https://github.com/1vecera/Mluva
Native dictation and rewriting for Linux. At home on Omarchy, with editable drafts and independent providers.
https://github.com/1vecera/Mluva
dictation gtk4 linux omarchy quickshell speech-to-text
Last synced: about 1 month ago
JSON representation
Native dictation and rewriting for Linux. At home on Omarchy, with editable drafts and independent providers.
- Host: GitHub
- URL: https://github.com/1vecera/Mluva
- Owner: 1vecera
- License: apache-2.0
- Created: 2026-08-21T12:36:03.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-10-09T01:09:48.000Z (about 1 month ago)
- Last Synced: 2026-10-09T02:35:59.581Z (about 1 month ago)
- Topics: dictation, gtk4, linux, omarchy, quickshell, speech-to-text
- Language: Rust
- Homepage: https://1vecera.github.io/Mluva/
- Size: 359 MB
- Stars: 5
- Watchers: 0
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Security: SECURITY.md
- Agents: AGENTS.md
- Claude: CLAUDE.md
Awesome Lists containing this project
- awesome-gtk - Mluva - Native dictation and rewrite workspace for Omarchy, with editable transcripts and searchable history `#python` `#gtk4` `#libadwaita`. (Audio / Audio Players)
README

**The most delightful dictation for Omarchy.**
Speak freely, turn the result into a useful draft, and keep every original word. Mluva brings recording, editable rewrites and searchable history into a quiet native workspace that follows your desktop theme.
**[Install](#install)** · **[Watch the intro](#meet-mluva)** · **[Choose your providers](docs/provider-selection.md)** · **[Release notes](https://github.com/1vecera/Mluva/releases/latest)**
## Meet Mluva
https://github.com/user-attachments/assets/ffac3582-3146-46f2-9244-2c28c5cbef26
72 seconds · Narrated by Daniel Vecera · Captions included. Video credits and demo notes.
## From a thought to a finished draft
| Start with… | Make it useful |
| --- | --- |
| A rough idea | Dictate with F9. Scribe streams words while you speak; completed text copies to the clipboard. |
| Text you already have | Paste into a conversation, then choose **Polish**, **Structure** or your own instruction. |
| A note that needs shape | Select **Live rewrite** once or continuously across recordings. Grilling keeps useful questions above evolving notes; task spec, structured note and custom templates remain available. |
| A draft worth keeping | Edit originals and rewrites, save with **Ctrl+S**, and keep raw recognition available separately. |
| Work to return to | Search the sidebar, click a title to rename it, or use the conversation menu to merge or delete chats. Reuse saved prompts and export Markdown or JSON. |
| Something on screen | Attach an Omarchy region screenshot with F10, then use **Add narration** in the optional upgraded Tensaku editor to place a spoken text box. Supported rewrite models can use the image as context. |
**A calm place to work.** Stable Live panes, restrained recording motion and subtle Markdown keep long notes readable. **Ctrl+P** finds actions; **Ctrl+Enter** sends a rewrite. Copy and scrolling preferences are yours to change.
**Your choice of engines.** Select speech and rewriting independently: ElevenLabs Scribe, Mluva-managed local speech models or a compatible transcription API; native Codex app-server or a LiteLLM/OpenAI-compatible service for rewriting. Availability, credentials and models are configured in Settings.
Live rewrite is opt-in and Experimental. It marks missing information, preserves manual edits and reconciles the draft with the final transcript when recording stops. [Explore the features and their limits →](docs/feature-story.md)
## Install
Mluva uses [native Rust executables](rust/README.md), with no Python runtime. Download the prepared Omarchy x86_64 bundle for an installation without a compiler. Source builds require the pinned Rust toolchain and a C compiler before setup provisions the remaining [development prerequisites](CONTRIBUTING.md#local-setup). See the [cutover verification and remaining acceptance limits](docs/verification/rust-cutover/README.md).
Run the setup from a source checkout or extracted [release archive](https://github.com/1vecera/Mluva/releases/latest). On **Omarchy Quattro**, it installs the desktop dependencies, native application and bundled shell widget together. Both parts share one repository, version and package. It shows the installation plan first; system packages may request your sudo password.
```sh
git clone --depth 1 https://github.com/1vecera/Mluva.git mluva
cd mluva
bash install.sh
```
Launch **Mluva** from the application menu, then choose your [speech and rewrite providers](docs/provider-selection.md). Cloud providers need an account and credentials; local recognition downloads its own model, with CPU support by default and optional NVIDIA GPU acceleration. Provider setup is separate from installing the app.
### Install with an agent
Copy this prompt into your coding agent:
```text
Install Mluva from https://github.com/1vecera/Mluva on this computer.
Use the repository's install.sh to install desktop dependencies and the native
app, including the mluva.dictation plugin on Omarchy Quattro. Inspect the setup
script, then run it with --yes. Preserve existing settings, conversations and
plugin customizations. If Mluva is running, ask me to quit it before upgrading.
Help me configure speech and rewriting with my chosen providers, installing
any required provider client and using existing credentials without revealing
them. Tell me how to launch Mluva and approve its recording shortcuts.
```
**Other install paths:** `bash install.sh --app-only` installs the native app without changing plugins. Fedora GNOME uses the same setup command, with the compatibility limits below. For a staged install or preinstalled dependencies, see the [Linux guide](linux/README.md#install-for-the-current-user).
