Mango Chat — Setup & Usage Guide

About Mango Chat

Free, open-source voice dictation for Windows.

Built in Rust. Under 50 MB of memory. Speak into any application—Mango Chat streams your audio to your chosen provider and transcribes in real time.

Supports Deepgram, AssemblyAI, OpenAI Realtime, and ElevenLabs—all via WebSocket for ultra-low-latency results.

Deepgram and AssemblyAI offer a combined $250+ in free credits—that's 700+ hours of dictation, no credit card required.

Download

Get the latest release from GitHub:

Latest Release
  • Windows 10 / 11
  • No admin privileges required
  • ~8 MB installer

Installation & Getting Started

  1. Download MangoChat-Setup-Latest.exe from the latest release link
  2. Run the installer — no admin privileges needed
  3. Accept defaults or change the install path
  4. Click Install, then Finish
  5. Search Mango Chat in the Windows search bar to launch the app

Settings are stored in %LOCALAPPDATA%\MangoChat\. Uninstalling preserves your settings and keys.

Configuration

  1. Open Settings (gear icon)
  2. Go to Providers
  3. Select a provider (Deepgram, AssemblyAI, OpenAI, or ElevenLabs)
  4. Paste your API key
  5. Click Verify to test the connection
  6. Click Save — you're done

API keys are encrypted with Windows DPAPI. Never stored in plain text.

Features

  • Screenshot capture — Right Alt to snip, three clipboard modes: copy path for terminals (paste into Claude Code, Codex), copy image for browsers and chat, or copy & edit to annotate first. No save dialogs, no file dragging.
  • Session safeguards — configurable max session length and inactivity timeout to prevent runaway provider costs
  • Voice commands — built-in commands like enter, back, undo, copy, paste, and more
  • Browser commands — open URLs by voice (e.g. "open github" launches a site in your chosen browser)
  • Program launcher — say a single word to open any app. Just say cursor or chrome on its own—no prefix needed. Map any word to an .exe path in Settings.
  • Text aliases — create text shortcuts that expand on dictation (e.g. "my email" types your address)
  • Appearance — change color theme, window opacity, and anchor position
  • Noise suppression — High (recommended) or Low. Filters background noise before sending audio to your provider.

Tips

Mango Chat is designed to bridge the gap between your voice and your tools—whether it's a coding agent, a research assistant, or any AI-powered application.

  • Speak naturally — don't worry about perfect transcription. AI tools like Claude and ChatGPT are smart enough to understand your intent from context, even with minor errors.
  • Don't redo, just correct — if you misspeak or see an error, don't start over. Simply say your correction and keep going. The AI will understand.
  • Keep sentences short — speak in short, focused sentences. This produces cleaner transcripts and faster responses.
  • Say enter when ready — this sends your input to the active application. Think of it as pressing the submit button with your voice.
  • Commands must be standalone — voice commands like cursor, chrome, undo, or open github only trigger when spoken by themselves, not as part of a sentence.
  • Combine with screenshots — dictate to your AI, take a quick screenshot with Right Alt, and let the AI see what you see. Voice + visuals is a powerful workflow.

Screenshots

Compact dictation bar showing provider status and waveform Compact View
Compact bar with screenshot, terminal, and clipboard tools With Screenshot Tools
Dictation Bar
Mango Chat compact bar anchored on Windows desktop, showing minimal footprint Desktop — Minimal Footprint
Provider settings showing Deepgram, OpenAI, ElevenLabs, and AssemblyAI configuration Provider Settings
Session settings with microphone, hotkey, noise suppression, and screenshot options Session Settings
Browser commands tab showing voice-activated URL shortcuts Browser Commands
Custom text aliases for text expansion shortcuts Custom Text Aliases
Built-in Mango Chat voice commands reference Voice Commands
Appearance settings with theme color, transparency, and monitor options Appearance
Usage dashboard tracking time and data across all providers Usage Dashboard
Built-in FAQ with common questions and answers In-App FAQ
About page with version info, updates, and diagnostics About & Updates

FAQ

When you start recording, Mango Chat listens for audio from your device and streams it to your selected provider for transcription. Place your cursor in a text field to begin dictating.
Right Ctrl to start and stop recording. Right Alt to capture a screenshot. Both can be enabled or disabled in Settings.
Deepgram, OpenAI Realtime, ElevenLabs Realtime, and AssemblyAI. All connect via WebSocket.
Mango Chat is free. You only pay your speech provider—on average around $0.50/hour or less. Deepgram and AssemblyAI both offer generous free credits (no credit card required) that can last hundreds of hours. Start with those.
Encrypted at rest and stored locally in AppData/Local/MangoChat. Only transmitted over secure connections when authenticating with your chosen provider.
No. Mango Chat has no built-in telemetry. During recording, audio is sent only to your selected provider for transcription.
enter, new line, back, undo, redo, copy, paste, cut, select all. Define custom aliases and URL commands in Settings.
Yes. You can edit browser commands, text aliases, and app locations from the Commands tab in Settings.
Enable screenshot capture in Settings, move your cursor to the monitor you want, press Right Alt, then drag to select the region.
Based on your selected mode, Mango Chat can copy the image path to your clipboard, copy the image content, or open it in an editor for annotation.
Use Open images folder in Settings to open the active screenshot directory.
These are provider-dependent and may be caused by audio quality, speech clarity, network latency, or inherent limitations of the model.
Yes. Settings > Session > Refresh to re-scan devices. No restart needed.
Mango Chat monitors your capture device mute state. Mute your headset to pause, unmute to resume.
Yes. MIT license. View source on GitHub.