Fontlore: typeface intelligence for AI agents

Why it exists

Font decisions fail on unverified facts: wrong licence assumptions, missing script coverage, stale catalog data. Fontlore forces verification at the foundry source and marks everything else as null or unverified.
AI agents need a base rate. Font fame is a weak prior; scoring on category fit, tone, coverage and licence gives a shortlist that can be defended line by line.
Typography knowledge is scattered across foundry pages, licence texts and memory. The 95 references consolidate the operational parts.

What is inside

Data layer

Three JSON files: 165 font records, 66 curated heading-and-body pairings and 71 tools. Every font record carries a source URL.

Decision engine

A standard-library Python CLI that hard-filters by script and licence, scores against a rubric and prints ranked reasons.

References

95 curated references across 9 categories: sourcing, selection, identification, licensing, scripts, technical, accessibility, applications and workflow.

Guides and site

632 generated guides and a static tabbed library of 732 pages with no cookies, no analytics and no external requests.

Quickstart

git clone https://github.com/izynu/fontlore.git
cd fontlore
python3 scripts/decide.py --project dashboard --script latin --tone neutral

The data rule. Unverifiable values are null, never invented. Licences are recorded as SPDX identifiers and verified at the font's source URL, not at aggregators.

Fontlore