Kartik Labhshetwar spent months watching subscription fees pile up from dictation software. Then he built an alternative and gave it away.
BetterWispr, his free dictation app for Mac, went live October 8, 2026, processing all speech recognition on the user's device without ever touching cloud servers. The open-source software arrived as subscription services like Wispr Flow—which caps free users at 2,000 words weekly before requiring payment—have turned voice-to-text into another recurring charge.
"I wanted to make voice typing more accessible without requiring a monthly subscription or sending every recording to a cloud server," Labhshetwar wrote on Hacker News after the launch.
The app operates through a global keyboard shortcut, Option-Space by default, and pastes cleaned-up text into any application. Users can toggle between hold-to-talk and press-to-toggle modes. Under the hood, BetterWispr offers three speech recognition engines that run entirely on the user's Mac: Apple's built-in framework, NVIDIA Parakeet (which handles English plus 25 European languages and Japanese), and Whisper large-v3-turbo through WhisperKit, according to documentation published on the project's GitHub repository.
Text cleanup happens at three levels of intensity. A light mode strips out English fillers, repeated words, and false starts. Medium mode can tap local AI models such as Apple Intelligence or Ollama, or connect to external services like Claude and OpenAI if the user explicitly configures an API key and accepts that text will then leave the device. The app also responds to voice commands: "comma," "question mark," "new paragraph," "scratch that."
Performance in Practice
Version 0.1.5, released October 12, 2026, introduced speaker detection for meeting transcripts. Internal tests on six two-person calls attributed 93% of words to the correct speaker, up from 85% with voice-change detection, according to release notes the project published. Four-person meetings proved trickier, with accuracy reaching 82% versus 77% baseline.
The recommended Parakeet Ultra model logged a 14.45% word error rate across 300 video-call segments before cleanup. Light cleanup nudged that to 14.95%, suggesting the filler-removal process occasionally discards legitimate words along with verbal tics.
The meeting feature records microphone input and, on macOS 14.2 or later, can capture audio from other applications separately. After a conversation ends, users can feed transcripts to Apple Intelligence or an Ollama model for summarization.
Labhshetwar, who lists his current role as Member of Technical Staff at Mem0 on his personal site betterwispr.com, released BetterWispr under the Apache 2.0 license through the opennook GitHub organization. As of October 14, 2026, the repository had gathered 318 stars and 20 forks.
A Crowded Field
BetterWispr enters a market where privacy trade-offs shape product design. Wispr Flow, its most direct competitor, routes dictation through cloud servers but offers cross-platform support across Mac, Windows, iOS, and Android, backed by SOC 2 Type II and ISO 27001 certifications. Superwhisper blends local and cloud models with paid tiers. MacWhisper charges a one-time fee for its Pro version. Two other open-source projects, FluidVoice and OpenWhispr, occupy similar territory with different licensing and feature sets.
The tension between convenience and control runs through user reactions. A Show HN post on Hacker News drew 218 points and 126 comments within three days of the October 11 announcement. "So far I dictated about 5 different things… it worked without a single error! Good choice to use parakeet v3," HN user GodelNumbering reported.
The app requires macOS 14 or later and runs best on Apple Silicon, though Intel builds are available through GitHub Releases. No account creation. No data collection unless you wire it up yourself. Audio stays local by default, which matters to some users more than the polish of cloud transcription—or perhaps more than Labhshetwar initially expected, given the early response.
