The all-in-one local AI studio for your desktop: chat, image and video generation and a coding agent in one free, open source app. Windows and Linux. No Docker, no terminal, no cloud required.
TypeScript agentai-chatai-studioall-in-onecoding-agentdesktop-appfluximage-generationlinuxllmlocal-ailocal-llmopen-sourceprivacyself-hostedstable-diffusionuncensoreduncensored-aivideo-generationwindows
// P3: the two Studio-only cloud routes. Same layer as every other cloud call
// (cloudFetch: CLOUD_BASE + bearer), never a bare fetch out of a component
// (Abschnitt 4 des Portplans).
import { cloudFetch, jsonOrError, CloudJobError } from './client'
import type { ModelRuntime } from '../../lib/render/runtime-format'
export interface StudioQuoteResult {
credits: number
seconds?: number
}
export interface StudioQuoteParams {
op: 'studio'
studio_options: Record<string, unknown>
// Whichever staged-upload path fields the picked model's schema needs
// (image_paths, source_path, mask_path, audio_path, audio2_path, video_path,
// last_image_path, duration, shot_type, ...); the exact set is per-model,
// studio-contract.ts is the vertrag, not this type.
[field: string]: unknown
}
const OLDER_SERVER_MESSAGE =
'This feature needs a newer LU Cloud server. Try again later.'
/** Thrown on a 409 from `/api/jobs/studio-quote` (`code: 'quote_changed'`):
* the price this call asked about is stale, and `credits` carries the
* server's fresh number, typed through instead of buried in `.message`. */
export class StudioQuoteChangedError extends CloudJobError {
readonly credits: number
readonly seconds?: number
constructor(message: string, credits: number, seconds?: number) {
super(message, 409, { code: 'quote_changed' })
this.name = 'StudioQuoteChangedError'
this.credits = credits
this.seconds = seconds
}
}
/**
* POST /api/jobs/studio-quote: the provider-confirmed price for one Studio
* run, asked before the run books. Never falls back to a client-side formula
* on failure: `studio-contract.ts` can PREVIEW a number for the controls, but
* only this call's number is what the server will actually charge, and
* booking against a guessed one risks a 409 after the customer already
* pressed start (Portplan Abschnitt 4/7, Risiko 1+2).
*
* Two failure shapes both mean "this server does not have Studio yet", and
* both get the same honest, English text instead of a technical one:
* - a plain 404 (the route answered, CORS present, but no handler, a
* server that shipped P0's CORS fix without this route existing yet), and
* - a raw network/CORS failure (the route has no `withCors`/`OPTIONS` at
* all, the browser never lets the response reach this code; the
* documented pre-P0 state of both routes today). cloudFetch already turns
* that into its own CloudJobError(status 0, "Could not reach the LU
* Cloud server...") for every OTHER cloud call; here it is reworded once
* more, because "try again in a moment" reads as a blip and this is a
* version gap. Any other rejection (401 not signed in, 408 timeout, a
* deliberate abort) is not a version gap and keeps its own status intact.
*/
export async function studioQuote(
model: string,
prompt: string,
params: StudioQuoteParams,
/** Vorab-Preis vor dem Hochladen (02.10.2026): nur die ZAHL der Bilder, nie
* eine Adresse. Ein Server, der das Feld nicht kennt, ignoriert es; der
* Aufrufer sendet es darum nur an einen Server, der Stufen fuehrt (also
* neu genug ist), siehe useStudioPrice. */
quoteImages?: number,
): Promise<StudioQuoteResult> {
let res: Response
try {
res = await cloudFetch('/api/jobs/studio-quote', {
method: 'POST',
headers: { 'content-type': 'application/json' },
body: JSON.stringify({ model, prompt, params, ...(quoteImages ? { quote_images: quoteImages } : {}) }),
})
} catch (err) {
if (err instanceof CloudJobError && err.status !== 0) throw err
throw new CloudJobError(OLDER_SERVER_MESSAGE, 0, { cause: err })
}
if (res.status === 404) throw new CloudJobError(OLDER_SERVER_MESSAGE, 404)
if (res.status === 409) {
const body = (await res.json().catch(() => ({}))) as {
error?: string
credits?: number
seconds?: number
}
throw new StudioQuoteChangedError(
typeof body.error === 'string'
? body.error
: 'The price has changed. Review the new quote before starting.',
typeof body.credits === 'number' ? body.credits : 0,
typeof body.seconds === 'number' ? body.seconds : undefined,
)
}
// 400 (invalid request/options) and 429 (quote rate limit) keep the
// server's own text verbatim: jsonOrError throws CloudJobError with
// `body.error` unchanged, nothing to reword or translate here.
return jsonOrError<StudioQuoteResult>(res)
}
/**
* GET /api/jobs/runtime: measured render times per model, for the "about
* 45 sec" hint next to the price. Unreachable (offline, an older server, or
* the same pre-P0 CORS gap as studio-quote) means silence, not a guess:
* `MIN_SAMPLES` in runtime-format.ts already says this app would rather say
* nothing than make one up, so any failure here just yields an empty map.
*/
export async function modelRuntimes(): Promise<Record<string, ModelRuntime>> {
try {
const res = await cloudFetch('/api/jobs/runtime')
const body = await jsonOrError<{ runtimes: Record<string, ModelRuntime> }>(res)
return body.runtimes ?? {}
} catch {
return {}
}
}