Skip to content

outsung/hebi

Repository files navigation

Hebi

Hebi

AI-driven research platform with automated experiment pipelines

Hebi Screenshot


Overview

Hebi is an Electron desktop app for managing AI-driven research workflows. It provides a chat-based interface powered by Claude and Codex agents, with a structured 3-step pipeline for running experiments, managing datasets, and iterating on results.

Topics are user-configured with custom input/output definitions and evaluation criteria, making Hebi applicable to any research domain.

Workflow: Analyze → Experiment → Optimize

Step Description
Analyze AI agent analyzes a paper, URL, or text to extract reproducible methods and tunable parameters
Experiment Generates implementation code based on analysis, runs it against a dataset, and produces results
Optimize Iteratively improves an existing implementation by adjusting parameters and comparing metrics

Each step is triggered separately by the user, giving full control over the research process.

Development

npm install
npm run dev

Credits

  • Kanna — Web UI for Claude Code & Codex CLIs
  • Feynman — AI research companion

About

AI-driven research platform with automated experiment pipelines

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors