Offline local models
Offline dictation with local, open-source AI models
TapWisper can run entirely on your computer with open-source models. It bundles whisper.cpp for speech-to-text and llama.cpp for AI cleanup, offers 18 speech models (Whisper, NVIDIA Parakeet, Qwen3-ASR, Voxtral) and 24 language models (Qwen, Gemma, Llama, Phi-4, Mistral, gpt-oss), and works offline with no API key.
How it works
Three steps. That’s it.
01.
Choose Local
In onboarding or Settings, pick Local for speech, for language models, or both.
02.
Download a model
Pick a model rated for your computer. Downloads come from Hugging Face, are SHA-256 verified and resume if interrupted.
03.
Dictate offline
Everything runs on your machine; nothing is sent anywhere.
Details
What you should know.
- GPU accelerated
- Metal on Apple Silicon, Vulkan on Windows and Linux (NVIDIA, AMD, Intel Arc), with a CPU fallback.
- Rated for your computer
- Each model is labelled “Recommended”, “May be slow”, “Needs N GB RAM” or “Not for your languages” based on your machine.
- Speech models
- Whisper from Tiny to Large v3 Turbo (99 languages, including English-only variants), NVIDIA Parakeet TDT 0.6B v3, Qwen3-ASR 0.6B and 1.7B, and Voxtral Mini 3B.
- Language models
- Qwen3.5 and Qwen3, Gemma 4 and 3, Llama 3.1 and 3.2, Phi-4 Mini, Mistral 7B, LFM2, SmolLM, Granite 3.3, Nemotron 3 Nano and gpt-oss 20B.
FAQ
Good questions.
Can TapWisper work without internet?
Yes. With local speech and language models selected, dictation works fully offline.
Is local transcription free?
Yes. Local models are free to download and run; there's no API key or usage cost.
Do local models leave the computer?
No. The local servers listen on 127.0.0.1 only and no audio or text is sent anywhere.
More features
Everything else in the app.
AI voice dictation
Hold a key, speak, and TapWisper types clean, punctuated text into any app — Slack, Gmail, Notion, Cursor and more. Free and open-source for Mac and Windows.
Learn moreEdit by voice
Select text, hold your key and say “make it shorter” or “translate to Spanish”. TapWisper rewrites the selection in place with AI, in any app.
Learn moreAI meeting notes
TapWisper records your mic and meeting audio on your computer for a live transcript and an AI summary with decisions and next steps. No bot joins the call.
Learn moreCloud & self-hosted models
Bring your own API key for OpenAI, Anthropic Claude, Google Gemini, Together AI and Soniox — or connect Ollama, LM Studio or vLLM. 46 cloud models supported.
Learn moreSmart formatting & tone
TapWisper removes filler words, fixes punctuation and adapts tone by app: casual in WhatsApp and Slack, formal in email. You choose the tone for each.
Learn moreCustom dictionary
Teach TapWisper your name, company, product names and technical terms, plus what they're often misheard as, so dictation spells them right every time.
Learn moreDownload · v1.2.0
Start talking in a minute.
Free and open-source, no account. Setup walks you through permissions, your microphone and a speech engine — then you just talk.