pkanderi-abio/youtube-autopilot

Purpose and Entry Points

YouTube Autopilot is a GitHub Actions-driven, unattended video-production system for multiple YouTube channels. It discovers topics, creates short- and long-form videos, schedules uploads, and records channel history. The primary runtime entry point is backend/src/run-pipeline.js; scheduled orchestration is defined in .github/workflows/pipeline.yml. The static dashboard begins at frontend/index.html.

Channel Configuration and Persistent State

Channels, their topic strategy, formats, and cadence are centrally defined in backend/config/channels.json. Pipeline state and publishing history are maintained as committed JSON data, including backend/data/history-channel1.json and backend/data/history-channel2.json. This makes GitHub repository state both the durable record and the dashboard’s primary data source.

Pipeline Orchestrator

backend/src/run-pipeline.js coordinates the numbered production stages, selects applicable channel/format work, tracks stage outcomes, and establishes the failure boundary for a run. Shared state handling is isolated in backend/src/lib/state.js, while channel-specific historical performance can be accessed through backend/src/lib/analytics.js.

Topic Discovery and Content Generation

The first stage, backend/src/steps/1-discover-topic.js, chooses either current trends or a configured evergreen pool. Trend acquisition is abstracted by backend/src/lib/trends.js. Script generation in backend/src/steps/2-generate-script.js uses the local LLM integration in backend/src/lib/llm.js, so generative text does not require a paid hosted model API. backend/src/steps/3-optimize-seo.js transforms the content into YouTube-oriented metadata.

Media Production

Voice generation is performed by backend/src/steps/4-generate-voice.js. Background creation in backend/src/steps/5-generate-background.js uses stock footage through backend/src/lib/stockFootage.js, with a gradient fallback when usable footage is unavailable. Video assembly is performed by backend/src/steps/6-assemble-video.js, implying a local ffmpeg runtime boundary. Nursery audio assets and their lookup manifest are held under backend/audio/nursery/ and backend/audio/nursery-manifest.json, accessed through backend/src/lib/nurseryAudio.js.

Publishing and External Boundaries

Thumbnail generation is a distinct late-stage artifact step in backend/src/steps/7-generate-thumbnail.js. backend/src/steps/8-upload-youtube.js publishes and schedules completed media through the YouTube integration in backend/src/lib/youtube.js. External service boundaries are YouTube APIs, Pexels’ stock-footage API, local Ollama, free TTS tooling, and ffmpeg; credentials are expected through environment configuration such as backend/.env.example.

Automation, Analytics, and Dashboard

GitHub Actions supplies cron execution and repository-write automation via .github/workflows/pipeline.yml. Post-publication analytics can be fetched with backend/scripts/fetch-analytics.js and persisted in files such as backend/data/analytics-channel1.json; .github/workflows/analytics.yml likely schedules this independently. The frontend is a static control panel whose data is assembled by frontend/build-data.js and rendered by frontend/index.html, separating operational visualization from the production runtime.