Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* The agent programs PortOS drives, as the browser needs to name them.
*
* Read straight off `PROVIDER_HARNESSES` in `server/lib/providerHarnesses.js`
* through `harnessById`, so a harness added to the registry is named here
* without a second edit. Only the label is read: the registry's other columns
* are decided server-side and arrive on the wire already resolved — `matches`
* classifies a provider RECORD (the browser is handed a `harnessId`), `modes`
* is already reflected by the routes a binding actually owns, and `protocol`
* is a transport decision no picker makes.
*/
import {
PROVIDER_HARNESS_IDS,
harnessById,
harnessForProvider,
} from '../../../server/lib/providerHarnesses.js';
/**
* What to call a binding's harness.
*
* `null` is the DIRECT API case and has a name of its own — it is a real
* binding with a real route, not a missing value. An id this build does not
* know is shown verbatim rather than hidden: an unmapped harness stays a
* visible legacy route, which is exactly what the graph promises.
*/
export const harnessLabel = (harnessId) => {
if (harnessId === null || harnessId === undefined) return 'Direct API';
return harnessById(harnessId)?.label || harnessId;
};
/**
* Where a preset is edited (#7567): `/ai/presets/:presetId`. Preset ids are
* slugified from the display name, so a preset named "New" gets the id `new`
* and would be shadowed by the static `/ai/presets/new` create route — that
* one record keeps the legacy `/ai/edit/:providerId` alias, which App.jsx
* serves as a working route for exactly this reason.
*/
export const presetEditPath = (presetId) => (
presetId === 'new' ? '/ai/edit/new' : `/ai/presets/${encodeURIComponent(presetId)}`
);
/** The group a preset with no recognizable harness lands in. */
export const OTHER_HARNESS_GROUP = 'other';
/**
* The harness a preset record runs on, for GROUPING: a derived preset (#7565)
* names it outright; a legacy preset is classified from its command/type the
* same way the server does (`harnessForProvider`), so the two never disagree
* on which group a record belongs to. `null` when neither applies.
*/
export const providerHarnessId = (provider) =>
provider?.harnessId ?? harnessForProvider(provider)?.id ?? null;
/**
* Presets bucketed by harness for the preset-first `ProviderModelSelector`
* (#7566): one `{ harnessId, label, providers }` per harness in registry
* order (`direct`, the API providers, is the registry's last row), then any
* harness id this build's registry does not know (a derived preset written by
* a newer install), and a final "Other" bucket for records no harness claims. Empty buckets are omitted and
* each record's position within its bucket is the input order, so a caller's
* own ordering survives the grouping.
*
* @param {object[]} providers
* @returns {{harnessId: string, label: string, providers: object[]}[]}
*/
export const groupProvidersByHarness = (providers) => {
const buckets = new Map();
for (const provider of Array.isArray(providers) ? providers : []) {
const harnessId = providerHarnessId(provider) ?? OTHER_HARNESS_GROUP;
if (!buckets.has(harnessId)) buckets.set(harnessId, []);
buckets.get(harnessId).push(provider);
}
const order = [
...PROVIDER_HARNESS_IDS,
...[...buckets.keys()].filter((id) => !PROVIDER_HARNESS_IDS.includes(id) && id !== OTHER_HARNESS_GROUP),
OTHER_HARNESS_GROUP,
];
return order
.filter((harnessId) => buckets.has(harnessId))
.map((harnessId) => ({
harnessId,
label: harnessId === OTHER_HARNESS_GROUP ? 'Other' : harnessLabel(harnessId),
providers: buckets.get(harnessId),
}));
};