A curated Angular 22 UI component library, forked from the last MIT release of PrimeNG (21.1.9), with the MIT provenance proven by machine rather than asserted in prose.
npm install @anguless/angulux @primeuix/themesimport { provideAngulux } from '@anguless/angulux/config';
import Aura from '@primeuix/themes/aura';
bootstrapApplication(App, {
providers: [provideAngulux({ theme: { preset: Aura } })]
});Why the second package. angulux ships no theme presets of its own, so the preset comes
from @primeuix/themes — PrimeTek's, and MIT through 2.0.3. angulux itself still installs
with zero PrimeTek packages in its dependency tree; drop the second package and
provideAngulux() runs unstyled but works.
It is declared as an optional peer ranged ^2.0.0, and the ceiling is deliberate:
3.0.0 is the first commercial release. Be clear about what that buys you — a peer range is
a warning, not a lock. npm install @primeuix/themes@3 still succeeds; it just prints an
ERESOLVE warning naming the range it broke. That is enough for the case that actually
happens (a routine npm update, or an agent bumping dependencies, sliding across a license
boundary in silence) and it is not enough for anything else. If the boundary matters to you,
pin the version exactly and run a licence guard in your own build. That gate is published as
its own MIT package, angulux-license-guard — it reads your
lockfile, fails closed on any PrimeTek package it cannot verify, and has no dependencies of
its own. It works in any project; it does not require angulux.
angulux is not a drop-in replacement for PrimeNG, and it is not trying to be the community's continuation of it. Three facts decide whether you should keep reading:
- It ships 65 of PrimeNG's 117 modules. Editor, Tree, OrganizationChart, PickList,
OrderList and Listbox are among those not ported — all 52 are listed in
packages/angulux/attic/. See What angulux is NOT. - Selectors were renamed.
p-buttonisagl-button,pTooltipisaglTooltip. CSS class names (p-*) were deliberately kept — see NOTICE for why. - Bus factor is 1. One maintainer, a permanent obligation of two Angular majors per year. MIT, best-effort, no SLA. If that is disqualifying for you, it should be.
If what you need is "I run PrimeNG 21 and I want to keep going with minimal change", the
honest answer is that Optimus UI fits that
better than angulux does: it keeps the p-* selectors and carries the full module set.
We would rather tell you that here than collect an issue about it later.
PrimeNG 22 moved to a commercial license. Every release up to and including 21.1.9 was published under MIT and stays MIT permanently — that is what a published license grant means. angulux is a fork taken at that boundary, for teams whose blocker is control and legal certainty, not price.
The interesting part is not the fork. It is that the boundary is enforced by machine.
Eighteen gates run on every commit and every pull request. They do not check style — each one closes a class of failure that has already happened in this repository.
| Gate | What it refuses to let through |
|---|---|
check:catalog |
a dependency version drifting off the pinned catalog |
check:scope |
the transitive closure drifting off the 65 warranted modules |
check:license |
any PrimeTek package from a post-MIT release entering the tree — and, separately, any licence-flagged material sitting in ref/ that PROVENANCE.md does not declare |
check:names |
PrimeNG names surviving in selector, API or trademark positions — including inside bare strings, JSDoc links and DOM attributes |
check:risk-coverage |
the browser gate's scope drifting off the risky decorators computed from source |
check:language |
non-English text reaching a public repository |
check:public-tree |
internal-only paths sitting in the git index (.gitignore does not untrack) |
check:scope-names |
a bare angulux/angulux-* package reference surviving where the scoped @anguless/* name belongs |
check:tsup |
the bundler leaking outside the four forks' own build scripts |
check:action-inputs |
a workflow passing a with: key that its action does not declare — YAML nobody validates until it silently does nothing |
check:release-plugins |
a release train listing the same plugin twice, or naming the wrong rail — the first published 22.2.1 with every changelog entry printed twice, and nothing else noticed because the version and the package were both correct |
check:facet-single-route |
a template slot growing a second route, after the whole library was collapsed onto one <ng-template #x> mechanism |
check:corpus |
corpus/corpus.json drifting off the source it is generated from — a generated artifact nobody regenerates is a lie with a timestamp, and nothing else breaks when it goes stale |
check:demo-code |
the documentation site claiming something the library does not do — a demo whose shown code stopped matching the demo that ran, a page entry pointing at a different file than it names, or a module documented on the public web that no release contains |
check:peer-licence |
angulux's own manifest telling a consumer's package manager to install a commercially-licensed PrimeTek package — check:license proves the tree installed here is clean, which is a different question from what the published version ranges admit |
check:install-scripts |
a dependency running arbitrary code during pnpm install without anyone deciding it should — the release job installs while holding id-token: write, so one compromised transitive package could mint an npm credential and publish as this project with a valid attestation |
check:module-counts |
the project's central claim — "65 of PrimeNG's 117 modules" — going stale in the fifty-odd places it is written down, including the MCP tool descriptions that leave the repository; the three counts are re-derived from the closure and the attic directory, and the attic cost table is compared as a set, so a promoted module cannot linger in it while every number on the page stays right |
check:gate-count |
this list going stale — the count and the gate names are re-derived from package.json and checked against every place they are written down |
npm run check # all eighteen, no build neededThree further checks run after the build rather than inside that suite, because none has anything to inspect until there is a build to inspect. All three look at the artifact instead of the source it came from:
| Check | What it refuses to let ship |
|---|---|
check:publishable |
packs every package and reads package.json back out of the tarball, so what gets inspected is the exact bytes npm would receive. It exists because a guard that read the source tree instead once passed while three packages were about to publish an uninstallable workspace: dependency |
check:dts |
compiles the emitted .d.ts with skipLibCheck: false. Every tsconfig here sets it to true, which means our own declarations are the one thing no build in this repository type-checks — a consumer who turns it off compiles them for real, and if they do not stand up alone that consumer's build breaks while ours stays green |
check:signal-forms |
compiles Signal Forms' [formField] on every form control against the built package, with strictTemplates on. [formField] binds a field's state into inputs by name — min typed as the field's value, pattern as readonly RegExp[] — so a narrower input is a type error in every consumer's template. Four controls shipped that way, invisible here: the library builds with strictTemplates off and the specs compile just in time. The form controls are read from the corpus, so a newly shipped one must join the fixture |
All three refuse to run against nothing, and two go further: check:dts and
check:signal-forms report success only after a deliberately broken file has been put through
the same program and rejected for the reason it was written for. A type-check aimed at zero
files does not say "I found nothing" — it says "no errors".
