Vibe editing: asynchronous voice-to-edit with AI agents

This project demonstrates an AI-powered, asynchronous workflow that transforms spoken ideas into precise, reviewable document edits.

Workflow diagram

graph TD
    subgraph " "
        IntegA["✨ Integrator Agent"]
        OV["📚 Obsidian Vault"]
        IntegA -- "Updates" --> OV

    end

    subgraph " "
        U["🧑‍💻 User"]
        VN["📝 Voice Note"]
        IA["🤖 Instruct Agent"]
        IIF["📃 Instructions File"]

        U --> VN
        VN --> IA
        IA --> IIF
        
    end

    IIF --> IntegA
    IntegA --> IIF

    style U fill:#D6EAF8,stroke:#333,stroke-width:2px
    style VN fill:#FCF3CF,stroke:#333,stroke-width:2px
    style IIF fill:#FCF3CF,stroke:#333,stroke-width:2px
    style OV fill:#E8DAEF,stroke:#333,stroke-width:2px
    style IA fill:#ABEBC6,stroke:#333,stroke-width:2px
    style IntegA fill:#ABEBC6,stroke:#333,stroke-width:2px

Loading

Setup

Installation

1. Clone the repository:

   git clone https://github.com/azhutov/vibe-editing.git
   cd vibe-editing

2. Install Python dependencies:

   pip install llm rich

3. Install an llm model plugin: The instruct_agent.py script uses gemini-2.5-flash-preview-04-17. Install llm-gemini and configure your API key (https://aistudio.google.com/apikey).

   llm install llm-gemini
   llm keys set gemini

4. Create data directories:

   mkdir -p data/voice-notes-demo data/transcripts data/integration_instructions

How to start the workflow in Cursor

The workflow involves two AI agents: the Instruct Agent (processes voice notes) and the Integrator Agent (waits for instructions).

1. Open Cursor in the vibe-editing project root.
2. Run the Integrator Agent in Cursor chat with claude-3.7-sonnet model:

   Run @integrator_workflow
   

   This will start both agents and guide you through the process.