**Narrated screenshots:** the release includes an optional prebuilt Tensaku editor for Omarchy x86_64. Verify and extract it, then run `bash linux/install-narrated-editor.sh --prebuilt-dir /absolute/path/to/mluva-2.2.2-tensaku-omarchy-x86_64` from the Mluva checkout or extracted source package. A source build remains available. Configure F10 as described in the [screenshot setup](docs/omarchy-integration.md#screenshot-context). Screenshot context is Experimental.
**Upgrading:** quit Mluva, update the checkout with `git pull --ff-only`, and rerun `bash install.sh`. Existing plugin customizations are preserved; setup checks for conflicting changes before installing. If a later plugin operation fails, the native app remains installed and setup reports how to retry. Upgrades from 0.x migrate the old product identities and retain a private backup of settings, conversations, drafts and audio. See the [migration guide](docs/identity-migration.md) for customized installations.
### Install through Omarchy's plugin manager
This same repository is also an Omarchy plugin. Its root `manifest.json` loads the bundled QML directly; there is no separate plugin repository or website. Omarchy's plugin manager installs the widget only. Install the native app and desktop dependencies explicitly from that same checkout:
```sh
omarchy plugin add https://github.com/1vecera/Mluva.git --enable
bash "$HOME/.config/omarchy/plugins/mluva.dictation/install.sh" --app-only
```
Launch Mluva from the application menu and choose your providers. The widget requires its `mluva-shell` bridge and does not install or launch the app itself. Quit Mluva before upgrading, run `omarchy plugin update mluva.dictation`, then rerun the `--app-only` command above so the app matches the updated widget. Keep using the combined setup if you already installed that way; Omarchy refuses duplicate `mluva.dictation` installations. See the [Omarchy guide](docs/omarchy-integration.md#installation) for both update paths and customization protection.
### Remove Mluva
Remove the widget before uninstalling the app. Both commands preserve Mluva's settings and saved conversations:
```sh
omarchy plugin remove mluva.dictation
mluva-uninstall
```
If you installed only the app, run just `mluva-uninstall`. The [desktop requirements](docs/linux-platform-profile.md), [provider guide](docs/provider-selection.md) and [third-party notices](THIRD_PARTY_NOTICES.md) document runtime dependencies, optional accounts/models and licenses.
### The Omarchy widget
The floating widget starts with five preview lines while you speak. A bare light sits at top left and the timer at top right. Drag the text box or status row to move it, choose a lower-left/bottom/lower-right preset, or focus it and press **Super+T** on Omarchy to tile it. Floating mode keeps it above other windows; resizing gives the preview more room. The main app uses the same monospace text and bare header above its content. When you finish, **Continue** adds speech to the same conversation; you can also rewrite, copy or open it in the workspace. The app offers **Continue recording** above the conversation. Hover, deliberate interaction, menus and active rewrites pause the widget's configurable four-second dismissal; inherited focus alone does not. **Shift+F9** reopens the latest conversation when configured.

Closing the main window keeps Mluva available; quit from the shell or application menu. The [Omarchy guide](docs/omarchy-integration.md) covers dependencies, existing manual installs and removal.
## Platform support
| Platform | Status |
| --- | --- |
| **Omarchy · Hyprland · Quickshell** | **Primary, maintained platform.** The dictation, rewrite and widget workflow is tested end to end and used daily by the maintainer. |
| Fedora · GNOME · Wayland | Compatibility is retained, but has not been tested for several releases. The last desktop acceptance used Fedora 44 and GNOME 50. |
Individual advanced features, including Live rewrite, Meeting mode and alternative provider routes, remain Experimental where indicated. Automatic insertion is off by default; clipboard delivery is the standard workflow. See the [feature status](docs/feature-maturity.md) and [desktop requirements](docs/linux-platform-profile.md).
## Your text and the cloud
- **Speech and rewriting:** your selected providers determine where processing happens. Local Whisper keeps recognition local; cloud speech sends audio to that provider. Rewriting sends the selected text to your rewrite provider.
- **Local history:** originals, completed rewrites and titles stay together in a local SQLite database. Rename, export or delete conversations. Automatic titles send up to 6,000 characters from each new conversation to the rewrite provider; disable them in Settings to use local labels.
- **Incognito:** no saved history or recovery audio, conversation rewriting or generated titles. Recognition still uses your selected speech provider; cancellation cannot recall audio already sent.
See the [product contract](docs/product-contract.md) for retention, recovery and privacy details.
## Development
Want to dictate from your phone into this PC? The optional [private browser recorder](docs/browser-recorder.md) installs on your home screen, opens straight to Record / Stop, and shares browser recordings with PC clipboard delivery through Cloudflare.
Mluva uses **Rust, GTK 4, Libadwaita and PipeWire**, with a QML plugin for Omarchy. Source, tests and runtime workers contain no Python. Start with the [code map and focused checks](CONTRIBUTING.md#code-map), [native overview](rust/README.md) or [Linux guide](linux/README.md).
```sh
make linux-test linux-shortcut-test
shellcheck install.sh linux/*.sh linux/tests/*.sh dev/*.sh linux/mluva-shell
```
Use the [isolated development runner](dev/README.md) for UI verification. Bug reports should include the platform, reproduction steps and expected behavior. Keep private recordings, transcripts and credentials out of public issues.
Mluva means *speech* or *manner of speaking* in Czech, pronounced roughly “MLOO-vah.” **[Apache License 2.0](LICENSE)** · [Third-party notices](THIRD_PARTY_NOTICES.md) · [Changelog](CHANGELOG.md)