Chatalogue

Chatalogue is a local-first workstation for ingesting, transcribing, diarizing, optimizing, searching, and editing long-form spoken content.

Core Features

• Channel ingest and automatic refreshing
• High-performance transcription and speaker diarization pipeline
• Transcript optimization workbench with repair, rebuild, retranscribe, benchmarks, and campaigns
• Channel-level transcript search
• Speaker profile management and merging
• AI episode cloning workbench with multi-variant generation
• Funny-moment detection and explanation
• AI-powered summaries and chapter generation
• Cleanup, ClearVoice, VoiceFixer, and conversation reconstruction runtimes
• Video clip editing, exporting, and uploading

---

Quick Start (Installation)

Installers clone the repo, create the backend virtual environment, install dependencies, and preload the transcription stack.

Prerequisites

Before installing, make sure your system has:

1. Git
2. Python 3.10+
   • On Windows, enable Add Python to PATH during install.
3. Node.js 18+

Windows Installation

1. Download the Windows installer: install_windows.zip
2. Extract the zip into the folder where you want Chatalogue installed.
3. Open the extracted folder and run install_windows.bat.

macOS Installation

1. Download the macOS installer: install_mac.zip
2. Extract the zip into the folder where you want Chatalogue installed.
3. Open Terminal, cd into the extracted folder, then run:

chmod +x install_mac.sh
./install_mac.sh

Optional installer environment variables

Set these before running the installer if you need custom behavior:

• INSTALL_PARAKEET=1|0 default 1
• SKIP_MODEL_PRELOAD=1|0 default 0
• PRELOAD_ENGINE=auto|whisper|parakeet default auto
• OLLAMA_MODELS="model1 model2 ..."
• CHATALOGUE_REPO_URL
• CHATALOGUE_REPO_BRANCH

Example on Windows:

set INSTALL_PARAKEET=0
set SKIP_MODEL_PRELOAD=1
install_windows.bat

---

Running the App

Once installation is complete, start Chatalogue at any time.

Windows

Double-click run_windows.bat inside the Chatalogue folder.

macOS

Open Terminal, cd into the Chatalogue folder, then run:

./run_mac.sh

The frontend opens at http://localhost:5173. Backend API docs are available at http://localhost:8011/docs.

---

Test-Ready Development Setup

The normal installers create a runtime-ready environment. If you also want local test execution:

cd backend
.\.venv\Scripts\python.exe -m pip install -r requirements-dev.txt

On macOS/Linux:

cd backend
.venv/bin/python -m pip install -r requirements-dev.txt

Then run tests with:

python -m pytest backend/src -q

---

Configuration

To override backend settings, copy backend/.env.example to backend/.env and adjust values as needed.

Common settings:

• HF_TOKEN: Hugging Face token for gated diarization models
• TRANSCRIPTION_ENGINE: auto, whisper, or parakeet
• WHISPER_BACKEND: faster_whisper or insanely_fast_whisper
• PARAKEET_MODEL: default nvidia/parakeet-tdt-0.6b-v2
• PARAKEET_ALLOW_WHISPER_FALLBACK: true or false
• MULTILINGUAL_ROUTING_ENABLED: route likely non-English episodes to Whisper automatically
• MULTILINGUAL_WHISPER_MODEL: multilingual Whisper model used for language-routed jobs
• LANGUAGE_DETECTION_SAMPLE_SECONDS: opening-audio probe window for language detection
• LANGUAGE_DETECTION_CONFIDENCE_THRESHOLD: threshold for metadata/audio language routing
• PIPELINE_EXECUTION_MODE: sequential or staged
• YOUTUBE_DATA_API_KEY: optional YouTube Data API key for view-count/popularity metadata and API-key test
• DB_PROVIDER: default postgres

---

Recommended Post-Install Setup

After first launch, open Settings and review these sections:

• Transcription
  • choose TRANSCRIPTION_ENGINE
  • choose WHISPER_BACKEND
  • test the engine/runtime path
• YouTube
  • add a YouTube Data API Key if you want stable popularity/view-count backfill
  • use Test API Key to verify connectivity
• Runtimes
  • install and self-test optional local runtimes such as ClearVoice, VoiceFixer, and Conversation Reconstruction

The YouTube Data API key is the preferred path for public view-count metadata. Browser-cookie scraping remains optional fallback behavior, not the recommended default.

---

Technology Stack

• Backend: FastAPI, SQLModel, Uvicorn, embedded PostgreSQL
• Frontend: React, Vite, TypeScript, Tailwind
• AI/ML: Whisper, NVIDIA Parakeet, pyannote.audio, Ollama
• Media Tools: yt-dlp, FFmpeg

License

MIT