Typeset LLM generated notes/reports (or GitHub flavored markdown (GFM))

This repo turns Markdown into publication-style PDFs with Pandoc + XeLaTeX. You get a small CLI for quick renders and a FastAPI endpoint for programmatic jobs. Fonts, a template, and a Lua filter ship with the tree so the output looks consistent without hunting for assets.

What’s here

• latexit/[Note: latexit was the local repo name I used] — active code: src/cli.py, src/api.py, template, Lua filter, Dockerfiles, tests, samples, fonts.
• Planning artifacts (coreIdea.txt, conceptmap.txt, md2pdf.ipynb, PDF exports) that show the early roadmap (now removed, can be requested over email).

Prereqs (host)

• Python 3.11
• Pandoc ≥ 3.x on PATH
• XeLaTeX toolchain (TinyTeX works. see latexit/tlpkgs.txt for the tested package set)
• Fontconfig can see the bundled fonts (assets/fonts/IBM_Plex_Serif_re, assets/fonts/Fira_Code, assets/fonts/Latin-Modern-Roman)
• Docker optional if you prefer everything pre-baked

Quick start (local)

cd latexit
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
# Make sure pandoc and xelatex run (install TinyTeX if needed).
# On Linux/macOS, copy fonts and refresh cache:
#   cp assets/fonts/*/*.{ttf,otf} ~/.local/share/fonts && fc-cache -fv

CLI usage

• Basic: python -m src.cli samples/sample.md -o out/sample.pdf
• STDIN: echo '# Title' | python -m src.cli --stdin -o out/stdin.pdf
• Pick a template: python -m src.cli note.md -t templates/vintage.tex -o out/note.pdf
• Wrapper script: ./convert.sh path/to/file.md → out/<name>.pdf

The CLI insists on real files, uses filters/landscape-6col.lua for wide tables, and raises clear errors if the source/template/filter is missing.

FastAPI service

cd latexit
uvicorn src.api:app --host 0.0.0.0 --port 8000 --reload

• /health → { "status": "ok" }
• /convert accepts JSON, form data, or raw body with markdown_text (or text) and optional theme; returns application/pdf.
• Guardrails: 10 kB payload cap, LaTeX sanitiser strips \\write18/\\input/openin/openout, SlowAPI rate limit of 60 req/hour/IP.

Example:

curl -o out/api.pdf \
  -H "Content-Type: application/json" \
  -d '{"markdown_text":"# Hi\n\nConverted by the API","theme":"vintage"}' \
  http://localhost:8000/convert

Docker (self contained)

Includes Pandoc, TinyTeX, fonts.

cd latexit
docker build -t gptnotes-pdf .
docker run --rm -p 8000:8000 gptnotes-pdf

Run the CLI inside the image:

docker run --rm -v "$PWD/samples":/data gptnotes-pdf \
  python -m src.cli /data/sample.md -o /data/sample.pdf

Tests

cd latexit
python -m pytest -q

The suite exercises both the CLI and API. Pandoc + XeLaTeX must be installed or the PDF assertions will fail.

Customisation

• Drop new .tex templates into templates/. pass -t (CLI) or theme=<name> (API). Unknown themes fall back to vintage.tex.
• Edit filters/landscape-6col.lua if you want different table widths or to disable the landscape flip.
• quickscripts/ holds small demo commands. samples/ contains real Markdown fixtures to tweak.
• Dockerfile.probe plus tlpkgs.txt capture the minimal TeX packages that successfully compile the template.

Notes

• Keep inputs under 10 kB to stay within the API limit and avoid long TeX runs.
• If PDFs come out blank, verify font availability and that xelatex works from your shell.
• Two trees (latexit and latexit copy) exist—do your work in latexit to match tests and Docker context.

Samples (Generated)