Speech to text, in your terminal

Open source voice‑to‑text
for Linux. And macOS.

Record from a hotkey, transcribe on your own machine with Parakeet or Whisper or with a cloud model, then paste the text into the app you are in, or send it to your clipboard, a file, an AI prompt or any shell command.

Free and open source MIT

Local models

Runs offline, on your own hardware.

A built-in engine runs NVIDIA Parakeet and OpenAI Whisper with full GPU support, and the installer picks the right build for your hardware. Once a model is downloaded, no audio leaves your machine. See Local Models and GPU Acceleration.

  • CUDANVIDIA
  • VulkanAMD, Intel, NVIDIA
  • MetalmacOS

# Download and select a local model$ ostt model local download parakeet-ultra$ ostt model select parakeet/parakeet-ultra# Keep it loaded between recordings$ ostt daemon start# Dictate into the focused app, fully offline$ ostt launch --paste

Workflow

From speech to useful output.

Record, transcribe with the model that fits the job, optionally process the text, and send it where it is needed.

Recordhotkey or fileTranscribelocal or cloudProcessAI or bashSendanywherepasteclipboardfilestdoutRecordhotkey or fileTranscribelocal or cloudProcessAI or bashSendanywherepasteclipboardfilestdout

Pipeline

Transcribe once. Route anywhere.

Dictate straight into the focused app, process audio files from scripts, hand the text to an AI agent, or pipe it through your own commands. Processing actions are one flag away. 

# Record, clean up, paste into the focused app$ ostt launch --paste -p clean# Transcribe a meeting, summarise it into a file$ ostt transcribe meeting.mp3 -p summary -o notes.md# Run the "cmd" action on the second latest recording$ ostt process 2 cmd

Everything else

Small tool, many uses.

Linux-first

Hotkeys for Omarchy and Hyprland, GNOME, KDE and other Linux desktops, with macOS support too.

Paste into any app

OSTT closes its popup, waits for focus to return, and pastes into the editor, browser, chat or terminal you were in.

Local or cloud models

Parakeet and Whisper on your machine, or OpenAI, Deepgram, Groq, DeepInfra, AssemblyAI, Berget, ElevenLabs and Mistral.

GPU acceleration

Metal on macOS, CUDA on NVIDIA, Vulkan on AMD and Intel. The install script picks the right build.

Daemon mode

Keep a local model loaded in the background, so every transcription starts at once.

Process with AI or bash

Run your own prompts on the agent you already use, or any shell command, before the text is sent on.

Retry without re-recording

Recordings are saved, so the same audio can be transcribed again with another model.

Keywords and replace rules

Teach the model your names and terms with ostt keyword, and fix casing with ostt replace.

File transcription in CI

Transcribe meetings, voicemails and audio files from scripts and pipelines, no microphone needed.

Terminal-native

Pipe the text through jq, sed or awk, and chain recording, processing and output in one line.

Agent-ready

Speak your prompts into OpenCode, Claude Code, Codex, Gemini CLI and other agents.

No subscription

Free and open source. The config is local, and the providers are the ones you choose.

On Omarchy

Dictation that takes instructions.

On Omarchy, OSTT becomes the system's dictation, and Ghostwriter reads your screen and writes replies with the AI agent you already use.

See OSTT on Omarchy

Try it.

One command installs OSTT, with the right build for your hardware. OSTT is free, requires no account, and is open source and MIT licensed.