Skillget

AI_MIDI

agent
Guvenlik Denetimi
Basarisiz
Health Uyari
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Basarisiz
  • network request — Outbound network request in frontend/js/app.js
  • network request — Outbound network request in frontend/js/arrangement/audio_engine.js
  • exec() — Shell command execution in frontend/js/arrangement/midi_parse.js
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

All you need is one API to let AI assist you in music creation—modify MIDI files, get suggestions, and more. No complex setup, just plug and play!

README.md

🎹 AI_MIDI — AI-Assisted Music Theory & MIDI Composition

English | 中文

AI_MIDI is a local-first desktop app that lets AI assist your music creation: add chords to a melody, translate lyrics, design vocal runs, generate new ideas — and arrange everything in an FL Studio-style arrangement view. All you need is one API key. Your projects, files and conversations never leave your computer.

v3.0.0 upgrades AI_MIDI from a web app to a standalone desktop application — install and go, no Python required.

✨ Features

💬 Multi-turn Chat (main workbench)

  • Project archive + workbench dual view — manage your creations like folders
  • The AI operates MIDI directly: say "Write an 8-bar C major melody with a syncopated rhythm" and it reads the built-in theory library, creates the file and organizes your project — every step visible and reversible
  • Workspace binding — link a local folder; files sync both ways
  • Undo / recycle bin — every file operation can be rolled back
  • Thinking process visualization — reasoning and tool calls stream live
  • Drafts auto-save per project; parchment / dark-blue dual theme

⚡ Quick Tasks (one-shot processing)

  • Add Chords — harmonize a melody into MIDI
  • Translate Lyrics — make translations fit the melody (Japanese kana supported)
  • Design Melisma — generate vocal ornaments & coloratura
  • Custom Requests — free-form tasks, optionally output as MIDI

🎚 Arrangement View (new in v3 · early beta)

  • FL Studio-style timeline: tracks, clip dragging, snapping, split, loop range
  • Double-click any clip to edit notes in the piano roll
  • Output rack — swap instrument sounds, import SF2 soundfonts, drag-to-assign
  • Global BPM — change it once, it applies everywhere; new AI-generated MIDI follows it automatically (existing files keep their own tempo)
  • Export to WAV with full tail decay
  • Auto-save keeps your arrangement safe

ℹ️ Early beta: VST plugins and AI-link are not supported yet — feedback welcome.

🔊 Audio Engine (Windows)

  • Built-in JUCE native engine: low-latency ASIO monitoring, live level meters, offline WAV bounce
  • Without the engine the app falls back to browser audio automatically (fully functional, slightly higher latency)

🚀 Quick Start

Windows (recommended)

  1. Grab AI_MIDI_Setup_3.0.0_windows_amd64.exe from Releases and run it — no admin rights needed
  2. On first launch, read the usage notice, then open Settings and enter your API key
  3. Create a project and start composing!

Linux

  1. Download AI_MIDI_v3.0.0_linux_amd64.tar.gz from Releases and unzip it
  2. chmod +x RUN.sh && ./RUN.sh (desktop mode needs webkit2gtk; otherwise ./AI_MIDI -browser opens in your browser)
  3. Linux uses browser audio synthesis — no engine needed

First-time Configuration

Open the Settings page and fill in your API key. The default connects to DeepSeek; any OpenAI-compatible provider works (OpenAI, Gemini, local LLMs…).

Field Description
API Key Your provider key (stored only in local settings.json, never uploaded)
Base URL Default https://api.deepseek.com; any OpenAI-compatible endpoint works
Model Default deepseek-v4-pro; click "Refresh" to fetch the list from your provider

🛠 Build from Source (developers)

Requirements: Go ≥ 1.27, Wails v2 (go install github.com/wailsapp/wails/v2/cmd/wails@latest), Windows 10+ (Linux: webkit2gtk for desktop mode).

git clone https://github.com/abab996/AI_MIDI.git
cd AI_MIDI
wails build        :: Windows -> buildin\AI_MIDI.exe
RUN.bat

Linux: go build -o AI_MIDI . && chmod +x RUN.sh && ./RUN.sh. Architecture & design documents live in docs/.

📖 Usage Guide

Quick Tasks

  1. On the chat page, click ⚡ Quick Task at the top
  2. Drop in a MIDI file → pick a task → fill in your requirements → Start
  3. Download the result when done

Multi-turn Chat

  1. Create a project in the archive (＋ New Project)
  2. Describe your goal — e.g. "Harmonize this melody with jazz chords"
  3. The AI reads the theory library first, then creates files step by step; undo anything you don't like

Arrangement View

  1. Click ▦ Arrange on the workbench
  2. Drag MIDI files from the left panel onto tracks, swap sounds from the output rack
  3. Set the global BPM, arrange your clips, hit ⬇ WAV to export

Tips

  • Switch theme (parchment ⇄ dark blue) via the top-right button
  • Switching projects auto-saves your draft; it is restored when you return
  • Arrangement data auto-saves — still keep backups of important projects

🔒 Privacy & Security

  • Your API key is stored only in local settings.json — never in code, never uploaded
  • All project data (conversations, MIDI files) lives on your machine
  • The key is sent only to the service address you configure
  • The app runs fully locally and collects no usage data

📄 License

Code licensed under Apache License 2.0.
The bundled audio engine (aimidi-engine.exe, Windows installer) is a closed-source component distributed via the installer.

Developers: architecture and design documents live in docs/.

AI_MIDI — compose smarter, theory-first. 🎵

Yorumlar (0)

Sonuc bulunamadi