Video To Checklist Converter — Free, On Your Own Machine, Nothing Uploaded (2026)

Local Toolchain Setup for Video-to-Checklist Extraction

Install yt-dlp for downloading, ffmpeg for audio extraction, and whisper.cpp (GGML large-v3 model) for transcription — all run offline. Use yt-dlp -x --audio-format wav "URL" to pull 16kHz mono WAV directly. Feed output to whisper.cpp/main -m models/ggml-large-v3.bin -f audio.wav --output-txt. For GPU acceleration, compile with GGML_CUDA=1 (NVIDIA) or GGML_METAL=1 (Apple Silicon); expect 15–25× realtime on RTX 4070 / M3 Pro. Skip cloud APIs — local inference keeps proprietary content off networks and eliminates per-minute costs. Store models in ~/models/ and alias whisper=~/whisper.cpp/main -m ~/models/ggml-large-v3.bin for repeat use.

Post-Processing Pipeline for Actionable Checklists

Pipe raw transcript through a local LLM (Ollama with llama3.1:8b-instruct-q4_K_M) using a strict prompt: "Extract only imperative action items from this transcript. Output as markdown checkboxes. No summaries. No categories. One task per line." Example command: cat transcript.txt | ollama run llama3.1:8b-instruct-q4_K_M "Extract only imperative action items...". Validate output with a second pass: grep -E '^\s*-\s*\[' checklist.md | wc -l to count items. For timestamped tasks, add --output-srt to whisper, then parse start_time --> end_time blocks alongside LLM output using a Python script that merges by line proximity. Save final checklist as checklist-$(date +