Mira Help

Comprehensive guide to all Mira features

What is Mira

Mira is an AI voice assistant for Windows 10/11. Just say "Hey Mira" to start conversations, get weather forecasts, launch apps via Voice Launcher, analyze screens, and much more.

With Cortana discontinued, Mira is the cutting-edge voice assistant for Windows users. With a local AI it can also work without internet, protecting your privacy while providing powerful AI capabilities.

System Requirements

• OS: Windows 10 (64-bit) or later / Windows 11
• .NET: .NET 9.0 Desktop Runtime (bundled)
• RAM: 4GB+ (8GB+ recommended for local AI)
• Microphone: Built-in or USB microphone
• Storage: ~500MB+ (including Whisper model. Recommended large-v3-turbo: 1.6GB)

Quick Start

1. Install and launch Mira. An "M" icon appears in the system tray.
2. On first launch, the Whisper speech recognition model (default small: 466MB) is automatically downloaded. You can switch to the recommended large-v3-turbo (1.6GB) in Settings.
3. Say "Hey Mira". A beep sounds and an overlay appears in the bottom-right corner.
4. Speak your question or command. Mira responds with AI and reads it aloud.
5. Double-click the tray icon to open Settings.

Wake Word Setup

Mira can be activated with any word you choose. Change it in the "Wake Word" tab in Settings.

Settings — Wake word and Memory

• Default: "OK Mira", "Hey Mira", "Mira please"
• Interrupt phrases: "Mira stop please", "Mira stop", "Stop Mira" to interrupt speech (longer phrases recommended to avoid false triggers)
• Register multiple phrases separated by commas
• These defaults apply when the Windows display language is not Japanese. On Japanese Windows the defaults are different (wake words 「ミラお願い」, 「ミラ質問です」, "Hey Mira please"; interrupt phrases 「ミラストップして」, 「ミラ止めてください」, "Mira stop please")

Tips to Reduce False Recognition

Background noise or speaker audio feeding back into the microphone can cause unintended wake or stop triggers. Setting longer phrases significantly reduces false recognition.

Microphone Setup NEW

Choose the microphone Mira uses on the "Wake Word" tab in Settings. The microphone you select is used for both wake word detection and voice input.

• Default: "Windows default microphone"
• The selected microphone is stored by name, so unplugging and reconnecting USB devices will not silently switch Mira to a different one
• Changes take effect as soon as you save (no restart needed)

Test Microphone

The "Test microphone" button records for 3 seconds and shows the actual input level.

Microphone Boost

If your voice is still too quiet with the Windows microphone level at maximum, Mira can amplify the input up to 10x.

1. First press "Test microphone" at x1.0 and speak at your normal distance and volume
2. If the average exceeds the required value, no boost is needed
3. If it falls short, raise the slider gradually until the average clears the required value while the peak stays under 0.5

AI Overview

Mira supports 7+ AI providers. Select your provider and model in the "AI" tab in Settings.

Recommended Models (Tested)

The following models have been tested on real hardware. Screen Vision was tested with the cloud models. Gemma 4 is for conversation and cannot be used for Screen Vision in v2.3.1.

Ollama (Local AI)

1. Install Ollama from ollama.com
2. Run ollama pull llama3.1:8b in Command Prompt to download a model
3. Mira Settings → AI → Provider: Ollama, Model: llama3.1:8b

LM Studio (Local AI)

1. Install LM Studio from lmstudio.ai
2. Download a model and start the local server (default: http://localhost:1234/v1)
3. Mira Settings → AI → Provider: LM Studio

Lemonade (AMD GPU)

Google Gemini

1. Get an API key from Google AI Studio (free with a Google account)
2. Mira Settings → AI → Provider: Gemini, enter API key

OpenAI

Anthropic (Claude)

Custom Endpoint

Connect to any service that provides an OpenAI-compatible API. Specify the Base URL and model name.

Speech Recognition (Whisper STT)

Mira uses Whisper.NET for local speech recognition. Audio data is never sent externally (except in Live mode).

Choose model size in the "Speech Recognition" tab:

Text-to-Speech (TTS)

Mira uses a 4-tier fallback chain for TTS. If the top engine is unavailable, it automatically switches to the next.

1. AivisSpeech (localhost:10101) — Best quality Japanese
2. VOICEVOX    (localhost:50021) — High quality Japanese
3. Edge TTS    (Python edge-tts) — Multilingual neural voice
4. System.Speech               — Always available

TTS Engine Setup

AivisSpeech

The highest quality Japanese TTS engine. Download from aivis-project.com and run on port 10101.

VOICEVOX

Alternative to AivisSpeech. Download from voicevox.hiroshiba.jp. Default port 50021.

