@decentnetwork/beagle-ui

The Beagle chat UI, minus the machine it runs on.

Two clients render these components: the desktop install (@decentnetwork/beagle, whose local server holds the identity as a file on disk) and the browser build (beagle-web, where the peer runs inside the tab). They were two copies of the same files, kept in sync by hand, until the drift was measured — 306 lines across three files, none of it two implementations of the same thing. This package is that measurement made permanent.

Using it

import { mount, setUiHost } from "@decentnetwork/beagle-ui";

setUiHost({
  fileUrl: (name) => "/api/file-download?name=" + encodeURIComponent(name),
  fileDownloadUrl: (name) => "/api/file-download?name=" + encodeURIComponent(name) + "&dl=1",
});
mount(document.getElementById("root"));

setUiHost must run before mount — the flags it sets are read during the first render.

React, ReactDOM, window.PeerWebRTC (calls) and window.ChatComponents (form messages) are expected as <script> globals. This bundle references them; it does not import them, so a page never ends up with two copies of React.

The host

Everything the two platforms cannot agree on. A host that supplies nothing gets the simple case: a local server that answers plain URLs.

| key | what it is for | |---|---| | fileUrl(name) | a stored file as a URL an <img> can use | | fileDownloadUrl(name) | the same, for <a download> | | prefetchThreadFiles(msgs) | resolve a thread's files before it renders | | DkWelcome | first-run flow | | DkLockedOut | another tab owns the identity | | DkBrowserNotice | standing caveat about running in a browser | | hostNeedsWelcome | run the first-run flow, and refuse a nameless friend request | | hostPrefersRtcFile | send files over a DataChannel when the peer can take one | | hostHasKeyBackup | offer to export the keypair |

In a browser, <img src> and <a download> issue native requests that bypass a patched fetch, so a plain /api/... URL reaches nobody when the peer lives in the tab — which is why fileUrl exists rather than being hardcoded.

Versioning

Consumers should pin an exact version. The release cadence is manual, and a UI change that reaches an already-published client through a caret range is a change nobody chose to ship.