What's Changed

🚀 Features & Enhancements

• Onboarding Redesign: Streamlined the initial setup with a consistent 3-step onboarding flow designed for a cleaner first time user experience.
• Studio Workspace Refactor: Updated the /studio layout to use a route-style project history table, ensuring a consistent UI between the "create" and "detail" views.
• TTS Workflow Polish: Refined the Text-to-Speech model selection flow and performed a server side cleanup to remove dead code warnings and improve stability.

⚡ Performance

• LFM2 Latency Optimization: Reduced "Time to First Token" (TTFT) for fallback chats by implementing adaptive prefill and optimizing decode hot-paths for faster response times.

📚 Documentation

• Docs Sync: Updated the user documentation to reflect the latest model catalog, CLI defaults, and current feature support, ensuring all guides are accurate to the current build.

Full Changelog: v0.1.0-beta-11...v0.1.0-beta-12

What's Changed

✨ New Features

• System Tray Integration: Added system tray controls for the desktop app, including settings for startup behavior and visibility.
• In-App Updater: Introduced a cross-platform updater for the beta channel, supported by a signed release-manifest CI pipeline and hardened runtime UI.
• Anonymous Analytics: Implemented opt-in, anonymous desktop analytics using Aptabase, complete with user consent controls.

🎨 UX & Workflow Improvements

• Transcription Overhaul: Refined the transcription index and detail workflow with a cleaner record page, background-safe polling, and a polished streaming upload modal.
• Unified Voice Experience: Merged the /voices and /text-to-speech flows to provide a standardized, model-driven UX and polished table interactions.
• Standardized Actions: Aligned history row actions across the transcription, diarization, and TTS modules for a more consistent user experience.
• Workflow Alignment: Updated the diarization module to align with the new route-based application workflow.
• Settings Refinement: Polished the settings interface and updated the version display for better clarity.

🛠️ Technical Refinements & Fixes

• Performance Optimization: Resolved polling regressions that were causing high CPU usage and page flickering.
• Architecture Update: Migrated application record IDs to UUIDs for improved data integrity and scalability.
• Bug Fixes: Resolved issues with stacked modal behavior in the transcription model selection.

Full Changelog: v0.1.0-beta-10...v0.1.0-beta-11

What's Changed

• Optimize Qwen3.5 inference end-to-end on Metal with safer SDPA/RoPE paths, adaptive dense→paged decode routing, and generic kernel telemetry compatibility by @zinyando in #106
• Native Qwen3 ASR GGUF integration with Metal hot-path decode/prefill optimizations by @zinyando in #108
• Add diarization summaries with unified Qwen3.5 model and stabilize loading/polling UX by @zinyando in #110
• Add manual segment insertion to Studio with inline add controls by @zinyando in #111

Full Changelog: v0.1.0-beta-9...v0.1.0-beta-10

What's Changed

• Qwen3.5 inference performance overhaul: batched prefill, DeltaNet/attention optimizations, and decode hot-path reductions by @zinyando in #95
• Redesign transcription and diarization into a shared two-state audio workspace by @zinyando in #96
• Fix diarization mic upload transcoding by @zinyando in #97
• Remove deprecated ASR variants and polish transcription/diarization review UX by @zinyando in #98
• Consolidate voice workflows into /voices, refresh the voices UX, and split TTS Projects into its own route by @zinyando in #99
• Rename Projects to Studio and streamline text-to-speech delivery controls by @zinyando in #100
• Overhaul Studio into a routed project workspace with full TTS project lifecycle support by @zinyando in #101
• feat(studio): add per-segment model/voice controls and harden settings state handling by @zinyando in #102
• Fix TTS voice workflows: route-aware generation dispatch, Qwen3-TTS GQA cache/decode alignment, and explicit designed-voice naming by @zinyando in #103
• Refactor Studio end-to-end naming and API contracts from tts_project to studio (remove legacy paths) by @zinyando in #104
• feat(ui): hide incomplete studio folders ui by @zinyando in #105

Full Changelog: v0.1.0-beta-8...v0.1.0-beta-9

What's Changed

• End-to-end Qwen3.5 GGUF integration with 4B/9B output-quality fixes across runtime, APIs, and UI by @zinyando in #90
• Add native LFM2.5 Audio GGUF support across the runtime, speech routes, and realtime voice by @zinyando in #91
• Refine Izwi’s UI with consistent selectors, calmer workflows, and route-aligned transcription surfaces PR Description by @zinyando in #92
• Improve Qwen3.5 chat performance and streaming reliability by @zinyando in #93
• Add first‑run onboarding modal with model setup and persisted completion state by @zinyando in #94

Full Changelog: v0.1.0-beta-7...v0.1.0-beta-8

What's Changed

• Stabilize the serve/runtime config contract by @zinyando in #81
• Add end-to-end diarization review workflow with corrections, exports, reruns, and polished history UI by @zinyando in #82
• Normalize persisted history APIs to RESTful /v1 resources and harden API routing under the UI fallback by @zinyando in #83
• Add timestamped transcription workflows with diarization-style review, history, and export UX by @zinyando in #85
• Build reusable voice workflows and redesign the TTS and voices experience by @zinyando in #87
• fix: use rustls instead of native tls by @vigsterkr in #86
• Improve /voice with persistent sessions, editable agent prompts, observational memory, and a cleaner configuration UI by @zinyando in #88

New Contributors

• @vigsterkr made their first contribution in #86

Full Changelog: v0.1.0-beta-6...v0.1.0-beta-7

What's Changed

• feat(backends): centralize backend routing and device policy across runtime, CLI, and server by @zinyando in #69
• Refactor core GGUF loading to use a shared backend-aware reader policy across Qwen and LFM2 models by @zinyando in #70
• refactor: centralize core runtime layers and extract app entrypoints by @zinyando in #72
• Improve inference latency, memory stability, and streaming backpressure across core runtime paths by @zinyando in #73
• refactor(ui): reorganize the frontend into a production Vite architecture by @zinyando in #74
• Improve Qwen3.5 runtime behavior and unify history UX across the UI by @zinyando in #76
• Refine chat UI, history management, and Qwen chat defaults by @zinyando in #78
• refactor(ui): move history actions into page headers and refresh the chat composer by @zinyando in #80

Full Changelog: v0.1.0-beta-5...v0.1.0-beta-6

What's Changed

• UI Improvements and Cleanup by @zinyando in #64
• Enhance TranscriptionPlayground UI with improved recording indicator by @zinyando in #65
• Add native Whisper-Large-v3-Turbo ASR end-to-end with upstream-aligned robust decoding and UI integration by @zinyando in #66
• Add LFM2.5 1.2B GGUF (Q4_K_M) support by @zinyando in #67
• Align lfm2.5 audio architecture by @zinyando in #68

Full Changelog: v0.1.0-beta-4...v0.1.0-beta-5

What's Changed

• Add Qwen 3.5 small model support by @zinyando in #61
• Refactor Qwen3.5 quantized backend with linear-attention, cache, and multimodal support improvements by @zinyando in #62
• refactor: add early media validation for non-Qwen3.5 models by @zinyando in #63

Full Changelog: v0.1.0-beta-3...v0.1.0-beta-4

What's Changed

• Improve diarization alignment plumbing and refinement defaults by @zinyando in #56
• Align Sortformer v2.1 diarization with NeMo streaming behavior and fix transcript speaker grouping by @zinyando in #57
• Online parakeet transcription by @zinyando in #58
• refactor: add graceful shutdown and model cleanup by @zinyando in #59

Full Changelog: v0.1.0-beta-2...v0.1.0-beta-3