (88% pass@1 accuracy) and a blog post for a discussion on other applications.

Check out an interesting type-inference implementation OpenTau that uses a variation of Reflexion to play a 2-player game with type-checker.

If you have any questions, please contact [e-mail]

Cite

@article{shinn2023reflexion,
  title={Reflexion: an autonomous agent with dynamic memory and self-reflection},
  author={Shinn, Noah and Labash, Beck and Gopinath, Ashwin},
  journal={arXiv preprint arXiv:2303.11366},
  year={2023}
}