v0.40.0

0.40.0 (2026-07-10)

Features

• admin: booth-log prompt summaries, shadcn scroll areas, segment fire pads (#958) (5897729)
• admin: library search paging, sounds-like search mode, URL-persisted state (#967) (5049787)
• admin: per-call TTS log in the debug panel's TTS routing card (#960) (a24a5c3)
• admin: support multi-file jingle import (#952) (990b035)
• app: web-player parity, sleep timer, Google Cast + AirPlay connectivity (#970) (0b060a3)
• broadcast: configurable Icecast listener cap (ICECAST_MAX_CLIENTS) (#978) (3a56a1b)
• connect: discoverable API — explorer, playground, HTTP MCP (#928) (6ac4b0c)
• dj: add ElevenLabs v3 audio-tag hint to the system prompt (#916) (718f5fa)
• library: manual playlist creation in admin library + Navidrome library-scoping docs (#704) (#953) (aa0f046)
• llm: pool-mode segment path, pick prompt diet, and llm-bench harness (#961) (42ca3ac)
• mix: tail-loudness-shaped exit canvas + boundary-key matching (#972) (69cea9a)
• observatory: WebGL galaxy renderer replaces SVG + canvas maps (#957) (35bcda2)
• personas: system-prompt template library with switching + modal editor (#955) (d1f00c3)
• search: Brave Search API as web-search backend (#623) (#984) (a82a23c)
• shows: multi-value genre/mood/energy/era show filters (#929) (#951) (53e0524)
• stations: add Radio Parra Vergara (#944) (5e06195)
• tools: import track analysis from AudioMuse-AI (#934) (5f81f9a)
• tts: choose which heavy TTS engines load via TTS_HEAVY_ENGINES (#956) (3078d4d)
• tts: expose ElevenLabs voice_settings globally under tts.cloud (#915) (5036bc1)
• tts: operator-editable speech corrections in admin settings (#981) (c34df62)
• web: now-playing UX — up-next tease, clock toggle, off-air state, cover placeholder (#968) (6bd35f6)

Bug Fixes

• address develop→main review findings (leaked reasoning, ElevenLabs, AudioMuse importer) (#954) (bbf1be5)
• analyzer: local backend reports real CLAP/Demucs capability + AIO-aware doctor hints (#983) (f6ec6ad)
• dj: slim the programme-plan capability menu to one line per kind (#950) (26fb586)
• doctor: analyzer heavy-image hints cover non-compose installs (#974) (ae6d476)
• llm: force parallel_tool_calls:false on openai-compatible tool requests (#945) (c2adb64), closes #940
• llm: honour forceNoThink on openai-compatible/locca picker legs (#935) (42f78ee)
• llm: never air truncated reasoning (handoff spoke its own instructions) (#949) (0ddc54f)
• llm: resolve GLM structured-output and thinking-suppression failures (#923) (1434e91)
• picker: tolerate small-model id transcription slips (#939) (#948) (233fea6)
• stations: Brisflix Radio showed offline — url carried a /listen path (#964) (9f8085f)
• tts: apply speech speed locally for openai-compatible cloud TTS (#982) (bb78a0d), closes #942
• tts: normalize display text to spoken form; never air intro fragments (#965) (04c3393)

Performance

• web: canvas waveform visualizer — log-frequency bars, calm-mode gating (#973) (4021b21)

Documentation

• skills: add subwave-discord-release skill (#941) (e7bc827)