Mluva 1.6.0 — Screenshot context and narrated annotations

Mluva 1.6.0 adds screenshot context and spoken text boxes, and fixes automatic paste through Omarchy shortcuts and Firefox.

• Bind F9 to mluva-shell global-record to capture the insertion target when recording starts. Bar/button recording remains clipboard-only. Firefox uses guarded keyboard paste, and an already focused field is recognized when the tracker starts.
• Bind F10 to mluva-shell screenshot to select an Omarchy region and attach it to the current narration or saved conversation. Supported rewrite models receive the ordered image pixels and recording offsets.
• In the upgraded Tensaku editor, choose Add narration, click or drag a text area, dictate a fresh phrase and press Stop narration. The text box saves into the PNG through the normal editor. Escape cancels without an annotation upload. Finish the annotation before rewriting with the image.
• Screenshots follow History retention and deletion and are unavailable in Incognito and Scratchpad. Neither attachment nor narrated annotation changes the clipboard.

The native app and Omarchy widget both report 1.6.0. The optional editor is the patched Tensaku 0.29.0, distributed with its source pin, complete patch, MPL license, authorship notice and Rust dependency notices.

Download mluva-1.6.0-source.tar.gz and SHA256SUMS. The prebuilt mluva-1.6.0-tensaku-omarchy-x86_64.tar.gz is optional and avoids compiling the editor on the tested Omarchy x86_64 platform. Verify the downloaded files, close Mluva, extract the source archive and run the combined installer:

sha256sum --ignore-missing -c SHA256SUMS
tar -xzf mluva-1.6.0-source.tar.gz
cd mluva-1.6.0
bash install.sh

To activate the prebuilt editor from the extracted source directory:

tar -xzf ../mluva-1.6.0-tensaku-omarchy-x86_64.tar.gz -C ..
bash linux/install-narrated-editor.sh --prebuilt-dir "$(cd ../mluva-1.6.0-tensaku-omarchy-x86_64 && pwd)"

The binary was tested with GTK 4.22.4 and Libadwaita 1.9.3; other platforms should use bash linux/install-narrated-editor.sh to build from the pinned source. Existing settings, saved conversations and Tensaku preferences are preserved. F9/F10 setup and upgrade details.

Validation: 631 Linux tests, Ruff, formatting, generated feature consistency, ShellCheck, plugin validation and clean archive installation passed. Patched Tensaku passed 270 application and 2 CLI tests and workspace Clippy; 10 upstream tests remain ignored. Disposable native GTK/Wayland checks covered real installed files, Omarchy capture, selected-area annotation, independent PNG OCR, cancellation, privacy and image payloads. Codex identified generated shapes and a random number from actual inline pixels. Paste evidence and screenshot evidence describe their test boundaries.

Physical F9/F10, the user's actual microphone and simultaneous real-microphone recording were not exercised. Automatic paste and screenshot context remain Experimental; unsupported image models retain the existing safe text fallback. Fedora GNOME has not been tested recently. Hosted CI is manual and was not triggered.

Changes: automatic paste, screenshot context and narrated text, release packaging.

Mluva 1.5.4 — Automatic paste fixes

Mluva 1.5.4 fixes automatic paste into Hyprland terminals and X11 applications.

• Hyprland terminals can be captured without an accessibility text field. Mluva verifies the terminal executable and requires the same window and process to remain focused immediately before paste.
• Symbolic terminal paste uses Ctrl+Shift+V, which reads the clipboard, including on default Foot where Shift+Insert reads the primary selection.
• X11 uses xclip and xdotool even when Wayland helpers are installed. Wayland keeps its own clipboard and input tools.

Text is copied before insertion. A changed target remains copy-only, and an uncertain paste never triggers a second attempt. Terminal dispatch remains unconfirmed; window capture cannot distinguish tabs, panes or shell input modes. Automatic paste remains Experimental and disabled by default; enable it in Settings → Capture when wanted.