Two of them — check:license and check:catalog — also run daily on a schedule, and
that is not redundancy. A pull-request run only proves the tree was clean at our last
commit, while the boundary those two enforce lives outside this repository: PrimeTek
decides which of their releases is the last MIT one, and that can change on a day when
nobody here pushes anything. The rest gate our own code, so a commit is the only
moment they can be wrong.
On top of that: provenance/manifest.json records SHA-256 checksums and registry publish
timestamps for every archived MIT artifact, and a browser gate renders 16 of 16
risk-flagged decorators in a real Chromium and asserts on the result — because a green
type-check cannot see a render bug.
It was 15 of 15 until the guard was asked what it actually matched. Angular 22 renamed
ChangeDetectionStrategy.Default to Eager and deprecated the old spelling; the guard
recomputed its risky set by grepping for the new one, so three CheckAlways components —
agl-scroller and two in table.ts — were outside the guarded set while the gate
reported full coverage. The enum makes the two spellings literally equal
({ OnPush: 0, Eager: 1, Default: 1 }). A guard that matches a token rather than a risk
class is a guard a rename can switch off.
That gate has a second half, for the bugs a green spec suite cannot see. It opens all 51 module pages of the showcase — the markup a reader copies, rather than the reduced markup a spec sets up — and measures whether what the library draws survives its own CSS: an element's rectangle intersected with every clipping ancestor. It exists because a badge shipped that was present, correct and 32% visible.
Evidence measured at 22.2.3, reproducible from a clean checkout:
18/18 gates exit 0
library build exit 0 · 211 entrypoints
inherited spec suite 3809 SUCCESS
browser gate 70 tests · 14 behaviour + 56 visibility
runtime deps tslib + four first-party angulux-* packages. Zero PrimeTek.
The commitment is a green build within 14 days of every Angular major RC, verified by a
nightly job that builds against @angular/core@next and reports publicly. angulux's major
version is locked to Angular's major, so angulux@22.x targets Angular 22 and nothing else.
Every shipped module is inside the closure that the gates cover. The other 52 live in
packages/angulux/attic/ — verbatim upstream source, not built,
not published, kept in the open as a roadmap. Promoting one is a well-defined contribution; see
CONTRIBUTING.md.
- Not a drop-in replacement. Selectors changed. A migration codemod is planned, not shipped.
- Not feature-parity with PrimeNG. 65 of 117 modules. Missing, among others: Editor, Tree, TreeSelect, OrganizationChart, PickList, OrderList, Listbox, Terminal, Dock. (TreeTable is shipped; Tree is not.)
- Not affiliated with PrimeTek or Google. See NOTICE.
- Not a supported product. MIT, best-effort, no SLA, no response-time promise.
- Not a fork made in anger. PrimeTek gave the community ten years of MIT work and confirmed that past MIT releases stay MIT. This project exists because of that grant, not in spite of it. Keep that tone in issues and pull requests.
| Angular | ^22.0.0 |
| TypeScript | 6.0.x (Angular 22's compiler requires >=6.0 <6.1) |
| Node | >=22 |
| Package manager | pnpm 9.6.0 (via corepack) for development; any manager to consume |
chart.js is an optional peer, needed only by the chart module. There is no
@angular/cdk peer — it was dropped on purpose; nothing in the warranted closure needs it.
| Package | Version | Purpose |
|---|---|---|
angulux |
22.x |
the component library — major locked to Angular's major |
angulux-styled |
1.x |
style engine, forked from @primeuix/styled |
angulux-utils |
1.x |
DOM/object utilities, forked from @primeuix/utils |
angulux-styles |
1.x |
base style definitions |
angulux-motion |
1.x |
animation primitives |
The four forked packages version independently on 1.x — they do not depend on Angular,
so locking them to Angular's major would be a lie about their compatibility.