Edge TTS

Uses Microsoft Edge neural voices. Requires internet. Uses Python edge-tts: pip install edge-tts

Emotion TTS

When using AivisSpeech or VOICEVOX, Mira automatically switches between 6 voice styles based on the AI response ("Emotion Style Auto-Switch" on the "Text-to-Speech" tab in Settings):

• Normal / Casual (ふつー) / Sweet (あまあま, happy) / Calm (おちつき) / Teasing (からかい) / Wistful (せつなめ, sad)

Mira switches only when the selected speaker has styles with these names (for example AivisSpeech's "まお"). Otherwise it keeps the style you chose. Normal is the style you chose in Settings.

Voice Command List

Mira recognizes your voice and instantly executes the commands below. Intent detection runs locally without calling any AI, so commands run right away. Any utterance that doesn't match a command is processed as a normal AI conversation (adding and deleting schedule entries is handled by the AI, so an AI provider is required).

Screen Vision

AI captures and analyzes your screen. Requires a Vision-capable AI provider.

Search & Weather

Ask about the weather by voice, or open Google search results in your browser.

Web Watch NEW

Register frequently-checked web pages with keywords and get their content fetched and summarized by voice. Set up keyword-URL pairs in the "Web Watch" tab in Settings.

Voice Launcher

Register keyword-command pairs in the "Voice Launcher" tab in Settings. When your speech contains a registered keyword, the corresponding app or URL is launched.

System Operations

Control volume, clipboard, date/time, and other system functions.

Timers & Alarms

Set and manage countdown timers and alarms by voice.

In v2.3.1, timers cannot be cancelled or listed by voice in English (only in Japanese).

Stored at: %LOCALAPPDATA%\Mira\alarms.json

Memory & Schedule

Mira automatically remembers important information from conversations. Say "Remember this" to explicitly save memories.

Stored at: %LOCALAPPDATA%\Mira\memory.json, schedules.json

Screen Vision

AI captures and analyzes your screen. See Screen Vision commands for voice command details.

Recommended providers: gemini-flash-latest (free tier available), Claude Haiku 5.5. In v2.3.1, Screen Vision works only with cloud AI (Gemini, Claude, OpenAI).

Gemini Live

Real-time voice dialogue mode using Google's Gemini Live API. STT + LLM + TTS integrated into a single WebSocket for natural conversation.

• Dedicated wake phrase: "Mira Live" to activate (separate from regular wake words)
• Stop phrase for instant termination (pausing from the tray also ends it)
• Enable in the "Live" tab in Settings
• Model: gemini-3.8-live (changeable in the "Live" tab in Settings)
• Language (listening & replies): choose Japanese, English or auto-detect. The default is Match Windows language (auto-detect when Windows uses a language other than Japanese or English). With a language set, replies are in that language and short phrases are less often mistaken for another language
• The microphone is not sent while Mira is speaking or using a tool (so it doesn't pick up its own voice from the speakers). You can't interrupt by voice mid-sentence, but the stop phrase still ends Live mode
• Screen Capture (off by default for new installs from v2.3.1): when turned on in the "Live" tab, the AI looks at your screen when you ask. Screenshots are sent to Google. As this relies on the AI's judgment, it may occasionally look without being asked

Custom Notes

Write additional instructions for Mira in Settings. Example: "Search Yahoo Weather when asked about weather" / "Use polite language"

Web Watch Setup NEW

Register keyword-URL pairs in the "Web Watch" tab in Settings.

1. Open Settings (double-click Mira tray icon → "Settings" button)
2. Select the "Web Watch" tab
3. Enter a keyword you'll use in voice commands (e.g., "latest news")
4. Enter the URL of the page to fetch (e.g., https://news.yahoo.co.jp/)
5. Click "Add"
6. Click "Save" to save settings (Cancel to discard)

Voice Launcher Setup

Register keyword-command pairs in the "Voice Launcher" tab in Settings. Launch apps and websites using voice commands that contain registered keywords.

1. Open Settings (double-click Mira tray icon → "Settings" button)
2. Select the "Voice Launcher" tab
3. Enter keywords you'll use in voice commands (e.g., "Chrome"). Comma-separated for multiple keywords
4. Enter the executable path or URL (e.g., chrome.exe or https://mail.google.com)
5. Click "Add"
6. Click "Save" to save settings

Startup Settings

Turn on "Start with Windows" on the "Notes" tab in Settings to start Mira automatically when you sign in to Windows (on by default; the Microsoft Store version may not start automatically, which we are checking). When Mira starts, it opens no window and stays in the system tray (the setup wizard appears only the first time).

License

Mira is 100% free software. No license key or trial period required. All features are available without any restrictions.

FAQ

Q: Speech recognition accuracy is low / too many misrecognitions

Try switching to a more accurate model in the "Speech Recognition" tab in Settings.

