Open-source voice-to-text for the terminal, OSTT addresses the need for seamless speech-to-text integration within command-line workflows. It empowers users, particularly those who rely heavily on the terminal, to convert spoken words into text that can be piped to AI, shell commands, or standard output. The primary benefit is enabling a fluid, text-based interaction model for terminal users, eliminating the friction of context switching.

How It Works

OSTT operates as a terminal-native application, capturing audio via global hotkeys. It offers a flexible transcription backend, supporting both local, offline Whisper-compatible models (with GPU acceleration via Metal on macOS, and CUDA/Vulkan on Linux) and various cloud-based providers like OpenAI, Deepgram, and Groq. After transcription, results can be directed to the clipboard, files, stdout, or processed further via AI prompts or shell commands, integrating voice input directly into existing terminal toolchains.

Quick Start & Requirements

Installation is straightforward via a curl script: curl -fsSL https://ostt.ai/install | bash. The installer handles runtime dependencies. For development, Rust and Cargo are required. Full documentation and setup guides are available at https://ostt.ai.

Highlighted Details

• Supports Linux (with desktop environment integration) and macOS.
• Configurable global hotkeys for recording and workflow initiation (e.g., Alt+Space).
• Extensive provider choice: OpenAI, Deepgram, Groq, AssemblyAI, Mistral, ElevenLabs, Berget, and local Whisper models.
• Local transcription supports offline use and GPU acceleration.
• Scriptable post-processing actions transform transcripts using AI or bash commands.
• Ability to retry transcriptions with different models or providers without re-recording.

Maintenance & Community

The project lists Kristofer, Claw Pastilhas, kristofernoaccess, and axo bot as contributors. No specific community links (like Discord or Slack) were found in the provided README snippet.

Licensing & Compatibility

OSTT is released under the MIT license, which is permissive and generally compatible with commercial use and closed-source applications.

Limitations & Caveats

While macOS is supported, the project emphasizes "Linux-first" development, suggesting potential differences in feature parity or stability. Users must manage their own API keys for cloud transcription services. Windows support is not mentioned. The project notes that deprecated configuration formats will "fail loudly," indicating potential breaking changes during updates for users with older configurations.