Summarize

• You need to quickly digest a long article, PDF report, or research paper without reading it in full.
• You want to extract the key points from a YouTube video without watching it.
• You work with multiple AI providers and want a single tool that works across OpenAI, Anthropic, xAI, and Google Gemini.
• You need machine-readable summaries (via --json) for downstream automation or data pipelines.

• You need deep document analysis, comparison across multiple files, or structured data extraction — summarization is the only output mode.
• The content is behind a heavy paywall or bot protection and you don't have a Firecrawl or Apify API key configured.

Summarize handles web URLs, local PDFs, images, audio files, and YouTube links in a single unified CLI. No need to preprocess or convert files before summarizing.

Choose from OpenAI, Anthropic, xAI (Grok), or Google Gemini by setting the appropriate API key and passing --model. The default model is google/gemini-3-flash-preview if no model is specified.

Use --length short|medium|long|xl|xxl|<chars> or --max-output-tokens to precisely control how long the output summary is, from a quick one-liner to an extended breakdown.

Set FIRECRAWL_API_KEY to automatically fall back to Firecrawl for sites that block standard scraping. For YouTube, APIFY_API_TOKEN enables transcript extraction via Apify when the default method fails.

The --json flag outputs structured summaries suitable for piping into scripts, automation workflows, or other tools without any extra parsing.

Point Summarize at a PDF or arXiv URL and get a concise summary in seconds, letting you decide quickly whether a paper warrants a full read.

Run summarize "https://youtu.be/..." --youtube auto to get a text summary of any YouTube video's content without watching it, using Apify as a fallback for transcript extraction.

Use --json output to pipe summaries of multiple URLs or files into a downstream script, building daily briefings or automated newsletters from curated sources.

Summarize any article or documentation page from your terminal in a single command, with Firecrawl fallback ensuring coverage even on bot-protected sites.