Playground

Roadmap

Where Monoline honestly stands today and what comes next.

This page is docs/ROADMAP.md, rendered from the repository. It says where the kit stands today, including what is not done.

Last updated 2026-11-12. This replaces the earlier task table. The component list in docs/research/coverage-matrix.md is complete for the 72 components that appear in three or more of the ten libraries we surveyed; this roadmap is about everything else that decides whether people adopt a UI kit.

Where we honestly are

AreaState todayEvidence
Components114 components and 23 hooksregistry/index.json, 137 entries
Tests1182 unit tests with axe and keyboard checks; every demo renders in a testnpm run check
Real-browser checksaxe (with colour contrast), overflow and console errors on every docs page in 3 themes at desktop and 375px, orange accentscripts/audit.mjs
Look options8 accents, 3 fonts, 3 corner styles, 3 densities, RTL passdocs/rtl.md, packages/css/src/options.css
Docspage per component, search, TOC rail, per-example code, props, a11y, stylinglive site
AI-readablemeta.json, registry/, llms.txt, llms-full.txt, Markdown per page, AGENTS.mdgenerated
Bundle sizeone Button is 1.7 KB gzip, all CSS 16.4 KB gzip (after the fix below)npm run size
Publishednot on npm yetn/a
Used by anyonenon/a

A bug we found by measuring. The first build shipped one file, so importing a single Button cost 103 KB gzip (every component and all of Radix). The package is now built one file per module and a bundler drops what the app does not use. npm run size enforces budgets so this cannot regress silently.

What we can and cannot compete on

MUI, Ant Design, Mantine and shadcn/ui have years of community, thousands of issues fixed and large ecosystems. We will not match that breadth of trust in months. Competing means being clearly better at a few things people care about, and then earning trust:

  1. Small and fast. Plain CSS, no runtime styling, per-component bundles. Target: a 10-component app under 25 KB gzip of JS and about 10 KB of CSS.
  2. Accessible, with proof. Axe plus a real-browser audit in CI across every theme and accent, and published results. Others claim it; we show it.
  3. AI-first. Registry, llms.txt, Markdown per page, an MCP server, and a CLI that copies source. Only shadcn/ui and Mantine do part of this today.
  4. One distinct, themeable identity. Monospace and thin rules by default, but font, corners, density and eight accents on one switch, so people who dislike the default still can use it.
  5. Copy or install. Both routes work from the same source.

Skip on purpose: 3D, WebGL, glow and aurora effects (Aceternity, parts of Magic UI); enterprise-only widgets such as Transfer, Cascader and Watermark until someone asks; Tailwind as a dependency.

Phase 0: ship it safely (weeks 1 to 2)

Goal: someone can install it, trust it and see it working.

StepWhy it mattersDone when
0.1 Per-module build and size budgetone Button must not cost 100 KBnpm run size passes in CI; Button under 2.5 KB gzip
0.2 Green CI on GitHubci, audit and visual jobs have never been fully greenall three jobs pass on master
0.3 Publish @monoline/css, @monoline/react, monoline to npmnothing else counts until people can install itnpm i @monoline/react works on a clean machine; release workflow with provenance
0.4 Real install test on clean projectsproves the READMEVite, Next.js (App Router), Astro, Remix starters each render a Button and a Dialog in CI
0.5 React 18 and 19 matrixpeer range says >=18tests run on both
0.6 Real-browser interaction testsjsdom cannot test focus, layout or popovers properlykeyboard flows for dialog, menu, select, combobox, tabs, date picker, command in real Chrome in CI; Firefox and Safari smoke
0.7 CSS splitpeople who use 3 components should not load 16 KB of CSScore.css plus per-component CSS files; @layer so app styles always win

Phase 1: make it consistent (weeks 2 to 5)

Goal: no surprises. This is where small libraries lose to big ones.

StepDetail
1.1 API auditsame prop names everywhere: size (sm/md/lg), variant, disabled, invalid, loading, asChild, controlled/uncontrolled pairs; compare against shadcn/ui, Mantine and Base UI conventions and fix outliers
1.2 State attributesevery stateful part exposes data-state, data-disabled, data-invalid the same way, so styling and tests are predictable
1.3 Slot and class namingsame naming for parts (-header, -content), documented in each page's Styling section
1.4 Token completenessz-index scale, motion tokens, focus ring tokens, elevation, breakpoints; no component hard-codes a size
1.5 Behaviour paritycheck each interactive component against its WAI-ARIA pattern: dialog, menu, select, combobox, tabs, tree, grid; fix gaps
1.6 Bug sweepevery issue found by the audit, the interaction tests and a manual pass at 375px and 1440px in all themes
1.7 Docs playgroundlive controls (variant, size, state) above the first example, like Mantine

Phase 2: fill the real gaps (weeks 5 to 9)

Components that Mantine, Ant Design or Chakra ship and we do not:

  • Date range picker, range calendar, time field
  • Tour (guided onboarding), resizable panels and splitter, mentions, tree select
  • Overflow list, copy button, image gallery and lightbox, notification centre
  • Sortable list with keyboard dragging, virtualised select and listbox
  • Real tests against react-hook-form and TanStack Form (needs an extra test workspace)
  • Full app shell: sidebar, navbar and breadcrumbs working together

Phase 3: ecosystem (weeks 9 to 14)

StepDetail
3.1 Templatesdashboard, auth, landing, settings and blog starters as real apps in examples/
3.2 Blocks gallerya page showing every block with copy buttons
3.3 MCP serveran MCP server over the registry (list, search, get a component) so any AI tool can use Monoline
3.4 Theme presetsa gallery of named presets (paper, terminal, slate, soft) and a theme builder that exports CSS and JSON
3.5 Figma kittokens and components for designers
3.6 Editor supportVS Code snippets and a props autocomplete
3.7 Versioned docs, changelog, migration guidesneeded the first time something breaks

Phase 4: earn trust (ongoing)

  • Manual screen-reader testing with NVDA, JAWS and VoiceOver on the core 15 components, results published
  • A benchmark page: bundle size, axe score and Lighthouse for Monoline against the other libraries on the same demo app
  • Governance: RFC process, maintainers, release train, security policy, good-first-issue labels, a public roadmap
  • Real users: dogfood on your own projects first; fix what they hit; then ask three outside projects to try it

Quality gates (always on)

  • npm run check: contrast (3 themes × 8 accents, code colours), no raw colours, no physical CSS properties, doc comments on every prop, 1182+ tests
  • npm run size: per-component and CSS budgets
  • node scripts/audit.mjs: real-browser axe, overflow, console errors
  • node scripts/audit.mjs --visual: screenshot comparison on pull requests

Measures of success

MeasureTarget
One Button, gzipat most 2.5 KB
Typical 10-component app, gzipJS under 25 KB, CSS under 10 KB
Axe violations across all pages, 27 theme/accent pairs, 2 viewports0, in CI
Time from npm create to a styled pageunder 2 minutes
Clean install works in Vite, Next.js, Astro, Remixall four, in CI
Docs: every component has installation, usage, example per variant, API, accessibility, styling100 percent

Risks

  • Solo maintenance. 137 units is a lot to keep correct. Mitigation: the gates above, strict scope, and not adding components without tests and a demo.
  • Trust. Nobody adopts an unproven kit. Mitigation: publish the audit results, ship templates, dogfood.
  • Radix dependency size. Select costs about 32 KB gzip and Dialog 14 KB because of Radix. If that matters, replace the heaviest primitives with native or smaller implementations one at a time, behind the same API and tests.
  • Version drift in Radix and React. Mitigation: React 18 and 19 matrix and dependency updates in CI.