kalengh ⚡

Real-Time Audio & Visual AI Copilot for Engineers and Problem Solvers

Engineered by exhuzaifa (Huzaifa Malik)

---

💡 Overview

kalengh is an ultra-fast, discreet desktop AI assistant built on Electron, designed by exhuzaifa to provide contextual, sub-second intelligence during technical interviews, live coding sessions, and complex architectural discussions.

Unlike standard AI chat interfaces, kalengh operates passively in the background:

• 🎙️ Real-Time Speech Detection: Hooks directly into incoming speaker audio via Voice Activity Detection (VAD) and Groq Whisper (whisper-large-v3-turbo) to transcribe questions the millisecond the speaker finishes.
• 👁️ Multimodal Screen Capture: Takes discreet snapshots of coding prompts, IDEs, or whiteboard diagrams without triggering browser screen-sharing banners.
• 🧠 Staff-Level Reasoning: Powered by ultra-fast LLMs (openai/gpt-oss-120b, llama-3.3-70b-versatile, or gemini-2.5-flash-lite) with structured coding rules that prioritize problem understanding, complexity derivations, and optimal implementations.
• 👻 Discreet Click-Through HUD: A floating glassmorphic window that can be rendered transparent, clicked through, or toggled with a single keystroke.
• 💻 100% Local Mode Available: Option to run completely offline on your own machine using Ollama and local Whisper models.

---

🛠️ Architecture & Working Mechanism

flowchart TD
    subgraph Capture ["Audio and Vision Capture"]
        A["Speaker / System Audio"] -->|Loopback / CoreAudio| B["Voice Activity Detector (VAD)"]
        B -->|Silence Gap Detected| C["Groq Whisper API / Local Whisper"]
        D["Screen / IDE"] -->|Native Desktop Capture| E["Screenshot Buffer & Vision Pipeline"]
    end

    subgraph CoreEngine ["kalengh Core Engine"]
        C -->|Transcribed Text| F["Question Classifier & Prompt Router"]
        E -->|Visual Context| F
        F --> G{"Provider Mode"}
        G -->|BYOK Cloud| H["Groq API Key Pool & Rotation"]
        G -->|Local AI| I["Local Ollama / LLaMA 3.1"]
        H --> J["High-Speed LLM Inference"]
    end

    subgraph Overlay ["Discreet Overlay"]
        J --> K["Format Filter: Zero Math Symbols & Staff Tone"]
        I --> K
        K --> L["Transparent Glassmorphic HUD"]
    end

Loading

Complete Pipeline

1. Audio Interception: On Windows, kalengh leverages loopback system capture to isolate speaker audio from user microphone audio. On macOS, it spawns a lightweight CoreAudio helper (SystemAudioDump).
2. Dynamic Key Rotation: Cloud mode utilizes a self-healing key pool. If a provider returns a 429 (rate-limit) or 401 error, kalengh automatically rotates to the next API key in milliseconds without dropping the session.
3. Optimized Interview Persona: The response engine strictly enforces concise, spoken-friendly answers without messy Markdown syntax or mathematical notation that cannot be vocalized naturally.

---

🚀 Getting Started & Installation

Option 1: Pre-Built Installers

• Windows: Download kalengh Setup.exe from the Latest Releases and double-click to install.
• macOS (Apple Silicon M1/M2/M3/M4): Download kalengh-arm64.dmg from Releases. Drag it into your Applications folder.
• macOS (Intel Mac): Download kalengh-x64.dmg from Releases. Drag it into your Applications folder.

  Note for macOS: On first launch, open System Settings > Privacy & Security > Screen Recording (and Microphone) and enable permissions for kalengh.

• Linux: Download the .deb package (kalengh_amd64.deb). Install with sudo dpkg -i kalengh_amd64.deb or launch directly.

---

🎧 Audio Setup & Permissions Guide (macOS & Windows)

kalengh isolates audio into two streams to ensure sub-second contextual responses without feedback loops:

• Interviewer Voice (Incoming): Captured directly from the system speaker output.
• Your Voice (Outgoing): Captured from your microphone (when enabled).

🍎 macOS Permissions & Setup

Important

Why is kalengh NOT listed under "Microphone" in macOS System Settings?
macOS routes interviewer audio through your speakers, not your microphone. To capture what the interviewer says without requiring virtual audio cables (like BlackHole or Soundflower), kalengh uses Apple's native ScreenCaptureKit CoreAudio pipeline.
In macOS, ScreenCaptureKit is governed by Screen & System Audio Recording, NOT the Microphone!

