SpeakHarness v0.2.0 — native desktop installers

The native desktop app is SpeakHarness's primary graphical interface. It opens its own window and runs the voice service locally; the browser dashboard is an optional secondary interface.

Installers: Windows x64 (NSIS .exe), Linux x64 and ARM64 (.AppImage and .deb), macOS Intel and Apple Silicon (.dmg). Bun, Node and native synthesis modules are bundled; Kokoro English and Piper Brazilian Portuguese voice models download once on first use. On NixOS, launch the Linux AppImage through appimage-run (install it with nix profile add nixpkgs#appimage-run if needed).

The five-platform validation installed each package, rendered its desktop window and generated WAV speech in both languages. CI has no physical speaker. Windows installers are unsigned; macOS apps are ad-hoc signed but not notarized. Linux packages target glibc desktops. SHA256SUMS-desktop.txt accompanies the installers.

The browser-only CLI, OpenTUI and OMP integration remain available from source or Nix.

SpeakHarness v0.1.6 — FlakeHub license metadata

No functional changes from v0.1.5. The FlakeHub publishing workflow now declares MIT AND GPL-3.0-or-later to reflect the original MIT source and bundled ephone GPL-3.0-or-later component. Install with nix profile add github:Matheuscara/speak-harness/v0.1.6, then speakh setup and speakh web. Source and Nix flake only; NixOS x86_64 exercised, ARM/macOS not exercised.

SpeakHarness v0.1.5 — speak final OMP replies

Adds speakh integrate omp, a local Oh My Pi extension that sends final main-session reply text to the running SpeakHarness audio owner through a user-owned Unix socket. Adds speakh daemon for UI-free playback and a deduplicated, latest-pending external speech queue. Install with nix profile add github:Matheuscara/speak-harness/v0.1.5, run speakh setup, then speakh integrate omp and start one speakh daemon (or keep speakh web running). Restart OMP once so it loads the installed extension. This release contains source and a Nix flake, not standalone binaries. Exercised on NixOS x86_64; ARM and macOS remain untested.

SpeakHarness v0.1.4 — live answers and voice previews

Sessions opens on All folders by default. The listening page refreshes on live transcript events, checks for missed changes, marks new activity, adds searchable answer history and a Latest answer action. Settings → Voices now separates English and Brazilian Portuguese voices, previews installed voices in the browser, installs missing models and lets you choose a voice per language. Install with nix profile add github:Matheuscara/speak-harness/v0.1.4, then run speakh setup and speakh web. Source and Nix flake only; no standalone binaries. Exercised on NixOS x86_64; ARM and macOS remain untested.

SpeakHarness v0.1.3 — full-page dashboard navigation

Sessions, Saved phrases and Settings are now full pages alongside Listen, with browser-history navigation and mobile layouts. Sessions has wide search/filter controls and conversation cards; Saved phrases has a dedicated archive and empty-state action; Settings lays controls out across the workspace. Install with nix profile add github:Matheuscara/speak-harness/v0.1.3, then run speakh setup and speakh web. Source and Nix flake only; no standalone binary assets. Exercised on NixOS x86_64; ARM and macOS remain untested.

SpeakHarness v0.1.2 — dashboard layout fixes

The listening desk now grows with its content so selected-session details are not clipped. The headline scales with the available panel width, and Ctrl+R is no longer intercepted by the repeat-sentence shortcut. Install with nix profile add github:Matheuscara/speak-harness/v0.1.2, then run speakh setup (downloads voices once) and speakh web. This release contains source and a Nix flake, not standalone binaries. Exercised on NixOS x86_64; ARM and macOS remain untested.

SpeakHarness v0.1.1 — English interface and screenshots

SpeakHarness now ships an English-only interface across the graphical dashboard, OpenTUI, settings, status messages and accessibility labels. The README screenshots have been recaptured from the real platform with English demo data. Brazilian Portuguese speech remains supported; your own transcript content is displayed in its original language.

Install the local Nix package:

nix profile add github:Matheuscara/speak-harness/v0.1.1
speakh setup
speakh web

Also available: speakh for OpenTUI. This release contains source and a Nix flake, not standalone binary assets or an npm publication. Verified on NixOS x86_64; ARM and macOS have not been smoke-tested here. Original code is MIT; the included ephone phonemizer is GPL-3.0-or-later (see THIRD_PARTY_NOTICES).

SpeakHarness v0.1.0 — local voice for coding agents

First public release of SpeakHarness.

• Graphical listening dashboard and OpenTUI reader for OMP, Pi, Codex and Claude Code.
• Local Kokoro English and Piper Brazilian-Portuguese voices; Markdown-aware speech and study controls.
• Read-only session adapters, searchable session explorer and an embedded-terminal fallback for other CLI harnesses.

Install with Nix:

nix profile add github:Matheuscara/speak-harness/v0.1.0
speakh setup
speakh web

This release contains source and a Nix flake, not standalone binary assets or an npm publication. Verified on NixOS x86_64; ARM and macOS builds are not yet exercised here. Original code is MIT; the bundled ephone phonemizer is GPL-3.0-or-later (see THIRD_PARTY_NOTICES).