Pin exact matching versions for torch, torchvision, and torchaudio to fix ABI mismatch

Force PyTorch CPU version to prevent AMD ROCm executable stack crash

Pin numpy and tensorflow to resolve dependency graph explosion

Fix numpy version conflict for Hugging Face build

Update history UI and integrate native DeepFilterNet for stronger noise reduction

Fix audio streaming by adding missing stream endpoint

revert: remove /api/stream endpoint per user request

feat: add missing /api/stream endpoint to fix inline audio playback on frontend

fix(critical): remove dead gradio_client import that was crashing backend boot

fix: change background temp file cleanup threshold from 1 hour to 24 hours

chore: remove unused dependencies and add automated background temp file cleanup to prevent storage issues

fix: remove dangling youtube_service imports from core background tasks

fix: resolve all ESLint warnings and errors across frontend components

refactor: completely remove youtube downloader feature from frontend and backend

feat: upgrade Pyannote diarization and fallback to original audio for voice separation

fix: implement create_custom_zip to fix ImportError on startup

build: pin exact pytorch CPU versions to fix pip resolution-too-deep error

fix: enforce CPU-only PyTorch to fix CUDA driver crash, and remove queue UI from frontend

refactor: remove queue for youtube download, return file directly

fix: enforce ml-dtypes>=0.4.0 to fix onnx import error

fix: ignore corrupted tasks_db.json on boot

fix: add gcc to Dockerfile for diffq c-extension compilation

refactor: modularize backend architecture with routers and core services

fix: add missing yt-dlp to requirements.txt

fix: resolve queue status bug and audio file download mime types

fix: resolve lucide-react missing exports

fix: resolve syntax error in youtube page template string

feat: add youtube to audio converter page and backend

feat: add beautiful queue waiting room ui

feat: add AI speaker separation diarization tool

feat: add pro tools backend integration

feat: unique pro tools UI and animated SVG extraction flow

fix: resolve missing icon import causing build failure

feat: firebase history tracking and id search

feat: added custom folder and zip naming to download UI

feat: next-level UI for custom format and chunked stem downloads

fix: add torchcodec and soundfile to requirements to fix torchaudio save crashes

fix: capture demucs stderr to diagnose crash

feat: replace Demucs HF Space API with completely native subprocess execution to eliminate all rate limits

fix: implement global Gradio Client singletons to prevent rapid WebSocket connections triggering 429 bans

fix: pin gradio_client version to 1.3.0 and restore hf_token to fix WS protocol errors

fix: remove deprecated hf_token kwarg from gradio_client

fix: add python-multipart to dependencies for FastAPI file uploads

fix: add Hugging Face YAML configuration block to README

fix: enforce login add rate-limit retry logic fix clerk ui and fix binary history

Massive UI/UX overhaul, added 4-stem separation, interactive audio player, and SEO metadata

Switch architecture from YouTube URLs to direct file uploads to bypass bot blocks

Fix frontend API URL concatenation bug

Add frontend source code

Set concurrent workers to 5