Skip to content

AI Block

This feature is Experimental

The ai { ... } block generates Markdown files using AI through OpenRouter. You describe what you want in a prompt and Scimon writes the resulting Markdown straight into your output folder.

Requirements

You need an OpenRouter API key set as the OPENROUTER_API_KEY environment variable in your Scimon .env file:

scimon options write-env

OPENROUTER_API_KEY="sk-or-..."

If the key is missing, the block reports a friendly error instead of stopping the run.

Usage

Each line inside the block is a prompt followed by the output file:

"<prompt>" as "<filename>.md"
  • The output extension decides the format:
  • .md (or no extension) saves the raw generated Markdown.
  • .pdf renders the generated Markdown to a styled PDF using the generic template (same pipeline as Markdown render). The style variable is honored for the PDF's CSS, falling back to the default stylesheet.
  • .epub packages the generated Markdown as an EPUB. The book title comes from the output file name and the author from package.yml (falling back to Scimon).
  • The .md extension is added automatically if you omit it.
  • Files are written inside the folder defined by the path variable, so they are also picked up by the server command.
  • Add !ignore to skip an entry without removing it.

Example

path "downloads/"

ai {
    "Write a short article about the Rust programming language" as "rust.md"
    "Explain quantum computing for beginners" as "quantum.pdf"
    "Write a beginner's guide to Rust" as "guide.epub"
    "Draft release notes for version 2.0" as "drafts/release.md" !ignore
}

Here rust.md is saved as Markdown, quantum.pdf is rendered to a PDF, and guide.epub is packaged as an EPUB.

Choosing a model

By default the block uses the openai/gpt-4o-mini model. You can override the model per entry with with "provider/model", using any model id available on OpenRouter:

ai {
    "Summarize the history of the internet" as "internet.md" with "anthropic/claude-3.5-sonnet"
    "Write a haiku about Rust" as "haiku.md" with "meta-llama/llama-3.1-8b-instruct"
}

Notes

  • The model is asked to reply with the Markdown body only (no surrounding code fence), so the output is ready to use as-is.
  • Combine it with the readme or markdown render features to turn the generated Markdown into styled PDF.
  • The block runs after downloads but before merge, covers and compress, so generated PDFs are merged, get a cover extracted and are included in the archive. The full order is: downloads → ai → merge → covers → compress → qrcode → math → convert.