Download mluva-1.5.4-source.tar.gz, verify it with SHA256SUMS, extract it, close Mluva and run bash install.sh in the extracted directory. The native app and Omarchy widget both report 1.5.4. Existing settings and saved conversations are preserved. Installation and upgrade details.

The main repository also supports Omarchy's plugin manager. After plugin updates, close Mluva and install the matching native app from that checkout:

bash "$HOME/.config/omarchy/plugins/mluva.dictation/install.sh" --app-only

Validated with 605 tests, Ruff, formatting, ShellCheck, private shortcut and native text-target checks, isolated Omarchy widget tests, both plugin manifests, and a clean archive installation. Real isolated targets passed exact Unicode insertion checks: GTK Entry/TextView, Chromium input/textarea/contenteditable, VS Code, Foot and Ghostty; X11 Chromium also passed. Physical F9, Fedora GNOME/Mutter, custom terminal bindings and the optional ydotool daemon were not exercised. Target evidence and limits. Hosted CI is manual and was not triggered.

The public marketplace listing still points at the retired repository. Its migration remains pending in omacom/omarchy-plugin-marketplace#9313; this release is available directly through the main repository and source archive.

Changes: automatic paste fixes, 1.5.4 release metadata.

Mluva 1.5.3 — Terminal paste fixes and one Omarchy repository

Mluva 1.5.3 fixes keyboard paste fallback and brings the native app and Omarchy plugin into one repository and release.

• Captured terminal targets use Shift+Insert: Foot, Alacritty, Ghostty, Kitty and WezTerm.
• Removed the hardcoded ydotool letter key that could send Ctrl+K on Dvorak. Ordinary apps use symbolic wtype or X11 xdotool; with only ydotool, text stays on the clipboard. Omarchy setup now includes wtype.
• The main repository now contains the root Omarchy plugin manifest. The existing plugin ID, mluva.dictation, is preserved; app and widget both report version 1.5.3.

Download mluva-1.5.3-source.tar.gz, verify it with SHA256SUMS, extract it, close Mluva and run bash install.sh in the extracted directory. Existing settings and saved conversations are preserved. Installation and upgrade details.

For installation through the Omarchy plugin manager, native-app setup is an explicit second step:

omarchy plugin add https://github.com/1vecera/Mluva.git --enable
bash "$HOME/.config/omarchy/plugins/mluva.dictation/install.sh" --app-only

After plugin-manager updates, close Mluva and repeat the --app-only command. The public marketplace listing still needs migration from the retired repository. The official marketplace scanner also rejects the bundled Mermaid renderer because it exceeds its 512 KiB per-file limit; official search publication remains blocked pending a supported packaging solution. Root and staged plugin manifests pass Omarchy validation.

Automatic insertion still requires a usable captured accessibility target. Terminals without one remain copy-only. Live Wayland and terminal paste have not been verified on the reporter's setup.

Validated with 587 tests, Ruff, formatting, ShellCheck, private D-Bus shortcut checks and the isolated Omarchy widget suite. Hosted CI is manual and was not run.

Changes: terminal and keyboard-layout fixes, single-repository Omarchy packaging.

Mluva 1.5.2 — Narrow window fixes

Mluva 1.5.2 keeps conversation controls usable in narrow tiled windows, including 360 logical pixels at 2× scaling.

• Hidden settings and meeting pages no longer force the conversation wider than its tile.
• Recording status sits above the buttons in compact windows. Rewrite controls reflow and long feedback wraps.
• Continue recording and final-rewrite cancellation use accessible icons in the compact header. Wide windows retain their text labels.
• The native app and Omarchy widget ship together at version 1.5.2; widget behavior and runtime dependencies are unchanged.

Download mluva-1.5.2-source.tar.gz, verify it with SHA256SUMS, extract it, quit Mluva and run bash install.sh in the extracted directory. Existing settings and saved conversations are preserved. Installation and upgrade details.

