Core architecture
A native macOS menu bar utility built around local Whisper models (whisper.cpp) with optional external API hooks for cloud LLM post-processing.
Comparison
Local transcription done well, with no knowledge layer behind it and cloud API keys waiting the moment you want real cleanup.
A native macOS menu bar utility built around local Whisper models (whisper.cpp) with optional external API hooks for cloud LLM post-processing.
Captures system audio locally and decodes it using quantized Whisper models directly on Apple Silicon. For advanced cleanup or formatting modes, it routes the raw transcript through user-supplied OpenAI/Anthropic API keys or local small models (such as s1-mini).
Historical dictations are saved as flat, unindexed logs. Superwhisper cannot connect related notes, track project specs, or act as an external brain.
Accessing capable LLM cleanup often requires routing text through external cloud API keys, reintroducing privacy concerns and variable token bills.
Purely a text-replacement tool. It cannot query system calendars, read application state, or execute agentic tool calls.
Free tier with small local models; Pro is $8.49/month, $84.99/year, or a $249.99 one-time lifetime license.
Unlimited dictation, free forever. Plus is $6/month or $49/year; Pro is $14/month or $119/year.
See full pricingSuperwhisper is a dictation utility; Tyne is an ambient desktop operating system. Tyne couples local Whisper transcription with an on-device 1.5B/3B LLM and a hardware-encrypted SQLite vault (AES-256 via SQLCipher) that indexes your ideas across vector embeddings, BM25 text search, and an entity-attribute-value knowledge graph.