v1.3.2

🚧 Breaking changes

• None

✨ New features

• Added expanded Groq TTS controls for playback format, sample rate, and speed, with direct native Groq output used for single-part unprocessed requests. (#31)

🐛 Bug fixes

• None

🔧 Improvements

• Kept ffmpeg handling for normalization, Long TTS stitching, and WAV compatibility repair while allowing compatible unprocessed TTS requests to use Groq's native output format directly. (#31)
• Added Indonesian (id-ID) to STT language hint selectors so it is available when configuring STT services and calling the groq.transcribe_audio action. (#33)

🔄 Other changes

• Updated README guidance, translations, diagnostics, and regression coverage for expanded TTS controls and STT language option parity. (#31, #33)

Full Changelog: CHANGELOG

v1.3.1

🚧 Breaking changes

• Groq response actions now require explicit target selection: service-level actions require service_id, and account-level actions require config_entry_id, preventing automations from silently changing behavior as accounts or configured services are added. (#29)

✨ New features

• Added support for the Groq qwen/qwen3.6-27b vision model and improved dynamic vision model discovery from Groq model metadata. (#26)

🐛 Bug fixes

• Replaced implicit single-account or single-service action fallback behavior with explicit action validation so service calls fail clearly when the required target is missing. (#29)

🔧 Improvements

• Refreshed discovered model handling so successful Groq /models refreshes replace stale built-ins, allowing newly visible models to appear and removed models to stop passing runtime validation. (#26)
• Moved Groq service action names and descriptions from services.yaml into integration translations in line with current Home Assistant guidance. (#27)
• Added Material Design Icons for every Groq service action so Home Assistant can show action-specific icons in the UI. (#28)

🔄 Other changes

• Updated README guidance for explicit Groq service and account action selection. (#29)
• Expanded regression coverage for dynamic model registry behavior, service translation metadata, service icons, and explicit action target validation. (#26, #27, #28, #29)

Full Changelog: CHANGELOG

v1.3.0

🚧 Breaking changes

• Raised the minimum supported Home Assistant version to 2026.6.0 to match the patched floor for CVE-2026-54317 / GHSA-x84v-g949-293w. (#23)

✨ New features

• None

🐛 Bug fixes

• Fixed repairs platform registration by adding an aborting repair flow for non-fixable Groq issues. (#22)

🔧 Improvements

• None

🔄 Other changes

• Updated installation documentation now that Groq is available as a default HACS repository. (#24)
• Expanded regression coverage for repairs platform fix-flow registration. (#22)

Full Changelog: CHANGELOG

v1.2.3

🚧 Breaking changes

• None

✨ New features

• None

🐛 Bug fixes

• Fixed TTS vocal direction option handling so users can clear defaults, select explicit None, and avoid storing or speaking invalid sentence-length directions. (#19)
• Fixed the Home Assistant 2026.6 config-entry reload deprecation by replacing update-listener reload behavior with explicit reload paths from config and options flows. (#20)

🔧 Improvements

• None

🔄 Other changes

• Expanded regression coverage for TTS vocal direction validation and Home Assistant config-entry reload behavior. (#19, #20)

Full Changelog: CHANGELOG

v1.2.2

🚧 Breaking changes

• None

✨ New features

• None

🐛 Bug fixes

• Fixed MP3 TTS playback on HomePod and Apple TV targets by converting MP3 output with a HomePod-tested 44.1 kHz mono 128 kbps profile. (#17)

🔧 Improvements

• None

🔄 Other changes

• Added regression coverage for the HomePod-safe MP3 ffmpeg conversion profile. (#17)

Full Changelog: CHANGELOG

v1.2.1

🚧 Breaking changes

• None

✨ New features

• None

🐛 Bug fixes

• Fixed HomePod and Apple TV TTS playback by validating Groq WAV output before serving it directly and rewriting malformed, non-WAV, or non-16-bit PCM WAV payloads through the ffmpeg WAV compatibility profile. (#15)

