Fast Summarization CLI

Problem

When summarizing web pages, local documents, or YouTube videos, the friction is often not just prompt wording: it is extracting readable content, handling different input types, and controlling output length. Copying raw pages into a model can miss the main article, subtitles, or accessibility constraints.

How it works

summarize23 is a CLI for three input types: URLs, local files, and YouTube links. It chooses an extraction path per input: ordinary URLs can be fetched and cleaned, YouTube links can use an automatic fallback, and local files are read directly. The extracted content is then sent to an LLM with a requested length and token cap.

Core capabilities:
- Multiple model providers: supports OPENAI_API_KEY, ANTHROPIC_API_KEY, XAI_API_KEY, and GEMINI_API_KEY; the default model is google/gemini-3-flash-preview.
- Output control: --length ranges from short to xxl, and --max-output-tokens caps generation.
- Structured output: --json emits machine-readable results for scripts.
- Extraction fallbacks: --firecrawl and --youtube can activate optional services such as FIRECRAWL_API_KEY or APIFY_API_TOKEN.

Boundaries

This fits lightweight extract-then-summarize workflows: reading public pages, organizing documents, and generating video takeaways. It is not a substitute for authentication, anti-bot handling, subtitle retrieval, or fact-checking when content is restricted or inaccurate. For engineers, it is closer to a content preprocessing pipeline than a full crawler system. Optional config lives in ~/.summarize/config.json.