Validated with 574 tests, Ruff, formatting, ShellCheck, private D-Bus shortcut checks, ten compact-window scenarios, the continuation lifecycle and the isolated Omarchy widget suite. The native app was also installed and rendered from a disposable prefix. UI checks use isolated X11 with synthetic device/provider boundaries; live Wayland, microphone and provider combinations were not retested. Hosted CI remains manual.

Fix: #58

Mluva 1.5.1 — app and widget in one package

Mluva 1.5.1 packages the native app and Omarchy widget together in one repository, version and source archive.

• Rename or delete chats from the sidebar menu, merge conversations, and click a conversation title to edit it.
• Install or update both app and widget with bash install.sh from this release. The widget no longer needs a separate Git repository or download.
• Clean legacy Git widget installs migrate automatically with a backup. Local widget edits and manual installs are preserved for reconciliation; --app-only leaves them alone.
• Updated QML loads through a new entry-point path without restarting the desktop shell. Failed widget setup restores the previous files.

Download mluva-1.5.1-source.tar.gz, verify it with SHA256SUMS, extract it, quit Mluva and run bash install.sh inside the extracted directory. Existing settings and saved conversations are preserved. Installation and upgrade details.

Validated with 574 tests, Ruff, formatting, ShellCheck, private D-Bus shortcut checks, the staged widget's isolated UI suite and the installed native conversation-management smoke test. The source archive was compared against the release commit and its staged installation. Hosted CI is manual; physical microphone and provider combinations were not retested.

The separate widget repository is retired. All current installation, source and releases live in Mluva.

Mluva 1.5.0 — Conversation management

App and widget now ship together. For installation and updates, use Mluva 1.5.1. The old widget repository has been retired; historical widget links below point to the matching source in Mluva.

Rename, merge or delete conversations directly from the sidebar. Click an open chat's title to edit it without leaving the conversation.

• Rename: Enter or the checkmark saves; Escape or Cancel discards the change. Manual titles take precedence over automatic titles while document edits and unsent prompts stay intact.
• Merge with…: Search for a destination chat, keep its title and append the other conversation after it. Working text and prompt drafts are combined; original transcripts, recordings and saved rewrites remain available in History and exports. A combined version keeps future rewrites grounded in both chats.
• Delete…: Confirm removal of the entire conversation, including merged or continued recordings and retained audio. Active recording, rewriting, recovery and Incognito prevent sidebar merge/delete operations.

Merge and deletion require confirmation. Merge has no automatic undo. These actions run locally without provider requests or clipboard changes.

The compatible Omarchy widget remains at 0.5.0. Existing settings and conversations are preserved. No widget update is needed if 0.5.0 is already installed.

Download mluva-1.5.0-source.tar.gz and SHA256SUMS into the same directory, save any pending edits and quit Mluva, then run:

sha256sum --check SHA256SUMS
tar -xzf mluva-1.5.0-source.tar.gz
cd mluva-1.5.0
bash install.sh

For an existing clean Git checkout, save pending edits, quit Mluva, run git pull --ff-only, then rerun bash install.sh. See the installation guide and changelog.

Verified locally: 563 tests, Ruff, generated-feature consistency, shortcut lifecycle checks, ShellCheck and the existing native conversation regression suite. The new controls were exercised at 420 × 520, 480 × 640 and 1060 × 780. Installation from the source archive into a disposable prefix passed the native rename/merge/delete and Incognito checks. All 400 archive files and 102 installed files matched the release source. Desktop checks used an isolated X11 session and synthetic content; real microphone and hosted-provider calls were not exercised. Hosted CI was not run.

Mluva 1.4.0 — Continued dictation

App and widget now ship together. For installation and updates, use Mluva 1.5.1. The old widget repository has been retired; historical widget links below point to the matching source in Mluva.

