Speak naturally.
AI handles the rest.

Open-source voice input with LLM-powered intent recognition. Dictate, edit, and send — all by voice. Bring your own AI provider. No account, no telemetry, no lock-in.

Speaking...
9:41
Message...
ABC
Clear
Send
Speech-to-Text

Your voice, transcribed instantly

Powered by Whisper API with switchable providers. Use OpenAI Whisper, or bring your own Whisper-compatible endpoint. Multi-language support follows your system settings — Chinese, English, Japanese, Korean, and more. The language hint is passed to Whisper for significantly better accuracy.

Schedule a meeting for tomorrow at 3 PM
LLM Intent Recognition

AI understands what you mean

Every voice segment is classified — content input, edit command, or send. No mode switching, no buttons to press. Just speak naturally and the AI figures out the rest. A dedicated edit button lets you record edit-only commands with a specialized prompt for precision edits.

content "Schedule a meeting for 3 PM"
edit "Change three to four"
send "OK, send it"
Bring Your Own Provider

Works with any AI service

Use OpenAI, Claude, or any OpenAI-compatible endpoint for the LLM. Use Whisper or any Whisper-compatible API for STT. Self-host your models, use a local LLM, or connect to your company's private endpoint — ChatyInput doesn't lock you in.

OpenAI GPT-4o · Whisper
Anthropic Claude Sonnet · Haiku
Compatible Ollama · vLLM · any endpoint
Privacy First

Your voice stays yours

No accounts. No telemetry. No data collection. Audio recordings are stored locally on your device for history playback — never uploaded to any server except your chosen STT provider. API keys live locally, encrypted at rest. Fully open source — audit every line.

No account required
Zero telemetry
Keys encrypted locally
Audio stored locally only
Custom Dictionary

Teach it your words

Add names, jargon, abbreviations, and domain-specific terms. The AI prioritizes phonetically similar matches from your dictionary — so "Kubernetes" doesn't become "Cooper Nettie's" and your colleague "Aoife" is always spelled right.

1 Kubernetes k8s
2 Aoife name
3 Thinkroid company
4 PostgreSQL tech
Cross-Platform

Native on every device

Full IME on Android. Menu bar app on macOS. Standalone app on iOS. System tray on Windows. Each version is built natively for its platform — no Electron, no web wrappers, no compromises. Text injection works directly into any app you're using.

macOS
iOS
Android
Windows

Voice to any app, in four steps

01

Speak

Hold the record button and speak naturally. Dictate content in any language — hold-to-record (PTT) or click-to-toggle, your choice.

02

Process

Whisper transcribes your speech. The LLM classifies intent, auto-corrects errors, applies your custom dictionary, and updates the buffer.

03

Edit

Hold the edit button to refine. Say "change 3 to 5" or "rewrite as a formal email" — the LLM modifies the buffer while preserving your intent.

04

Send

Say "send" and the text goes directly into the active app — Slack, Gmail, VS Code, any text field. Native text input on every platform.

Get ChatyInput

macOS

Menu bar + floating panel

Coming Soon

iOS

Standalone app

Coming Soon

Android

Full IME keyboard

Download

Windows

System tray app

Coming Soon

Built in the open

MIT licensed. Every line of code is auditable. Fork it, self-host it, modify it, contribute back. No proprietary cloud, no subscription, no strings attached.

MIT
License
4
Platforms
3
Languages