Omarchy Dictation ​

On Omarchy 4.5 or later, the OSTT dictation plugin makes OSTT the program behind Omarchy's own dictation keys. Hold a key and speak; OSTT transcribes with the model you chose, local or cloud, and pastes the text into the focused app. No terminal opens. The plugin installs OSTT for you, so you do not need the installer first.

In apps you choose, dictation can also take instructions. Say "reply saying I can't make it" in Slack, and an AI agent you already use writes the reply. See Context-Aware Dictation.

Install ​

Add the plugin:

omarchy plugin add https://github.com/kristoferlund/ostt-omarchy-dictate.git --enable

Then click the microphone that appears in the bar. The setup opens in a terminal and:

• installs OSTT from the AUR if it is missing: the CUDA build (ostt-cuda-bin) on an NVIDIA GPU with Arch's cuda 13 already installed, the Vulkan build (ostt-vulkan-bin) where Vulkan works, and ostt-bin otherwise, plus ffmpeg, python, wtype and grim;
• asks you to choose a transcription model if none is selected, and for a local model offers to keep it loaded with the daemon so dictation starts instantly;
• offers context-aware dictation;
• makes OSTT Omarchy's dictation backend.

Afterwards OSTT is listed in Setup > Defaults > Dictation. Selecting it there runs the setup again, to finish one you stopped or to change your answers.

On an NVIDIA GPU, ostt-cuda-bin is the faster build for local models. The setup picks it only when cuda is already installed, because CUDA is several GB. See GPU Acceleration.

Dictate ​

OSTT records from the default input device; pick it in Omarchy's audio settings. While you speak, a pill at the bottom of the screen shows your level between two buttons: ✕ cancels, and ✓ stops and transcribes. The pill never takes keyboard focus, so the text lands in the app you were typing in.

The microphone in the bar shows the state:

Left-click starts dictation, or stops it and pastes. Right-click cancels a recording or a pending transcription, and otherwise opens the per-app settings.

Music and videos that are playing pause while you record, and play on when you stop or cancel; players you had paused stay paused. To turn this off, use the switch at the bottom of the per-app settings, or put pause_media = false at the top of ~/.config/ostt/dictate.toml. It works with players that support MPRIS, which most do.

Dictated text is saved to ostt history. If a transcription fails, the recording is kept and the notification shows how to retry it.

Context-Aware Dictation ​

In apps you choose, dictation goes through a processing action that runs an AI agent you already use, such as Claude Code, Codex or OpenCode. The agent tidies up your words and carries out spoken instructions. If you allow it for that app, the agent also gets a screenshot of the window, so it can match names, terms and tone on screen. Every other app keeps plain transcription.

The agent runs with its tools switched off, so text on screen cannot make it run commands or touch files. ostt agent list shows the agents OSTT can run this way, which are installed, and which take images.

The setup adds two actions to ~/.config/ostt/ostt.toml, using the agent and model you pick. Edit their prompts there:

• Ghostwriter (ghostwriter) tidies up your words and carries out spoken instructions.
• Text cleanup (clean) only fixes misheard words, punctuation and filler.

If you already have an action with one of those ids, the setup uses it as it is.

Choose the Apps ​

Right-click the microphone in the bar. The card lists the apps on the visible workspaces, and any already set up. Expand a row to choose one of your processing actions for that app, and whether to send a screenshot. The screenshot is off unless you turn it on, and it is offered only for actions that can use it.

The choices are saved in ~/.config/ostt/dictate.toml by Hyprland window class, which you can also edit by hand:

[apps]
"Slack" = { action = "ghostwriter", screenshot = true }
"obsidian" = { action = "clean" }

Any processing action works, including your own. A bash action gets the screenshot's path in $OSTT_ATTACHMENT.

When the Agent Fails ​

The agent takes a few seconds, and the text is pasted into whichever window is focused when it finishes. Super + Ctrl + Alt + X cancels a dictation that is taking too long.

If the agent fails or runs past its time limit (60 seconds for the actions the setup adds), OSTT pastes your words as it heard them, and a notification says why. If an app's entry is wrong or the screenshot fails, that dictation falls back to plain transcription, also with a notification.

The screenshot is a capture of the window's area of the screen, so anything on top of it, such as a notification, is included. It is kept in $XDG_RUNTIME_DIR/ostt-dictate, readable only by you, until the dictation finishes.

Update and Remove ​

Update the plugin:

omarchy plugin update io.github.kristoferlund.ostt-dictate --yes

The bar widget and the recording popup keep their old version until the shell restarts. Run omarchy restart shell, or log out and back in.

Remove it:

~/.config/omarchy/plugins/io.github.kristoferlund.ostt-dictate/uninstall
omarchy plugin remove io.github.kristoferlund.ostt-dictate

uninstall switches dictation away from OSTT and removes the plugin's scripts. OSTT itself stays installed, and so do ~/.config/ostt/dictate.toml and the ghostwriter and clean actions.

The OSTT Popup on Omarchy ​

The plugin and the installer's popup hotkeys use different keys and can be used together: the plugin for dictating into apps, and ALT+SPACE or ALT+SHIFT+SPACE for the OSTT popup, clipboard output, or a spoken prompt to your default coding agent.

The plugin's README covers the ostt-dictate command and the bar widget settings.