Continue a finished dictation in the same conversation. The previous text stays visible while new speech appears underneath, and History keeps the combined result together with raw recordings available for recovery.

• Continue recording is available in the app and Continue in the finished Omarchy widget.
• Live rewrite cycles through Off, Once and Continuous. Once resets after one recording; Continuous stays selected across recordings.
• Thinking level follows the selected Codex model's catalog. Compatible providers use advertised choices or explicit overrides; Provider default keeps the server's normal behavior.
• The widget's dismissal timer starts after completion even when keyboard focus was inherited. Hovering, deliberate interaction, menus and active rewrites still pause it.

Update the app and the matching Omarchy widget 0.5.0 together. Existing settings and conversations are preserved.

Download mluva-1.4.0-source.tar.gz and SHA256SUMS into the same directory, quit Mluva, then run:

sha256sum --check SHA256SUMS
tar -xzf mluva-1.4.0-source.tar.gz
cd mluva-1.4.0
bash install.sh

For an existing clean Git checkout, quit Mluva, run git pull --ff-only, then rerun bash install.sh. See the installation guide and changelog.

Verified locally: 558 tests, Ruff, generated feature documentation, the production widget on a private display/session bus, and installation from the source archive into a disposable prefix. The installed app passed the continuation regression check through recording, finalization, grouped History, copying, cancellation, Live modes and thinking controls. All 398 archive files and 102 installed source files matched the release. These checks use synthetic speech/provider boundaries; real microphone and hosted provider calls were not exercised. The previously reported unrelated Mermaid screenshot check still has an empty GTK render-node limitation and is not claimed as passing.

Mluva 1.3.0 — Reliable startup

Mluva 1.3.0 fixes desktop launches blocked by a retired managed credential reference, even when the current speech credential is already available in the local cache.

• The launcher now prefers an enabled local credential snapshot and loads only the supported speech token. Existing process credentials, supported credential-name overrides, and installations without snapshots retain their existing paths.
• The app introduction now includes Daniel's narration and inline GitHub playback.

This release retains the security fixes from 1.2.1. The matching Omarchy widget remains version 0.4.0.

Download mluva-1.3.0-source.tar.gz and SHA256SUMS below into the same directory, then run:

sha256sum --check SHA256SUMS
tar -xzf mluva-1.3.0-source.tar.gz
cd mluva-1.3.0
bash install.sh

Quit Mluva before upgrading. Setup preserves settings and saved conversations. Managed snapshot users need a ready snapshot containing the selected speech credential. See the installation guide.

The unpacked source archive passes 546 Linux tests, generated-feature checks, Ruff lint and formatting, private D-Bus shortcut checks, and installation into a disposable prefix. Installed source modules, renderer assets, launcher and version metadata match the archive. ShellCheck passes, and the packaged application opens and exits cleanly in an isolated X11 session.

This is a Linux source release. GUI verification used a synthetic credential with providers disabled; live transcription and physical Wayland shortcuts were not retested. Fedora GNOME compatibility has not received recent live desktop acceptance. Hosted product CI was not dispatched.

Changelog · Startup fix · Changes since 1.2.1

Mluva 1.2.1 — Security update

Mluva 1.2.1 fixes the findings from the security audit and includes the identity, prompt editing and Live workspace improvements prepared for 1.2.0.

• Native Codex rewrites run without tools, inherited global instructions or integrations. Mluva checks the session's capabilities before sending text and rejects unexpected tool activity. Existing Codex sign-in is preserved.
• Incognito dictation, meeting and batch-preview audio use private memory-backed storage with crash cleanup and no disk fallback. Operating-system swap and crash dumps remain outside Mluva's control; the selected cloud speech provider still receives submitted audio.
• ElevenLabs uploads reject redirects, preventing a redirect destination from receiving credentials or recordings.
• Scratchpad, settings and personalization files are private before content is written and are replaced atomically. Application data directories are restricted to their owner.
• The local Mermaid renderer uses patched DOMPurify 3.4.15 and lodash-es 4.18.1, with a reproducible locked build, checksum, component inventory and license notices.