• small (default, 466MB): Fast but lower accuracy
• large-v3-turbo (1.6GB, recommended): Fast + accurate. Best balance
• large-v3 (2.9GB): Highest accuracy but slow. For high-spec PCs

Q: Mira responds even when I didn't call it

Short wake words can be accidentally triggered by ambient noise or TV audio. Try these solutions:

• Use a longer wake word: "Mira" → "Hey Mira please", "Mira question" — 5+ syllables recommended
• Longer stop phrase too: "Stop" → "Mira stop please" to prevent false stops
• If your mic and speakers are close, Mira's own voice may feed back. Using a headset can help

Q: Web Watch isn't working properly

Check the following:

• Try different keywords: Change to words that are reliably recognized. E.g., if "news" is misrecognized, try "latest news"
• Register multiple keywords: Use comma-separated keywords like "news, latest news" to increase match rate
• Verify the URL: Open the URL directly in a browser to make sure it loads content
• Check AI provider is configured: The summary feature requires an AI provider

Q: Voice Launcher doesn't launch apps

Check the following:

• Is the keyword recognized correctly?: Check the actual recognized text in the "STT result" lines of the log file (%LOCALAPPDATA%\Mira\diag.log)
• Is the command path correct?: Use a full path or a command that's in your system PATH
• Include launch verbs like "open", "launch", "start" along with the keyword

Q: Screen Vision doesn't work

You need a Vision-capable cloud AI (in v2.3.1, Screen Vision does not work with local AI). Set up one of the following:

• gemini-flash-latest (recommended, free tier available)
• Claude Haiku 5.5 / Sonnet 5.5 / Opus 5.5 (high accuracy)

Q: Mira asks me to set my location when I ask about weather

Enter your location in the "Default Weather Location" field on the "Notes" tab in Settings (e.g., "Tokyo", "Osaka", "New York"). When you say "What's the weather?", this location's weather will be returned. If empty, Mira will ask you to set it.

Q: AI responses are slow

Try the following:

• Use a cloud API: Gemini Flash has a free tier and is fast (daily limits vary by model)
• For local: Switch to a smaller model (e.g., gemma3:4b) or enable GPU acceleration
• Lemonade (AMD NPU): Fast inference via Lemonade on AMD Ryzen AI PCs

Q: Can I use Mira without internet?

Yes. With Ollama (AI) + Whisper (STT) + AivisSpeech/VOICEVOX (TTS), conversations stay on your PC. The speech recognition model must be downloaded once. Weather, web search, Screen Vision and Live mode use the internet. If your local AI is not running when Mira starts (or when you save Settings), Mira switches to a cloud AI if a cloud API key is available, including one set in an environment variable.

Q: What data does Mira send?

With a cloud AI, Mira sends your transcribed text, the recent conversation and what the AI uses to answer (memory, schedule, search results, etc.) to that AI. With a local AI, your conversations are not sent externally. However, if your local AI is not running when Mira starts (or when you save Settings) and a cloud API key is available (including one set in an environment variable), Mira switches to that cloud AI and sends your conversations there until you restart Mira or save Settings again (the switch is not shown on screen). Web search, weather, Edge TTS, Screen Vision and Live mode also send data to outside services. See Data Flow on the home page for a summary, and our Privacy Policy for details.

Q: Do I need to install VOICEVOX / AivisSpeech separately?

Yes. These are separate software. They must be installed and running. Without them, Mira speaks with Edge TTS (requires Python edge-tts) or the built-in Windows voice (System.Speech).

Q: Are there any free AI providers?

Yes. The following are available for free:

• Gemini Flash: Free tier available; daily limits vary by model and may change (API key setup guided in first-run wizard)
• Ollama: Completely free local AI. Just install on your PC
• LM Studio: Local AI similar to Ollama. Easy GUI interface

Q: How to disable the memory feature

Say "Forget everything" or "Clear memory" to delete all saved memories. There's no setting to disable the memory feature entirely, but you can delete individual memories by saying "Forget [topic]".

Troubleshooting

Wake word not responding

First, check that the right microphone is selected. This is by far the most common cause when Mira does not respond at all.

1. Open Settings → "Wake Word" tab
2. Select the microphone you actually use under "Microphone"
3. Press "Test microphone", speak at your normal distance, and check that the average clears the required value
4. If it falls short, raise "Microphone boost" or move closer (see Microphone Setup)

If that does not help:

• Check if microphone is muted
• Check Windows microphone permission (Settings → Privacy → Microphone)
• Check whether Mira is paused (grey tray icon). It is orange when running and blue while listening

No TTS audio output

• Check if speakers are muted
• Check if AivisSpeech / VOICEVOX is running
• Check the TTS engine on the "Text-to-Speech" tab in Settings

AI not responding

• Check if AI provider is correctly configured in Settings
• For Ollama: run ollama list to verify model is installed
• For cloud providers: verify API key is correct

Support

For questions or bug reports, please contact support@mojosoft.biz.