Skip to content
ZteinPublic

About

Documents to LLM-ready text, including what only the figures say. Fully local.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

figmark

Turn documents into text that an LLM can use — including what only the figures say: charts, diagrams and process sketches become structured, readable text. Runs fully locally.

Status: phase 1 done, at its decision point — a first converter on Docling, measured against the floor and ceiling (docs/phase-1.md); the yardstick is in docs/phase-0.md.

uv sync
cp figmark.example.yaml figmark.yaml     # point `figures` at an OpenAI-compatible VLM
uv run figmark convert report.pdf slides.pptx -o out/
uv run figmark cache stats             # descriptions are cached locally, keyed on image + model + prompt
  • docs/PRD.md — what is being built, how it is measured, and the hypotheses each phase has to pass (in Swedish).
  • eval/ — the public evaluation documents and question sets.

This is a new project. The first figmark (2026) is archived at Ztein/figmark-legacy.

MIT licensed.

About

Documents to LLM-ready text, including what only the figures say. Fully local.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages