Changelog

2.3.0 - Configurable playback boost

Added a bounded playback boost with output limiting for generated voice audio, making quieter voices easier to hear without changing the app-owned playback safeguards.
Added the [voice] gain_db TOML setting with validated 0 through 12 dB values and a 6 dB default.
Added relay config set --playback-gain-db <0-12> for changing playback boost from the CLI, including decimal values.
Added a compact, auto-saving Playback volume boost slider to 设置 > Voice for whole-decibel adjustments.
Preserved the emergency direct-say fallback without amplification.
Updated the app and CLI version to 2.3.0.

Kept app-owned playback available while output-only audio is active and while temporary input capture is safely coordinated.
Published bounded playback observations for local integrations without moving speech ownership out of the app.
Added persistent voice reservations so new Relay lines receive stable, unused provider voices when available.
Added relay voice ensure --line <line> [--voice-id <voice-id>] for persona-agnostic automatic or explicit line-to-voice assignment.
Validated explicit voice IDs against the active provider catalog before persisting them.
Fixed PATH-invoked relay commands so bundled Kokoro and Speechify helpers resolve from the real app bundle rather than the current working directory.
Updated the app and CLI version to 2.2.0.

Fixed playback hanging indefinitely when a speech subprocess never exits, which left the menu bar showing playing with no audio until the app was restarted.
Added a playback watchdog that force-recovers stalled playback, terminating the stalled process, resetting voice-command resolution, the audio player, and the synthesizer, and marking the stalled relay failed.
The stall deadline scales with message length, and each playback phase gets its own deadline, so a long relay is never cut off mid-sentence and a slow voice synthesis never eats the budget for the audio that follows.
Playback callbacks now verify that they still own the process, audio player, or utterance that produced them, so a late callback from recovered playback can no longer mark or interrupt the relay that replaced it.
Stall recovery waits for the cancelled speech process to exit before starting the next relay, so recovery cannot briefly overlap two voices.
Fixed the inactive-line combiner requeueing a relay that was skipped or cleared while the combiner was still running.
Fixed relay clear so it also removes relays stuck in speaking and stops audio that is currently playing, instead of leaving the app speaking a relay that was already cleared.
Added relay ff [--line <line>] to fast-forward past every queued relay by marking it skipped instead of deleting it, so skipping the backlog no longer destroys relay history.
Updated the app and CLI version to 2.1.1.

Added a Kokoro-compatible voice helper for direct-download builds that can use a user-installed local Kokoro venv without bundling Kokoro packages, model weights, voices, spaCy models, or caches.
Added an invisible same-user Kokoro helper server that keeps KPipeline warm across relays and automatically shuts down when the active voice provider is no longer kokoro.
Kept the Kokoro helper server alive when a client disconnects before reading its response.
Changed BYO voice output to a Core Audio-friendly relay.audio path so Kokoro can return WAV bytes without conversion.
Documented the optional Kokoro install path, missing-install failure mode, provider TOML example, local helper lifecycle, and Apache-2.0 Kokoro package/model license boundary.
Updated the app and CLI version to 2.1.0.

Made relay config the single CLI surface for TOML-backed advanced configuration.
Added relay config set --voice-command, --combiner-command, and --cleanup-retention-minutes.
Removed the duplicate relay settings command and made relay combiner read-only; use relay config set --combiner-command to change the combiner.
Preserved 1.1.2 upgrade migration from SQLite settings into config.toml.
Added BYO voice command playback while keeping app-owned speech, mute, Focus, Ready, and Live safeguards.
Added Speechify-compatible voice synthesis support for the direct-download app, including Keychain-based API-key lookup and rate-limit handling.
Added provider voice configuration with sticky per-line voice IDs and optional stable assignment for new lines.
Added local spoken-usage counters by provider, model, voice, and line so custom voice cost can be estimated without storing another copy of relay text.
Added cleanup controls for old relay rows, spoken-usage buckets, and temporary voice audio files.
Added 设置 UI capture and roundtrip verification for release-quality 设置 changes.

Added Darwin notifications so the app wakes promptly when the CLI changes the relay queue or playback state.
Optimized relay status refresh queries to reduce work while keeping 设置 and menu state current.
Added Live mode to the overview docs.
Updated the app and CLI version to 1.1.2.

Fixed configured inactive-line combiners so the direct-download app executes the configured command instead of always using latest-only replacement.
Kept combiner execution shell-free, bounded by timeout, and validated through JSON output before queuing the combined relay.
Preserved latest-only fallback when no combiner is configured, when native combiner execution fails, or in the App Store-safe profile.
Added regression coverage for CLI and native app enqueue paths, placeholder handling, and active-line changes while combining.

Added Live mode for automatic playback of new relays.
Plays queued relays in bounded batches by line so chatty lines do not starve other lines.
Added Start Live and Stop Live menu controls.
Added a green-dot menu bar status for Live mode.
Updated the app and CLI version to 1.1.0.

Shipped a signed direct-download macOS app.
Bundled the relay CLI with matching app and CLI versioning.
Added first-start setup for CLI installation, shortcut selection, voice selection, and Open at Login.
Added Focus, Ready, and Mute queue controls with app-owned playback.