lineate

Paste a link and Lineate turns it into a reading page designed to help you decide what is worth your attention and get more out of what you do read. Each output starts with concise takeaways and a strong critique, then includes either the full content or a complete summary.

What it does

• Turns links into cleaner, easier-to-read pages.
• Pulls the main takeaways to the top so you can triage quickly.
• Adds a devil’s-advocate critique to pressure-test the source's claims.
• Includes either the full content or a complete summary, depending on how you run it.
• Accepts links, multiple links pasted together, pasted text containing links, and pasted markdown.
• Processed URL(s) are opened in your browser by default.

Output structure

Each output begins with:

• Highlights: the main claims, conclusions, and important implications.
• Not in highlights: what you would miss by reading only the highlights.
• Best rebuttal: the strongest critique of the source's own case.

After that, Lineate includes either the full content or a semantically complete summary.

PERSONA.txt still informs the audience framing for the highlights prompt, but there is no longer a separate "skip this" section.

Supported inputs and adapters include: You can paste any text containing URLs, a block of markdown, or a local file path for .pdf, .mhtml, .mht, .mp3, or .mp4.

• Some converters only run when you pass --force-convert-all or append ## to a specific URL.
• Sources that can be turned into reading pages include YouTube (/watch?, /live/, youtu.be, /shorts/), Twitter/X (/status/), Discord (discord.com), Telegram (t.me), PDFs (.pdf, arxiv.org/pdf/...), DocSend (docsend.com/view/...), GitBook (gitbook, docs.), Discourse (/t/...), Medium articles, Substack, Apple Podcasts, SoundCloud, Streameth, Rumble, Notion, and direct .mp3/.mp4.
• URL normalisers include Medium tracking cleanup, Google Docs export, Wikipedia mobile, Reddit (old.reddit.com), and GreaterWrong (lesswrong.com).
• Warpcast (warpcast.com) is passed through unchanged.

Quick start

1. Install Python deps (managed via pyproject.toml):
   • uv sync
2. Ensure external tools are installed (see below).
3. Fill in .env with required keys.
4. Set the LLM model in config.json.
5. Run:
   • uv run --env-file .env -m lineate "https://..."
   • or just run with no args to use the clipboard.

Usage

uv run --env-file .env -m lineate [text-or-url]

Options:

• --force-convert-all forces conversion for all URLs.
• --summarise uses a semantically complete summary instead of the extracted body.
• --no-summarise keeps the full extracted body.
• --no-open prevents opening processed URLs in your browser.
• --force-refresh forces refresh for all converters (equivalent to appending ###).
• --force-no-convert skips conversion for all URLs.

Environment variables (.env)

Put these in this repo’s .env

Required for core functionality:

• CLOUDFLARE_ACCOUNT_ID
• R2_ACCESS_KEY_ID
• R2_SECRET_ACCESS_KEY

Required for LLM summarisation/title generation:

• LLM_API_KEY – used for text-generation calls to the configured llm.api_base_url.

Required for audio/video transcription:

• OPENAI_API_KEY – used for Whisper transcription.

Required for specific sources:

• Discord: DISCORD_AUTH_TOKEN
• Telegram: TELEGRAM_API_ID, TELEGRAM_API_HASH, TELEGRAM_SESSION_NAME
• Twitter/X: TWITTER_BEARER_TOKEN, TWITTER_CT0_TOKEN, TWITTER_COOKIE
  • Optional: TWITTER_USER_AGENT, TWITTER_XCLIENTTXID, TWITTER_XCLIENTUUID

Notes:

• Telethon will create a .session file named after TELEGRAM_SESSION_NAME in the repo.
• Twitter/X credentials need to be refreshed if the session expires.

How to get each key (links + minimal steps):

• Cloudflare R2 (CLOUDFLARE_ACCOUNT_ID, R2_ACCESS_KEY_ID, R2_SECRET_ACCESS_KEY): create an R2 API token with access to the configured bucket, then copy the account id and credentials from Cloudflare. Docs: Cloudflare R2 authentication.
• LLM_API_KEY: create a key for whichever OpenAI-compatible LLM endpoint you configure. For OpenRouter, see: Authentication.
• OPENAI_API_KEY: create/view your key in the OpenAI API keys page for Whisper transcription only. Doc: Where do I find my OpenAI API key?.
• Telegram (TELEGRAM_API_ID, TELEGRAM_API_HASH): create an app on Telegram’s developer portal and copy the values from API development tools. Doc: Creating your Telegram Application.
• Discord (DISCORD_AUTH_TOKEN) — unofficial: extract your user token from the browser’s developer tools (Network → request headers → Authorization). Doc: discordpy-self “Authenticating”.
• Twitter/X (TWITTER_BEARER_TOKEN, TWITTER_CT0_TOKEN, TWITTER_COOKIE) — unofficial: use browser devtools on x.com to copy cookies (auth_token, ct0) and the request Authorization: Bearer ... header. Docs: Export X cookies (auth_token, ct0), Extract Bearer token from request headers.

Config (config.json)

Set the LLM endpoint and model under llm:

{
  "llm": {
    "api_base_url": "https://openrouter.ai/api/v1",
    "reasoning_effort": "high",
    "model": "qwen/qwen3.6-plus-preview:free",
    "pdf_model": "openai/gpt-5.4",
    "pdf_reasoning_effort": "medium"
  }
}

llm.model and llm.reasoning_effort are the defaults for the non-PDF text-generation paths. llm.pdf_model and llm.pdf_reasoning_effort are used only for the PDF-to-Markdown conversion path.

System dependencies

• uv – Python package manager used to sync/install deps.
• ffmpeg – required by pydub and yt-dlp for audio conversion.
• yt-dlp – required for some sources (e.g., Rumble).
• Clipboard helper for pyperclip:
  • Linux (X11): xclip or xsel (or wl-clipboard on Wayland)
  • macOS: pbcopy/pbpaste (built-in)

Install (CLI dependencies):

• macOS (Homebrew):
  • brew install ffmpeg yt-dlp uv
• Linux (Fedora):
  • sudo dnf install ffmpeg yt-dlp uv

Python dependencies used by lineate

Installed via uv sync from pyproject.toml. Key runtime packages:

• requests, loguru, python-dotenv, pyperclip
• openai (OpenRouter/OpenAI-compatible client for LLM calls, plus Whisper transcription)
• pydub (audio slicing)
• youtube-transcript-api, bs4
• telethon (Telegram)
• yt-dlp (Rumble)
• soundcloud-lib
• python-dateutil (Discord timestamps)