Skip to content

Mluva 0.3.0 — Speak, structure, refine

Choose a tag to compare

View all tags
@1vecera 1vecera released this 09 Sep 10:32
· 195 commits to main since this release
v0.3.0
4ce8dc4
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Mluva 0.3.0 turns dictation into an editable workspace, with live structured drafts and independent speech and rewrite engines.

  • Edit original documents and completed rewrites, save with compact icons or Ctrl+S, and use those edits in follow-ups, search and exports. Raw recognition stays available separately.
  • Completed dictation and rewrites copy automatically by default. Copy behavior, action icons, scrolling and the four-second Omarchy review timeout are configurable in Settings and the JSON dotfile.
  • Choose native Codex or a LiteLLM/OpenAI-compatible rewrite service. For speech, choose ElevenLabs Scribe, local Voxtype/Whisper, or a compatible transcription endpoint. Credentials remain environment references.
  • Enable Live beside Dictate to turn speech into a task spec, structured note, faithful polish, or custom format. Drafts mark missing information, preserve manual edits and save one final version when recording stops.
  • Read longer documents with smooth following, room for incoming text and preserved manual scroll position. The Omarchy widget shares the configurable motion and review behavior.

Live rewrite is opt-in. Its default scheduler waits for 160 new characters and a four-second interval between requests. Local and proxy batch speech engines also collect provisional audio chunks before a full recording is finalized. Response time depends on the chosen engine and model. Provider and Live guide.

Install or upgrade

Install the Linux dependencies, then download the source archive and checksum below into one directory:

sha256sum --check SHA256SUMS
tar -xzf mluva-0.3.0-source.tar.gz
cd mluva-0.3.0
bash linux/install.sh
mluva

Quit a running Mluva before upgrading. Settings, saved conversations and original recognition are preserved. See the Omarchy integration guide for shell integration. Configure only the credentials required by the selected provider; local Voxtype needs no cloud speech key.

Validation

The combined release passed 318 Linux tests, generated feature metadata checks, Ruff lint and formatting on Lenovo/Omarchy. The runtime changes also passed isolated real GTK conversation and live-rewrite workflows, shortcut and AT-SPI insertion checks, production Omarchy QML animation/configuration checks, staged installation and installed-entrypoint verification. Real native Codex structured rewriting and local Voxtype transcription were exercised. Existing local conversations survived installation.

LiteLLM transports were checked against controlled HTTP endpoints; every cloud account/model was not live-tested. Omarchy integration, new providers and live rewrite remain Experimental. Hosted CI was not run; macOS and a nested GNOME Shell check were unavailable on this host. This is a Linux source release with an installer; macOS remains a source preview.

The promotion toolkit from #9 is included with explicit v0.1.1 capture provenance. New v0.3.0 Omarchy screenshots and real audio-driven videos are being prepared separately; the older clips do not demonstrate the new live mode.

Changes: PR #9 · PR #12 · Full comparison.

Assets 4
Loading