🔧 Improvements

• Optimized prompt cache expiry handling with heap-backed expiry bookkeeping and stale-entry compaction, avoiding full-cache scans on cache hits. (#14)
• Deferred Home Assistant camera and media-source helper imports until service paths need them, reducing import-time overhead during service registration. (#14)

🔄 Other changes

• Expanded tests for prompt-cache stale expiry compaction and TTS WAV compatibility handling. (#14, #15)

Full Changelog: v1.2.0...v1.2.1

v1.2.0

🚧 Breaking changes

• None

✨ New features

• Added opt-in Compound built-in tool controls with validation for dedicated options and raw compound_custom.tools.enabled_tools payloads. (#10)
• Added optional Long TTS announcements that split long Orpheus text into Groq-sized chunks, synthesize them sequentially, and stitch the result with ffmpeg. (#11)
• Added selectable TTS playback output formats, keeping Groq Orpheus requests in WAV while allowing local ffmpeg conversion to MP3 or FLAC for speaker compatibility. (#12)

🐛 Bug fixes

• Handled Groq TTS request timeouts as expected network failures with clearer unavailable-state and Home Assistant error reporting. (#8)
• Preflighted ffmpeg before spending Groq quota for converted TTS output and kept missing-ffmpeg repair issues aligned with the configured audio processing state. (#12)

🔧 Improvements

• Migrated TTS synthesis into the shared GroqApiClient, removing the standalone TTS engine and reusing the common HTTP session, rate-limit, and network-error paths. (#9)
• Tightened dynamic TTS model capability inference so unsupported TTS-looking models are not offered in voice/model pickers. (#9)
• Sent explicit empty Compound tool allow-lists by default and Groq-Model-Version: latest only when latest-only Compound tools are enabled. (#10)
• Disabled Long TTS and audio normalization options when ffmpeg is unavailable, and validated Long TTS batches before sending partial requests. (#11)

🔄 Other changes

• Updated README, architecture notes, quality-scale metadata, translation strings, and TTS benchmark helpers for the new TTS and Compound tool behavior. (#9, #10, #11, #12)
• Expanded tests for TTS timeout handling, shared API-client synthesis, Compound tools, Long TTS chunking/stitching, TTS output conversion, diagnostics, translations, and coverage paths. (#8, #9, #10, #11, #12)

Full Changelog: v1.1.0...v1.2.0

v1.1.0

v1.1.0 - 2026-05-16

🚧 Breaking changes

• None

✨ New features

• Added Home Assistant AI Task support for Groq text and structured data generation. (#5)
• Added Home Assistant LLM tool-calling support for Assist and AI Tasks, including tool request/result conversion and guardrails for unsupported models. (#5)
• Added multimodal attachment handling for supported Assist and AI Task image inputs. (#5)

🐛 Bug fixes

• Hardened Groq service input validation, reserved request-body option handling, and Assist context handling. (#3)

🔧 Improvements

• Improved Groq API and TTS request performance by preloading Home Assistant's shared aiohttp session helper and reusing the managed session. (#6)
• Added expanded translation coverage for Bulgarian, Danish, English regional variants, Spanish, Estonian, Finnish, French, Hungarian, Italian, Lithuanian, Latvian, Norwegian Bokmal, Dutch, Polish, Brazilian Portuguese, Romanian, and Swedish. (#4)
• Updated README documentation for AI Tasks, image/audio workflows, and current Home Assistant development expectations. (#2)

🔄 Other changes

• Added a TTS rate-limit benchmark helper for local performance checks. (#6)
• Declared the jsonschema runtime dependency required for AI Task structured output validation. (#5)
• Expanded tests for AI Tasks, tool calls, translation coverage, manifest metadata, rate-limit handling, and preload fallbacks. (#3, #4, #5, #6)

Full Changelog: CHANGELOG

v1.0.2

HACS default release

v1.0.0

Initial release