kristoferlund/ostt

What it solves

OSTT provides a terminal-native way to capture voice input and convert it into text that can be integrated into any shell workflow. It eliminates vendor lock-in by allowing users to switch between various local and cloud-based speech-to-text providers without re-recording audio.

How it works

Users record audio via a global hotkey or CLI command. The tool then transcribes the audio using either local Whisper-compatible models (with GPU acceleration via Metal, CUDA, or Vulkan) or cloud APIs (such as OpenAI, Deepgram, Groq, and Mistral). The resulting text can be sent to the clipboard, a file, stdout, or processed further through custom bash commands or AI prompts.

Who it’s for

Developers and power users on Linux and macOS who use the terminal as their primary interface and want a scriptable, open-source voice-to-text system.

Highlights

• Provider Flexibility: Support for multiple cloud APIs and local offline models via Hugging Face or direct files.
• Terminal Integration: Seamlessly pipes transcription results into stdout, files, or the clipboard.
• Scriptable Post-Processing: Ability to transform transcripts using AI prompts or shell commands.
• Non-Destructive Workflow: Saves recordings locally, allowing users to retry transcription with different models or providers.
• Keyword Boosting: Support for custom vocabulary to improve recognition of technical terms and names.