Fontlore

Open-source skill - MIT licensed

Typeface intelligence
for AI agents.

Fontlore gives AI assistants a curated knowledge base to research, select, pair, license, and implement fonts with evidence. Real catalogs, real licenses, real tradeoffs. No vibes.

Browse the library View on GitHub

MIT · 727 guides and references · no cookies, no analytics

727Guides and references
165Font records
66Vetted pairings
71Tools and sources
9Reference categories

The library

Nine curated reference categories

Every category answers one operational question. Each guide follows one house style, cross-links its neighbors, and ends in a checkable artifact.

Accessibility

Legibility, contrast, WCAG, dynamic type

10 guides →

Applications

Typography per medium: web, print, email, UI, and more

10 guides →

Identification

Naming a font from an image or specimen

10 guides →

Scripts

Per-script typography: Latin, CJK, Arabic, Devanagari, and more

10 guides →

Selection

Choosing between candidates with a rubric

10 guides →

Sourcing

Where to find fonts and how to read catalogs

10 guides →

Technical

Formats, loading, performance, variable fonts

10 guides →

The engine

From brief to handoff, deterministically

01

Brief

Project type, audience, tone, medium.

02

Constraints

Scripts, license scope, sizes, delivery.

03

Score

Rubric ranking with printed reasons.

04

Pair

Curated heading plus body, with rationale.

05

Handoff

CSS, tokens, and evidence to re-check.

Read the full pipeline in MASTER-SKILL.md →

Data integrity

Verified, or null

Source URLs on every record

Each font points at its own foundry page or repository. Verified at the source, not at aggregators.

Null, never guessed

Unverifiable values stay null. The schema makes missing data visible instead of plausible.

CI-gated on every commit

validate.py checks JSON schemas, pairing integrity, house style, and every cross-link before merge.

Licenses as SPDX ids

OFL-1.1, Apache-2.0, UFL-1.0, ITF-FFL-1.0, MIT. Anything else must be read in full, in the guide.

Quick start

Run the decision engine

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

Hard-filters by script and license, scores against the rubric, prints ranked reasons, and suggests curated pairings.

Build the site yourself

This entire site is generated from the data layer by a single stdlib-only script. After any data change:

python3 scripts/build_site.py

727 pages, no dependencies, no tracking. Deployed to GitHub Pages by the included workflow.