What it does summarize is a TypeScript CLI and browser extension that ingests almost anything—web pages, PDFs, images, audio, video, YouTube, podcasts, RSS feeds, even stdin pipes—and returns a condensed summary via your choice of LLM provider. It runs in the terminal, a Chrome Side Panel, or a Firefox Sidebar, with streaming Markdown output and cost metrics.

The interesting bit The browser extension is deliberately thin: it offloads heavy lifting (yt-dlp, ffmpeg, OCR, transcription) to a local daemon that autostarts via launchd/systemd/Scheduled Task. This lets the sidebar feel native while still chewing through video slide extraction and whisper.cpp transcription on your own hardware. The CLI and extension share a core library, so behavior stays consistent.

Key highlights

• Video slide extraction: screenshots + OCR + timestamped transcript cards for YouTube and local video files
• Transcript-first media: uses published transcripts when available, falls back to Groq/ONNX/whisper.cpp/AssemblyAI/Gemini/OpenAI/FAL
• Broad provider support: Codex, Claude, Gemini, Cursor Agent, OpenClaw, OpenCode, OpenRouter free tier, plus local OpenAI-compatible endpoints
• Flexible output: length presets (short through xxl), JSON diagnostics, extract-only mode, cost and timing estimates
• Smart default: if content is shorter than requested length, returns it as-is instead of wasting tokens (override with --force-summary)
• Gateway-style model IDs: openai/gpt-5.4, anthropic/claude-sonnet-4-5, google/gemini-3-flash, etc.

Caveats

• Requires Node 24+; media-heavy features need local installs of ffmpeg, yt-dlp, and optionally tesseract
• Browser extension needs the CLI daemon running locally—it’s not a standalone install
• Some providers reject certain file types (xAI models don’t handle PDFs; Google is noted as most reliable for PDFs)
• Spotify episode support is best-effort and may fail for exclusives

Verdict Worth a look if you consume a lot of long-form content across formats and want one consistent interface. Skip it if you just need occasional web page summaries—browser-native AI features are catching up fast.