GPT-Voice v2.4.0-alpha.1

Install GPT-Voice by downloading the package for your operating system below.

GPT-Voice v2.3.0

What's Changed

• feat(prettify): add transformation profiles and quick apply by @swimmwatch in #57

Full Changelog: v2.2.0...v2.3.0

GPT-Voice v2.2.0

GPT-Voice 2.2.0 expands the prompt-writing workflow with translation, stronger provider diagnostics, and clearer connection readiness while preserving the existing Voice and Prettify experience.

Added

• Translation providers for Google, Bing, and Yandex, including target-language selection, startup warm-up, connection state, and first-request readiness.
• Metadata-only audit lifecycles for Voice, Prettify, and Translation providers.
• Optional local diagnostic capture, privacy-safe export, and audit-log settings with explicit deletion controls.
• Repository abstractions and process-owned dependency injection for database and external-source access.

Changed

• Provider initialization and connection status now expose clearer, human-readable failure reasons.
• Settings and startup presentation are more stable and accessible.
• README now presents GPT-Voice as a focused Speak, Polish, Translate workflow for writing prompts faster.

Fixed

• Linux and Windows packaged startup benchmarks no longer wait for provider readiness in benchmark mode.
• CI checks now use complete Git history where history-dependent validation requires it.
• Cross-platform lifecycle, IPC, packaging, and security boundaries received focused hardening and coverage.

Full changelog: v2.1.0...v2.2.0

GPT-Voice v2.1.0

Added

• Claude Web live PCM streaming during recording, with bounded buffering, cancellation, and explicit buffered Retry.
• Claude CLI and experimental Codex CLI Prettify providers with privacy-preserving process isolation and provider-owned authentication.
• Eleven application languages and a System language selector.
• Main-window Prettify provider selection, provider-specific summaries, and connection-state visibility.

Changed

• Prettify settings and controls are capability-driven across Ollama, vLLM, Claude CLI, and Codex CLI.
• Dependencies, GitHub Actions, release packaging policy, and startup benchmark cleanup have been updated.

Fixed

• Provider-switching, CLI availability, status presentation, and release startup-benchmark reliability improvements.

Known limitation

Claude Web live streaming has a known reliability issue: some recordings can finish without a transcript or take longer than expected after Stop. The integration remains a private, browser-owned integration that can change. Use another supported transcription provider when a reliable result is required.

Codex CLI remains experimental. CLI authentication is managed only by the installed CLI; GPT-Voice does not read or store CLI account credentials.

Linux and Windows artifacts are provided below. macOS artifacts are not available pending signing and notarization.

GPT-Voice v2.0.0

What’s Changed

• Redesigned the desktop controls, settings, history, and About windows for a clearer day-to-day workflow.
• Added robust hotkey capture and conflict validation, including a retry-transcription shortcut for failed requests.
• Added in-memory deduplication for completed repeated transcriptions, avoiding duplicate provider calls for the same audio while preserving privacy.
• Hardened browser recovery, provider and settings validation, selected-text prettification, and user-facing error handling.
• Improved Linux and Windows release packaging, runtime verification, size measurement, and cold-start benchmarking.

Upgrade notes

This is a major desktop update. Review configured hotkeys and provider settings after upgrading. Retry audio and successful-result caches are memory-only and are cleared when the app exits.

Verification

• npm run format:check
• npm run lint
• npm run typecheck
• npm run test:types
• npm test (322 tests)
• npm run validate:dependabot
• npm run audit:prod
• npm run build:prod
• npm run prepare:cloakbrowser
• npm run smoke:cloakbrowser
• npm run pack
• npm run verify:packaged
• npm run dist:fedora

GPT-Voice v1.4.0

What's Changed

• Added retry for failed voice transcription uploads with a configurable Ctrl+F9 hotkey.
• Kept retry audio in memory only and enabled the retry shortcut only while retryable audio exists.
• Added distinct native success and error sounds for transcription, translation, and prettify completion notifications.
• Added in-memory caching for repeated selected-text translation and prettify actions to avoid duplicate provider calls.
• Added action-specific handling for empty selected text in translate and prettify flows.
• Added busy-state gating so recording, transcription, translation, and prettify hotkeys do not spam duplicate work.

Verification

• npm run format:check
• npm run lint
• npm run typecheck
• npm run test:types
• npm test
• npm run validate:dependabot
• npm run audit:prod
• npm run build:prod

GPT-Voice v1.3.0

GPT-Voice v1.3.0 adds Prettify Text, configurable text action settings, improved App Settings validation/logging, and related reliability fixes.

Release v1.2.0

Highlights

• Improved ChatGPT Web recognition reliability by restoring reliable WAV uploads and clearer ASR failure handling.
• Added first-class 429/rate-limit handling and OS notifications for recognition failures.
• Added an app settings window for CloakBrowser behavior, identity, and proxy settings.
• Reworked translation into an F11 selected-text workflow that copies translated text to the clipboard.
• Fixed expired ChatGPT session state so the app no longer shows stale connected status.
• Updated npm and GitHub Actions dependencies.

Validation

• PR quality gates passed.
• Fedora Linux package smoke passed.
• Windows package smoke passed.

Installers will be attached by the release artifact workflow after it completes.

Release v1.1.0

Added

• Added RPM package support for Fedora, RHEL, CentOS, openSUSE, and other RPM-family Linux distributions.
• Added a Fedora Docker release container for reproducible Linux packaging.

Changed

• Linux release artifacts now include AppImage, deb, and rpm packages.
• macOS release artifacts are paused until Developer ID signing and notarization are configured.

Fixed

• Transcription uploads now use WAV encoding to address provider-side transcription failures.
• Electron runtime and logging are lazy-loaded for packaged/runtime stability.

Documentation

• Updated Linux RPM install, update, uninstall, verification, and package maintenance guidance.

Release v1.0.6

GPT-Voice 1.0.6

Fixed

• Fixed release artifact versioning so installer filenames, package metadata, AppStream metadata, DMG title, Debian metadata, and generated license files use the GitHub release tag version.
• Release builds no longer depend on the static version field in package.json as the source of truth.
• Prevented mismatches like releasing v1.0.5 while generated installers still contain 1.0.1 in their filenames or metadata.

Changed

• Added an automatic release-version injection step to the release pipeline.
• Added support for passing a release_tag when running the release workflow manually through workflow_dispatch.
• Added a dedicated release-version script that validates semver tags such as v1.0.6, updates the temporary build manifest, and exports RELEASE_TAG / RELEASE_VERSION for later pipeline steps.

Notes

• This release does not change the desktop application runtime behavior.
• The fix affects release packaging and generated installer metadata only.