Roadmap

The authoritative plan is PLAN.md in the repository. This page is the summary. Milestones are named for what they let you do, not for versions; the version number is set by the release process from the commit history.

M0, foundations: done, 2026-09-03

The library and the command line. Every container format (TTF, OTF, TTC, WOFF, WOFF2), the full metadata model, the SQLite index with full-text search, duplicate detection across containers, health checks, CSS export, the HTML specimen, coverage queries, JSON output against published schemas, and a CI matrix on Linux, macOS and Windows with license and advisory auditing. Decision records for fontations, SQLite, the deferred desktop shell, the license and WOFF decoding.

M1, manage: done, 2026-09-04

The font manager proper, in the terminal. Re-scoped on 2026-09-03 from a desktop application to command line plus terminal browser.

1. Organise. Tags, collections with JSON import and export, sources, family grouping, facet counts, richer filters.
2. Activate. Native activate, deactivate, install and uninstall on Linux, macOS and Windows; conflict detection with --replace; activation state in the index; restore for login agents.
3. Watch. source add scans immediately; watch follows every watched source with debounced incremental rescans.
4. Preview. Shaped, rasterised previews in the terminal over kitty graphics, iTerm2 images, sixel or half-block text, with axis coordinates and feature toggles.
5. Browse. fontina ui: search, facets, families and faces, details, previews, tagging and activation from the keyboard.
6. Ship. Completions and man pages in the archives; .deb and .rpm from the release workflow.

Still to do before 1.0: a Homebrew formula, winget and Scoop manifests, an AUR PKGBUILD, and the rename.

M2, typography

In the browser: axis sliders with named-instance snapping, feature toggles, a glyph map by block with codepoint search, a license viewer. The compare and waterfall views were built and then removed: a terminal cell is one pixel wide and two tall, so type drawn into one is a picture of something that is not the font (ADR 0010). The browser shows what is in a font; s opens a real specimen in a browser and fontina preview draws a true image where the terminal has a protocol for one. check grows toward fontbakery parity where it is cheap; identifiers never change. Optional login-agent packaging (systemd user unit, LaunchAgent, Run key), off by default. An optional offline Google Fonts index, separately packaged, opt-in.

M3, ecosystem and shells

Team sharing through plain folders (collection JSON with relative paths over Syncthing, git or any synced directory), tag synchronisation with Finder tags and Windows file properties, and a plugin surface that is only the command line and JSON. A graphical shell, as one more client of the core, only if the terminal browser leaves a real gap; it would have to meet the same budgets and follow each platform’s own design conventions, Linux first.

M4, ask

Everything the index knows, askable. fontina has read the whole of a font since M0 — every variable axis and its range, the language systems each script declares, how many codepoints of a script a face actually covers — and stores it all. The filters lag behind it in three places: a variable font is matched on its default instance, so a face whose weight axis spans 200 to 800 is not found by a search for 400; there is no way to ask for a language at all; and a script filter cannot ask for two scripts at once or for how much of one a face covers.

Spacing is the fourth: whether a face is monospaced is read from the font and stored, and cannot be filtered or counted, which on a working developer’s library is the most useful single division there is.

None of that needs a new parse or a rescan. The data is already in the index; only the questions are missing.

The fifth gap is a different kind. fontina groups faces by what the font declares — the family in its name table — and by what a person asserts, with tags and collections. It derives no grouping from what a face actually is, so where a declared family is wrong there is nothing to fall back on. A real library of 149 fonts reports twenty families and holds about eight typefaces. The answer is not a second kind of family the index invents and stores, which would mean encoding somebody else’s naming convention as fact; it is a question you can ask of one face — what else here covers nearly the same characters — and answer from evidence the index already has.

M5, ship

Four milestones in, you can install fontina with a package manager on Linux and with none on macOS or Windows. The release builds a .deb and an .rpm; everywhere else there is an archive to download and a binary to move onto your PATH by hand. Every desktop is meant to be first class, and on the evidence of how you install it, that is not yet true.

So: a Homebrew tap, a Scoop bucket, a winget manifest, an AUR package — kept current by the release itself rather than by somebody remembering, and each one installed for real in a clean machine and tested there, the way the .deb and the .rpm already are.

Two performance budgets are written down and not measured, because both need a real terminal to measure: how much memory the browser holds at rest, and how long it takes to repaint. A budget nothing measures is a wish with a table row, so either they get a harness or they stop being called budgets.

Nothing here makes fontina do anything new with a font. It makes fontina something you can install, which after four milestones of features is the question worth asking.

The product programme

Everything above is what fontina does. Running alongside it is a second programme about what it is like to use, and the evidence for it is one screen:

$ fontina
error: 'fontina' requires a subcommand but one was not provided
  [subcommands: scan, list, families, facets, tag, collection, source, activate, ...]

The word error, thirty-four nouns, and no opening move. Nothing there is a bug — every part is a reasonable default — and together they are a toolkit rather than a product.

Five milestones, in the order somebody meets them.

Arrival is the first five minutes. fontina on its own does the obvious thing; an empty index is a state with a plan rather than an error to report.

One verb is acting on the thing rather than on a sentence about it: a spine through the thirty-four commands, the browser showing the command it would have run, and the words a person would use instead of the schema’s.

Judgement is the actual job — helping somebody choose. Comparison that does not lie, differences stated in the words of the difference, and is this the right one answered with evidence rather than deflected to a table.

The shelf is a library rather than a pile: sets you assemble by acting, that show what is in them, and that open without this program.

Plain sight is the half that keeps the rest honest. What fontina did and how to undo it, what a font lets you do, and where it came from — all legible without asking.

The line all five are measured against: decide the defaults, show the reasoning, never hide the lever. A default may be chosen for you; a capability may not be taken away, and nothing the interface can do is unavailable to a script.

Never

Font editing. Format conversion or subsetting. Cloud synchronisation. Accounts. Telemetry. An Electron shell.