Download mluva-1.2.1-source.tar.gz and SHA256SUMS below into the same directory, then run:

sha256sum --check SHA256SUMS
tar -xzf mluva-1.2.1-source.tar.gz
cd mluva-1.2.1
bash install.sh

Quit Mluva before upgrading. Setup preserves settings and saved conversations and installs the new Bubblewrap dependency. The matching Omarchy widget remains version 0.4.0. Native Codex isolation was verified with Codex 0.155.1; unsupported older protocols fail closed. See the installation guide.

The unpacked source archive passes 540 Linux tests, generated-feature checks, Ruff lint and formatting, and a real installation into a disposable prefix. Installed source files and renderer assets match the archive, and its data directory has owner-only permissions. The security changes also passed isolated native Codex, shortcut, conversation and Mermaid UI checks, ShellCheck, and a clean renderer rebuild with zero npm audit vulnerabilities.

This is a Linux source release. Omarchy is the primary platform; Fedora GNOME compatibility has not received recent live desktop acceptance. Tests used synthetic audio and local protocol providers; live microphone, real provider accounts and production Wayland interactions were not retested. Hosted product CI was not dispatched.

Changelog · Security fixes · Changes since the previous published release

Mluva 1.1.0 — Live Grilling and a quieter workspace

App and widget now ship together. For installation and updates, use Mluva 1.5.1. The old widget repository has been retired; historical widget links below point to the matching source in Mluva.

Mluva 1.1 lets you turn on Live rewrite while speaking, develop an idea with Grilling, and browse saved conversations while recording continues. It ships with the matching Omarchy widget 0.4.0.

• Live Grilling: unanswered questions stay above evolving architecture notes. The model is instructed to retire answered questions and add intent, constraints, preferences and technologies only as you supply them. Grilling is the default for new preferences; existing template choices are preserved and Live remains opt-in.
• A quieter workspace: matching monospace text, a hidden-by-default history sidebar, a bare recording light and timer, and a small bottom status line. The floating recorder gains three bottom placement presets and smoother recording motion while remaining draggable.
• Local Mermaid sketches: diagrams render in live and saved drafts. Editing, copying and exporting retain exact Markdown, and incomplete or failed sketches remain visible source. WebKitGTK 6.0 is included in dependency setup.
• Navigation and controls: incoming speech does not steal your history selection. Dates have a weekday/day/month format and optional AM/PM clock. Ctrl+P finds every settings row, with direct commands for common presentation and Live choices.
• Pause without losing work: pause/resume and template changes preserve manual edits. Stopping while paused saves the draft as not reconciled, without automatically copying it. Batch previews include speech captured before enabling Live and do not issue preview requests while Live is off.

Download mluva-1.1.0-source.tar.gz and SHA256SUMS below, then run:

sha256sum --check SHA256SUMS
tar -xzf mluva-1.1.0-source.tar.gz
cd mluva-1.1.0
bash install.sh

Quit Mluva before upgrading. Setup installs the native app and updates the Omarchy plugin while preserving settings and conversations; conflicting plugin customizations stop for reconciliation. Provider setup remains separate. See the installation guide.

This is a Linux source release. Omarchy is the primary platform; Fedora GNOME compatibility is retained but has not recently received live desktop acceptance. Optional promotion media and screenshot evidence are available in the full tagged repository.

Both the release tree and unpacked source archive passed 419 Linux tests, generated-feature checks and Ruff. Private shortcut checks, ShellCheck and installation from the archive passed. Existing isolated GTK/QML/Mermaid checks cover the unchanged feature implementation. Real-provider Grilling quality, live Hyprland/multi-monitor behavior and production browser sandbox operation remain unverified; see the verification report and screenshots. Hosted CI was not dispatched.

Changelog · Changes since 1.0.0