Step-by-Step macOS Audio Setup:

1. Open System Settings ( > System Settings).
2. Navigate to Privacy & Security → Screen & System Audio Recording (or Screen Recording on macOS Ventura 13.x / Monterey).
3. Find kalengh in the list and toggle the switch to ON.
   • If kalengh is not in the list, click the + button and add kalengh.app from your /Applications folder.
4. ⚠️ MANDATORY RESTART: After enabling the toggle, press Cmd + Q to Quit kalengh completely and reopen it. macOS requires the application to restart for the audio capture permissions to attach to the process.
5. Gatekeeper note (Ventura / OpenCore): If macOS says "App is damaged or cannot be opened", open Terminal and run:

   xattr -cr /Applications/kalengh.app

---

🪟 Windows Setup & Bluetooth Headphone Fix

On Windows, kalengh uses WASAPI System Loopback to record speaker audio directly without requiring permission prompts. However, watch out for the Bluetooth Profile Trap:

The Bluetooth Headphone Trap:

When using Bluetooth headphones (AirPods, Sony, Bose, etc.), Windows splits audio into two profiles:

1. Headphones (Stereo) — Default multimedia audio.
2. Headset (Hands-Free AG Audio) — Meeting communication audio.

Warning

If Zoom, Google Meet, or Microsoft Teams routes call audio to the Hands-Free device while Windows default output is set to Stereo, kalengh will record silence!

How to Fix:

• Inside your meeting app (Zoom / Google Meet / Teams), go to Audio Settings → Speaker.
• Set Speaker to "Same as System" or select your default playback device.
• In kalengh → Customize, set Audio Mode to:
  • "Speaker Only": Listens exclusively to the interviewer.
  • "Both": Listens to both the interviewer and your microphone.
• Ensure your system output volume is audible so the Voice Activity Detector (VAD) threshold triggers naturally.

---

📦 Source Code Note (node_modules)

• Pre-built Installers: All dependencies and binaries are pre-compiled and bundled inside the .exe and .dmg installers. End users never need to install or download node_modules.
• Running from Source: If cloning this repository, always run npm install first to create your local platform-specific node_modules before running npm start.

Option 2: Running & Building from Source

Ensure you have Node.js (v18 or higher) and npm installed.

1. Clone the repository:

   git clone https://github.com/Techsynx/kalengh.git
   cd kalengh

2. Install dependencies:

   npm install

3. Start the development application:

   npm start

4. Package distributable installers:

   # Generates .exe on Windows, .dmg on macOS, or .AppImage on Linux
   npm run make

   Artifacts are output to the out/make directory.

---

🔑 API Keys & Configuration

kalengh is completely free and open-source. It supports two operation modes:

1. Bring Your Own Key (BYOK) Mode

• Gemini API Key (Required for vision and multimodal analysis): Get a free API key at Google AI Studio.
• Groq API Key (Recommended for sub-second Whisper audio transcription & LLM speed): Get a free API key at Groq Console. You can provide comma-separated keys for automatic round-robin rotation.

2. Local AI Mode (100% Offline)

• Run models entirely on your own GPU/CPU with Ollama.
• Set your Ollama Host (default http://127.0.0.1:11434) and choose models like llama3.1 or qwen2.5-coder.

---

⌨️ Default Keyboard Shortcuts

Shortcut (Windows/Linux)	Shortcut (macOS)	Action
Ctrl + \	Cmd + \	Toggle HUD Visibility (Instantly show/hide)
Ctrl + M	Cmd + M	Toggle Click-Through (Interact with background windows)
Ctrl + Enter	Cmd + Enter	Submit / Trigger Next Step
Ctrl + Up / Down	Alt + Up / Down	Move HUD Window
Ctrl + [ / ]	Cmd + [ / ]	Navigate Previous / Next Response
Ctrl + Shift + Up / Down	Cmd + Shift + Up / Down	Scroll Response

---

👨‍💻 Author & Creator

kalengh was conceived, architected, and engineered by exhuzaifa (Huzaifa Malik).

• 🌐 Portfolio: exhuzaifa.vercel.app
• 💼 LinkedIn: linkedin.com/in/huzaifa-malikk
• ⚡ Contra: contra.com/huzaifa_malik
• 🐙 GitHub: @Techsynx / @exhuzaifa
• ✍️ Dev.to: @exhuzaifa
• 📧 Contact: exhuzaifa@gmail.com

If you find kalengh helpful or inspiring, please give the repository a ⭐ star!

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.
Copyright (c) 2026 exhuzaifa (Huzaifa Malik).