DeepSeek Harness Remote
DeepSeek Harness Remote 可將手機或另一台電腦與你運行 DeepSeek Harness 的電腦安全連接,隨時隨地繼續你的工作,方便隨時查看進度並繼續工作。
AI 翻譯
尚未收集儲存庫資訊。
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
Dictation for the Mac: local speech-to-text, live preview, model matched to your Mac. Fork of VoiceInk, GPL-3.0.
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗Download Diktilo.dmg, one universal app for Apple Silicon and Intel Macs.
.dmg and drag Diktilo into Applications.Requires macOS 14.4 Sonoma or later. Local models run best on Apple Silicon.
After installing, Diktilo downloads a speech recognition model to your disk. You download it once; from then on it runs on the Mac itself. The recording never leaves the computer and dictation works without internet.
| Where the recording goes | |
|---|---|
| Local model (default) | Nowhere. Speech becomes text on your own processor. |
| API key (only if you enter one) | To the provider you chose (Groq, OpenAI, ElevenLabs, Deepgram, Mistral, Gemini…), which sends back text. You pay the provider and need internet. |
| Your own server | Only to it, e.g. speaches on your VPS, with live streaming over the OpenAI Realtime API. |
An API key is your ID with an outside provider: you open an account there, copy the key and paste it into Diktilo. AI enhancement of the text (punctuation, rewording) follows the same rule: locally through Ollama, or with a provider's key, in which case only the text is sent, never the recording.
On first launch Diktilo reads the chip, memory, macOS version and the languages you dictate in, and suggests a model:
| Your Mac and languages | Suggested model |
|---|---|
| Apple Silicon, English only | Parakeet V2 (V3 as an alternative) |
| Apple Silicon, European languages (Polish, German, French, Ukrainian and 21 more) | Parakeet V3: fast, light on memory, transcribes live, recognises which language you speak |
| Apple Silicon, other languages, 8 GB RAM or more | Whisper Large v3 Turbo |
| Apple Silicon, other languages, less than 8 GB | Whisper Base |
| Apple Silicon on macOS 26, one of de/en/es/fr/it/ja/ko/pt/zh | Apple Speech as an alternative, built into the system |
| Intel | a small Whisper model, with a hint that a cloud model will be faster |
You can always pick another model by hand.
While you hold the key, a small recorder shows the waveform and the text as it is being written. Mini (the default) is a bar at the bottom of the screen; Notch slides out of the MacBook's screen notch. Below it, a strip shows the shortcuts that finish the recording and send the text somewhere specific.
If AI enhancement is on, Diktilo starts working on it while you are still speaking, so the corrected text is ready sooner when you let go.
More destinations are off by default and are turned on in Shortcuts and Modes (Save Targets, Advanced outputs):
⌘↩ finishes the recording in the current mode; once Modes are turned on, ⌥1…⌥0 switch them.
Diktilo is in beta. If something breaks, behaves oddly or is missing, tell me:
make local builds an unsigned app for your own use (needs Xcode); see BUILDING.md.
Diktilo is a fork of VoiceInk by Pax, licensed under the GNU General Public License v3.0 (see LICENSE). It builds on whisper.cpp, FluidAudio (Parakeet), KeyboardShortcuts, LaunchAtLogin, MediaRemoteAdapter, Zip, SelectedTextKit and Swift Atomics.