corepack pnpm install
npm run check # the 18 gates
# the four forked packages build BEFORE the library — dependency order matters
for p in utils styled motion styles; do
(cd packages/angulux-$p && corepack pnpm run build)
done
corepack pnpm --filter angulux run build # → 210 entrypoints
corepack pnpm --filter angulux run test:unit # the inherited spec suite
corepack pnpm --filter @angulux/verify run build
npx playwright test --config e2e/playwright.config.ts # browser gatepnpm runs through corepack and is not on PATH; use corepack pnpm. Karma needs
CHROME_BIN set if Chrome is not at the default location.
https://angulux.anguless.com is the documentation site: a page per module with runnable
demos, and an API reference for all 65 modules generated from corpus/corpus.json — the same
corpus that produces the files in the next section, so the two cannot say different things
about the same API.
It is built from apps/showcase, which is a workspace app and is never published to npm.
Its demos compile against packages/angulux/dist, not against src, so every demo on the
site is a demo of the package a user actually installs. The code shown under each demo is
EXTRACTED from the demo that just ran rather than written alongside it, and check:demo-code
fails the build if a demo stops being extractable, if a page entry names one demo and loads
another, or if a module is documented that no release contains. See apps/showcase/README.md
for the rules a demo file has to follow.
Angulux is new, so no model has seen it. Ask an assistant about it and it will answer with
PrimeNG's p-* API, because that is the only thing in its training data. These files exist
so it does not have to guess:
| URL | What it is |
|---|---|
| the index page | what is here, for a reader who trimmed the URL back to the host. The host root now serves the documentation site for humans, so this page lives at /llms |
llms.txt |
the index, in the llms.txt format |
llms-full.txt |
every module's API in one file |
<module>.md |
one page per module, e.g. button.md |
All of it is generated from this repository's own TypeScript into corpus/corpus.json, and
check:corpus fails the build if the committed corpus is not byte-identical to a fresh
generation. Nothing here is hand-written, so do not edit the corpus or the pages — the
next generation overwrites them. Fix the JSDoc in the component instead.
Three things worth knowing before you rely on it:
llms-full.txtis not part of the llms.txt specification. The spec defines one file,llms.txt. The full variant is a widely-followed convention, and we serve it because assistants look for it — not because a standard requires it.- Most inputs do not document a default. Only 127 of 1205 declare
@defaultValue, so pages say not documented rather than leaving a blank that reads as "there is no default". - 69 inputs are deprecated and each page names the replacement. That number is the single best reason for this to exist: without it, an assistant recommends them confidently.
The generator can only read packages/angulux/** and tools/** and makes no network
request. That boundary is enforced in code and asserted by a test, because the PrimeNG
documentation site was never MIT — only the source was, and only up to 21.1.9.
packages/angulux-mcp serves the same corpus as five tools over MCP stdio, so an assistant can
query it instead of reading a page. It is not published — run it from a checkout:
claude mcp add angulux -- node /absolute/path/to/angulux/packages/angulux-mcp/bin/angulux-mcp.mjs
The tool worth knowing about is check_usage. Give it a selector, an import specifier, a module
name or a list of inputs, and it answers before you write the code:
check_usage({ selector: 'p-button', entrypoint: 'primeng/button' })
→ selector `p-button` is PrimeNG's; angulux uses `agl-button`
→ import specifier `primeng/button` does not resolve; use `@anguless/angulux/button`
A test drives the real binary over stdio and proves all 24 benchmark questions are answerable in at most two tool calls, with PrimeNG's answer absent from the reply. That is a claim about the data, not about assistants: whether a model actually consults the server instead of guessing is not measured — see the package README.
See packages/angulux-mcp/README.md for the other four tools,
the non-Claude client snippet, and why the package is unpublished.
PROVENANCE.md records the exact upstream commits, the archived MIT
artifacts and their checksums, and the command to verify them yourself. You are not asked to
take the MIT claim on trust — you are given the means to check it. That is the whole point
of the project.
Read CONTRIBUTING.md before opening a pull request. It is short, but the provenance clause is not optional: contributions must not carry code, documentation or generated output derived from any post-MIT PrimeNG release. One contaminated pull request would invalidate the entire argument above, so every pull request carries a Developer Certificate of Origin sign-off and an explicit provenance attestation.
Also relevant: TRIAGE.md (what happens after you file) · CODE_OF_CONDUCT.md · SECURITY.md · SUPPORT.md · GOVERNANCE.md
Then that is the single most valuable thing you can report here, and it is handled as P0: open a Provenance concern.
A project that asks you not to trust it has to mean it. The claim gets checked, not defended; if it holds, the affected code is removed rather than argued for; if a gate should have caught it, a gate gets written; and the outcome is published either way, including when the report turns out to be wrong. If you are a rights holder making a legal claim, use the private channel instead.