kayba-ai/recursive-improve

Open Source

Autonomous Agents Updated Apr 5, 2026

🔄 Updated Apr 2026 🖥️ Self-hostable

Overview

The kayba-ai/recursive-improve is a recursive self-improvement framework for AI agents. It captures execution traces, analyzes failure patterns, and applies targeted fixes. This framework uses a keep-or-revert evaluation to assess the effectiveness of the applied fixes.

Problem It Solves

Improving the performance and reliability of AI agents through self-improvement

Target Audience: AI researchers and developers

Inputs

• Execution traces
• Failure patterns
• Agent performance metrics

Outputs

• Improved agent performance
• Targeted fixes
• Evaluation results

Example Workflow

1 Execution trace capture
2 Failure pattern analysis
3 Targeted fix application
4 Keep-or-revert evaluation
5 Performance metric assessment
6 Agent update

Sample System Prompt


              Analyze the execution traces of the current agent and apply targeted fixes to improve its performance on the given task

Tools & Technologies

Scikit-learn TensorFlow PyTorch

Alternatives

• Google's AutoML
• Microsoft's NNI
• Facebook's NAS

FAQs

Is this agent open-source?: Yes
Can this agent be self-hosted?: Yes
What skill level is required?: Advanced

Rate This Agent

Your rating:

Reviews

Loading reviews...