document.md

Student & Developer Documentation

Overview

Welcome to the RemiAI Framework! This document is designed to help you understand how to customize, configure, and make this application your own. This framework is built to be "Plug-and-Play"—meaning you don't need to know Python or complex AI coding to use it. It includes Text Generation (chat with AI), Text-to-Speech (TTS — convert text to voice), and Speech-to-Text (STT — extract text from audio files), all running 100% offline.

🛠️ Setup & How to Customize

0. Quick Setup (Important!)

Before running the app, you must ensure the AI engine files are downloaded correctly. GitHub does not store large files directly, so we use Git LFS.

1. Install Git LFS:
   • Download and install from git-lfs.com.
   • Open a terminal and run: git lfs install
2. Pull Files:
   • Run: git lfs pull inside the project folder.
   • Why? Without this, the app will say "RemiAI Engine Missing" or "Connection Refused".

1. Changing the AI Name

Want to name the AI "Jarvis" or "MyBot"?

1. Open index.html in any text editor (VS Code, Notepad, etc.).
2. Search for "RemiAI" or "Bujji".
3. Replace the text with your desired name.
4. Save the file.
5. Restart the app (npm start), and your new name will appear!

2. Replacing the AI Models (LLM, TTS, STT)

This framework is a Universal Wrapper. You can swap out any of the three "brains" (Text, Speech, Hearing) to build your own dedicated application.

A. Changing the Chat Model (Text Generation)

1. Download: Get a .gguf model from Hugging Face (e.g., Llama-3-8B-GGUF).
2. Rename: Rename it to model.gguf.
3. Replace: Overwrite the existing model.gguf in the root folder.
4. Restart: Run npm start.

B. Changing the Text-to-Speech (TTS) Voice

The framework uses Piper TTS.

1. Download: Get a voice model (.onnx) and its config (.json) from Piper Voices.
2. Place Files: Put both files in engine/piper/ (e.g., my-voice.onnx and my-voice.onnx.json).
3. Update Code:
   • Open main.js.
   • Search for: engine/piper/en_US-lessac-medium.onnx
   • Replace the filename with your new .onnx file name.
4. Restart: Run npm start.

C. Changing the Speech-to-Text (STT) Engine

The framework uses Whisper.cpp.

1. Download: Get a model in GGML/Binary format (ggml-*.bin) from Hugging Face (ggerganov/whisper.cpp).
2. Place File: Put the file in engine/whisper/.
3. Update Code:
   • Open main.js.
   • Search for: engine/whisper/ggml-base.en.bin
   • Replace the filename with your new .bin file name.
4. Restart: Run npm start.

Hardware Warning:

• Good Configuration: i3 (8GB RAM) for basic usage.
• Recommended: i5 (16GB RAM) for larger models.
• Note: Using models larger than your RAM capacity will crash your computer. Stick to "Q4_K_M" quantizations for the best balance.

3. Customizing the UI

All styles are in styles.css (or within index.html).

• Colors: Change the background colors or chat bubble colors in the CSS.
• Icons: Replace remiai.ico with your own .ico file to change the app icon.

4. Using Text-to-Speech (TTS)

The TTS feature converts typed text into natural-sounding English speech using the Piper engine.

How to Use:

1. Click the 🔊 Speaker icon in the sidebar.
2. Type the text you want to hear in the text area.
3. Click "Speak" — the audio will generate and play automatically.
4. Click "Download Audio" to save the .wav file to your preferred location (a native Save dialog will appear).

Customization:

• The TTS voice model is stored at engine/piper/en_US-lessac-medium.onnx.
• You can replace it with other Piper ONNX voice models from Piper Voices.
• Download a new .onnx model + its .json config file and place them in engine/piper/.

5. Using Speech-to-Text (STT)

The STT feature extracts text from audio files using the Whisper engine (runs as a local server).

How to Use:

1. Click the 🎙️ Microphone icon in the sidebar.
2. Click "Browse Audio File" to select your audio file.
3. Supported formats: .wav, .mp3, .m4a, .ogg, .flac.
4. Click "Transcribe" — wait for processing (10-30 seconds depending on file length).
5. The transcribed text will appear below. Click "Copy" to copy it to your clipboard.

Requirements:

• ffmpeg.exe and ffmpeg.dll must be present in the bin/ folder for audio format conversion.
• If missing, download FFmpeg from ffmpeg.org and place the files in bin/.

6. Dynamic Resource Management (New!)

To ensure the application runs smoothly even on lower-end devices, we implemented a dynamic resource management system.

• Behavior: When you are in the Chat tab, the heavy AI model (Text Generation) is loaded into RAM.
• Optimization: When you switch to TTS, STT, or Web Browser tabs, the main AI model is automatically unloaded/stopped. This frees up to 2GB+ of RAM and significant CPU usage, allowing the TTS/STT engines to run faster and the browser to be more responsive.
• Reloading: When you switch back to the Chat tab, the model automatically restarts.
• Note: You might see "Connecting..." for a few seconds. If it stays stuck, click the "Refresh App" button.

7. Offline Dependencies

All libraries are bundled locally — no internet needed after initial setup:

• Lucide Icons: Loaded from node_modules/lucide/ (not from CDN).
• Marked.js: Loaded from node_modules/marked/ (not from CDN).
• If icons or markdown rendering is broken, simply run npm install to restore them.

❓ Frequently Asked Questions (FAQ)

Q: Do I need Python? A: No. The application comes with a pre-compiled engine (bujji_engine.exe / llama-server.exe) that runs the model directly.

Q: Why does it say "AVX2"? A: AVX2 is a feature in modern CPUs that makes the AI run faster. The app automatically detects if you have it. If not, it switches to a slower but compatible mode (AVX).

Q: The app opens but doesn't reply / "RemiAI Engine Missing" Error. A:

1. Git LFS Issue: This usually means you downloaded "pointers" (tiny files) instead of the real engine. Open a terminal in the folder and run git lfs pull.
2. Model Issue: Check if model.gguf exists in the engine folder.
3. Console Check: Open Developer Tools (Ctrl+Shift+I) to see errors.

Q: I see "Content Security Policy" warnings in the console. A: We have configured safeguards (index.html meta tags) to block malicious scripts. The CSP is set to only allow local resources ('self') and the local API server (127.0.0.1:5000). All external CDN dependencies have been removed.

Q: How do I build it into an .exe file? A: Run the command:

npm run dist

This will create an installer in the release folder that you can share with friends!

if you are facing errors while building open the power shell as an administrator and run the above command then it will works 100%

Q: TTS says "Piper TTS executable not found". A: Make sure piper.exe exists in engine/cpu_avx2/ (or engine/cpu_avx/). Run git lfs pull to download all engine binaries.

Q: STT says "Whisper server failed to start". A:

1. Check that whisper.exe exists in engine/cpu_avx2/ (or engine/cpu_avx/).
2. Check that ffmpeg.exe and ffmpeg.dll are present in the bin/ folder. The Whisper server needs FFmpeg for audio conversion.
3. Run git lfs pull to ensure all files are fully downloaded.

Q: STT says "No speech detected". A: Make sure your audio file contains clear English speech. Background noise or non-English audio may cause transcription failures. Try with a clear .wav recording first.

Q: Can I use TTS and STT together? A: Yes! You can generate speech with TTS, save the .wav file, then upload it to STT to verify the transcription. They work independently and can be used simultaneously.

Q: Does the app need internet to work? A: No. After the initial npm install and git lfs pull setup, the app runs 100% offline. All models, engines, icons, and libraries are bundled locally.