Why OSTT?
OSTT exists for developers and power users who want speech-to-text to behave like the rest of their toolchain: transparent, scriptable, provider-agnostic, and comfortable on Linux.
Many dictation apps are polished GUI products built around a proprietary cloud, a subscription, or a Mac-first workflow. OSTT takes a smaller approach. It is open source and terminal-native, runs entirely offline or with your own API keys, and is designed so transcription output can move through shell pipelines, AI prompts, files, clipboards, hotkeys, and scripts.
Where OSTT Is Different
Linux Is a First-Class Target
Many popular dictation tools focus on macOS, Windows, and mobile. OSTT is built with Linux desktop workflows in mind: Wayland clipboard handling, Omarchy/Hyprland setup, GNOME and KDE guides, AUR packaging, native Linux packages, and a popup launcher that works from global hotkeys.
Local Models on Your Own Hardware
Parakeet and Whisper models run on OSTT's built-in engine, with full GPU support through the Metal, CUDA, and Vulkan builds, or on the CPU. Once a model is downloaded, audio never leaves your machine. Daemon mode keeps the model loaded between recordings, so dictation does not wait for it to load.
Provider Choice
OSTT does not force one transcription backend. Alongside local models, you can choose from OpenAI, Deepgram, Groq, DeepInfra, AssemblyAI, Berget, ElevenLabs, and Mistral in the cloud, or connect your own engine, then switch models when your needs change.
This matters for quality, latency, price, language support, and regulatory requirements. For example, Berget provides Swedish/EU-hosted transcription options for users who care about data residency.
Retry With Another Model
OSTT stores recordings locally so you can re-transcribe the same audio later.
ostt retry
ostt retry 3 -cIf one model gets a term wrong, retry the same recording with another one, for example ostt retry -m deepgram/nova-3, without speaking again.
Scriptable Post-Processing
OSTT can run processing actions after transcription. AI actions work like configurable dictation modes in GUI tools: transcription goes into a prompt, processed text comes out.
Bash actions are the unusual part. They let you pipe transcribed text through any command or script, so processing can be as simple as text cleanup or as specific as your own local workflow.
Terminal-Native Workflow
OSTT treats stdout, files, clipboard output, command aliases, logs, shell completions, and pipelines as core features rather than afterthoughts.
ostt | grep invoice
ostt -p clean -c
ostt transcribe meeting.mp3 -p summary -o summary.txt
ostt process 2 cmdFor users who live in the terminal, this is the main idea.
Comparison
This table is a high-level comparison against common voice-to-text tools. Capabilities change over time, so treat it as positioning rather than a permanent claim.
| Capability | OSTT | Typical GUI Dictation Apps |
|---|---|---|
| Linux-first setup | Yes | Usually no |
| macOS support | Yes | Usually yes |
| Open source | Yes | Usually no |
| BYO API key | Yes | Sometimes |
| Multiple transcription providers | Yes | Often one provider or model family |
| Terminal-native workflow | Yes | No |
| Global hotkey popup | Yes | Yes |
| File transcription | Yes | Sometimes |
| Local transcription history | Yes | Usually yes |
| Retry same recording with another model | Yes | Rare |
| AI post-processing | Yes | Often yes |
| Bash/shell post-processing | Yes | No |
| Context/screen awareness | Yes, per app and opt-in, on Omarchy | Often yes, sometimes paid |
| Streaming transcription | No | Often yes |
| Paste into the focused app | Yes | Often yes |
| On-device transcription | Yes | Sometimes |
What OSTT Does Not Try To Be
OSTT is not trying to be a mobile dictation app or a real-time streaming transcription overlay. Those are valid tools, but they pull in a different direction from a terminal-native, scriptable tool. Screen context is opt-in: with the Omarchy dictation plugin, only the apps you choose send a screenshot to your agent.
OSTT optimizes for control, composability, provider choice, and developer workflows. If you want voice input that behaves like a command-line tool, OSTT is built for that niche.
Compared With Other Open Source Dictation Tools
OSTT overlaps with dictation tools such as Voxtype, Hyprwhspr and Superwhisper, which Omarchy also offers. What OSTT adds is dictation that takes instructions, carried out by the AI agent you already use, with optional screen context per app, plus provider switching, file transcription, retry, stdout, and shell processing.
For a more detailed comparison, see: