99languages
Multilingual foundation
Whisper is built for automatic speech recognition and translation to English output.
Privacy-first dictation for Mac
Dictate everything. No cloud.
KimiTalk is the private dictation app for your Mac.
No cloud upload for recognition. Recognition runs locally.
Currently v0.9.19 · See all changes →
Time saved calculator
Speaking can save time once the text gets longer.
Measure or estimate your typing speed and see how much time speech input can save compared with desktop keyboard entry. The baseline values are backed by published keyboard and speech-rate research.
Your typing speed
Estimate it, enter it, or measure it in a short test.Type freely for 30 seconds. The estimate uses the common WPM standard: 5 characters count as 1 word.
30 sec
Ready for a short desktop test.
Workflow presets
Baseline: 52 words/min typing + 150 words/min speech
Estimated saved time
2.1 hrs saved per weekThe longer the text gets, the more speaking starts to pull ahead of typing.
Weekly time needed
Typing takes 194 min
Typing
Speech
time saved
1:07 hrs
Your advantage
2.9× faster Speech compared with your typingWorks anywhere you can type.
Gmail
Notion
Proton Mail
Standard Notes
Tuta
Signal
Obsidian
Tresorit
Cryptomator
Slack
VS Code
Linear
Cursor
Claude
ChatGPT
Write anywhere
Dictate anywhere you can type.
KimiTalk places text into the active text field: Mail, Notes, browsers, Slack, Notion, VS Code, ChatGPT, Claude, Cursor, and similar Mac apps. AI tools are destination apps, not a requirement. KimiTalk does not upload raw audio for recognition; text you paste or send in cloud apps follows the rules of that destination app.
Active text field
You say
Hi everyone thanks for the update next let's clarify the open points review the changes and agree on the release date
KimiTalk writes
Hi everyone, thanks for the update. Next, let's clarify the open points, review the changes, and agree on the release date.
KimiTalk works in the background
If you can type there, you can dictate there.
System-wide hotkey: dictate directly into the active app.
Multilingual in one dictation: German, English, mixed.
Pro shapes long rough dictation locally into clean text for different destination apps.
Why it matters
Privacy starts before the first word.
Focus
Trust The study by Vimalkumar et al. shows that trust helps decide whether people use speech systems. KimiTalk strengthens that trust because recognition runs locally on your Mac.
Local recognition instead of external processing
No wake word, no always-listening assistant
Clear boundary: voice, Mac, text field
Trust through fewer data paths
System recommendation
What do you want to do with KimiTalk?
Dictation is always the base. Once you add rewriting or translation, your Mac needs more reserve because Gemma 4 runs locally alongside Whisper.
- DictateWhisper runs locally, 16 GB unified memory is the baseline.
- RewriteAdd Gemma 4 for local text work, 24 GB recommended.
- TranslateWhisper plus Gemma 4 with more reserve, 32 GB ideal.
KimiTalk chooses the right model locally on your Mac when the app starts. This website does not probe your hardware automatically and stores nothing here.
Your needs
16 GB minimum
What do you want to do?
Choose what you want to add to local dictation.Recommended baseline
16 GB unified memory
For local dictation with Whisper. A good fit if you mainly want to turn speech directly into text.
Whisper
Turbo
Gemma 4
not needed
Memory
16 GB
Dictation
16 GB minimum
Whisper local
Dictation always stays local. Gemma 4 is added only when you use local text features or translation.
Model basis
How your voice becomes text locally.
The app chooses the variant that fits your device, so recognition can stay local and smooth.
Voice
raw audio
Whisper
local on Mac
Text
dictation or English
32→4decoder layers
Turbo for speed
Large-v3-Turbo greatly reduces the decoder stack, making it optimized for fast local transcription.
16 GBrecommended
Everyday Mac use
For smooth dictation while other apps are running, we recommend Apple silicon with 16 GB RAM or more.
Sources: OpenAI Whisper, large-v3-turbo model card.
Basic vs. Pro
Choose the tier that fits your workflow.
Basic covers local dictation and file transcription. Pro adds history, speaker-aware transcripts, long-form capture, and local tools for shaping text after you speak.
Basic
Dictate locally.
For users who want private speech-to-text, app switching, mixed-language dictation, file transcription, and cleanup.
€24.90 lifetime licensePro
Dictate, rewrite, translate.
For users who turn spoken thoughts and long-form audio into structured transcripts, projects, documents, messages, and prompts when needed.
from €6.99/monthCore dictation
Local dictation on Mac
Included
Included
Hotkey recording across apps
Included
Included
Dictate while switching apps
Included
Included
Menu bar app and floating menu
Included
Included
German + English in one dictation
Included
Included
Mixed-language output as spoken
Included
Included
Whisper model choice and ANE/GPU routing
Included
Included
Cleanup style
Included
Included
Basic power tools
Custom vocabulary
Included
Included
Smart punctuation
Included
Included
Filler removal
Included
Included
Privacy
WhisperKit/CoreML transcription
Included
Included
No dictation backend
Included
Included
No ad tracking, no content analytics
Included
Included
Local-first processing
Included
Included
Raw audio stays local
Included
Included
File transcription and export
File transcription
Included
Included
Plain text and simple SRT export
Included
Included
VTT / JSON export
Included
Included
CLI transcription tool
Included
Included
Speaker recognition
Not included
Included
Speaker renaming and aliases
Not included
Included
Local text tools
Local text engine
Not included
Included
Advanced rewrite styles
Not included
Included
Email, note, and document shaping
Not included
Included
Local translation*
Not included
Included
Summarization
Not included
Included
Mixed-language input to English output*
Not included
Included
Custom instructions and presets
Not included
Included
Workflow memory
History for dictations and files
Not included
Included
Projects for organizing history
Not included
Included
Deferred paste with preview editing
Not included
Included
Multi-burst sessions
Not included
Included
Transcript chat
Not included
Included
Voice commands and local structuring
Not included
Included
Long-form capture
Meeting capture
Not included
Included
Batch processing
Not included
Included
Translation to English output requires a Large-v3 model with Whisper Translate support, such as quantized Large-v3 (~626 MB). Turbo models are optimized for fast transcription and do not support this Translate mode.
Buying options
Buy or subscribe.
Checkout is not live yet. These buttons will redirect to Lemon Squeezy once payments are available.
Basic
€24.90 lifetime license
Local dictation, cleanup style, and updates until the next major version.
Pro subscription
€6.99 /month
Basic plus history, speaker-aware transcripts, long-form capture, local text tools, and all future Pro features.
Pro lifetime license
€99 one-time
Pay once and keep Pro. Usually pays for itself after roughly two years.
Good to know
Good to know before you start.
The essentials about trial, Pro, licensing, and local work.
Yes. After onboarding, you can try KimiTalk with 1,000 free words. Your settings remain available afterwards; you can keep dictating as soon as you activate a license.
Pro is for rough dictation, files, and longer recordings that should become cleaner, more structured, or better suited to their context. It includes history, speaker-aware transcripts, long-form capture, local text tools, and future Pro features.
One license covers up to three Macs. KimiTalk validates signed license tokens locally and only refreshes them occasionally online.
Recognition runs on your Mac with the downloaded model. Raw audio is not uploaded for recognition. If you paste or send text into cloud apps like ChatGPT, Claude, or Cursor, that text follows the rules of the destination app.