CanvasAnvil is an open source AI diagram maker and flowchart generator - an infinite-canvas diagram editor and self-hosted drawio alternative that writes editable .drawio files. Text to diagram, then fix nodes, layout and labels in place. Also AI presentations, interior design CAD/BOM, posters and infographics.
TypeScript ai-agentsai-diagramai-presentationbomcadcanvasdiagramdiagram-generatordrawioflowchartinfographicspptppt-generatorpresentation-generatorreactsvgedittext-to-diagramtypescriptworkflow
refactor(ai): collapse the multi-provider layer onto one OpenAI transport
codingFeng1018ef6f1f879eb
diff --git a/package.json b/package.jsonindex 4720d8f..e471442 100644--- a/package.json+++ b/package.json@@ -10,6 +10,7 @@"start": "tsx api/server.ts","build": "tsc -b && vite build","lint": "eslint .",+ "test": "tsx --test tests/*.test.ts","preview": "vite preview --host 0.0.0.0 --port 8001","check": "tsc -b --noEmit"},diff --git a/src/ai/browser-image.ts b/src/ai/browser-image.tsnew file mode 100644index 0000000..daab9f3--- /dev/null+++ b/src/ai/browser-image.ts@@ -0,0 +1,135 @@+/**+ * Browser-side normalisation for images handed to a model.+ *+ * Anything the user drops into a workspace can be a blob: URL, an oversized+ * screenshot, or an already-inlined data URL. Models reject or choke on large+ * payloads, so references are inlined, downscaled, and de-duplicated here+ * before they reach the server proxy.+ */++const MAX_DIMENSION = 1536;+const MAX_DATA_URL_LENGTH = 1_800_000;+const JPEG_QUALITY = 0.86;++export const MAX_REFERENCE_IMAGES = 3;++export function cleanUrl(url: string) {+ if (!url) return null;+ if (url.startsWith('http')) return url;+ // If it looks like a path but not absolute URL, return as is (might be base64 or relative)+ if (url.startsWith('data:image')) return url;+ return null;+}++async function objectUrlToDataUrl(objectUrl: string) {+ if (typeof window === "undefined") return null;+ if (!objectUrl || !objectUrl.startsWith("blob:")) return null;+ try {+ const resp = await fetch(objectUrl);+ const blob = await resp.blob();+ return await new Promise<string | null>((resolve) => {+ const reader = new FileReader();+ reader.onerror = () => resolve(null);+ reader.onloadend = () => {+ const result = reader.result;+ resolve(typeof result === "string" ? result : null);+ };+ reader.readAsDataURL(blob);+ });+ } catch {+ return null;+ }+}++function getImageMimeFromDataUrl(dataUrl: string) {+ const match = String(dataUrl || "").match(/^data:(image\/[a-zA-Z0-9.+-]+);base64,/i);+ return match?.[1]?.toLowerCase() || "image/png";+}++async function compressDataImageUrlForModel(dataUrl: string) {+ if (typeof window === "undefined" || typeof document === "undefined") return dataUrl;+ if (!String(dataUrl || "").startsWith("data:image")) return dataUrl;++ return await new Promise<string>((resolve) => {+ const img = new Image();+ img.onload = () => {+ try {+ const width = Number(img.naturalWidth || img.width || 0);+ const height = Number(img.naturalHeight || img.height || 0);+ if (!width || !height) {+ resolve(dataUrl);+ return;+ }++ const longestSide = Math.max(width, height);+ const needsResize = longestSide > MAX_DIMENSION;+ const needsReencode = dataUrl.length > MAX_DATA_URL_LENGTH;++ if (!needsResize && !needsReencode) {+ resolve(dataUrl);+ return;+ }++ const scale = needsResize ? MAX_DIMENSION / longestSide : 1;+ const targetWidth = Math.max(1, Math.round(width * scale));+ const targetHeight = Math.max(1, Math.round(height * scale));+ const canvas = document.createElement("canvas");+ canvas.width = targetWidth;+ canvas.height = targetHeight;+ const ctx = canvas.getContext("2d");+ if (!ctx) {+ resolve(dataUrl);+ return;+ }++ ctx.imageSmoothingEnabled = true;+ ctx.imageSmoothingQuality = "high";+ ctx.drawImage(img, 0, 0, targetWidth, targetHeight);++ const originalMime = getImageMimeFromDataUrl(dataUrl);+ const targetMime =+ originalMime === "image/jpeg" || originalMime === "image/jpg" || needsReencode+ ? "image/jpeg"+ : originalMime;+ const compressed =+ targetMime === "image/jpeg"+ ? canvas.toDataURL(targetMime, JPEG_QUALITY)+ : canvas.toDataURL(targetMime);++ resolve(compressed.length < dataUrl.length ? compressed : dataUrl);+ } catch {+ resolve(dataUrl);+ }+ };+ img.onerror = () => resolve(dataUrl);+ img.src = dataUrl;+ });+}++export async function normalizeImageUrlForModel(url: string) {+ const raw = String(url || "").trim();+ if (!raw) return null;++ if (raw.startsWith("blob:")) {+ const dataUrl = await objectUrlToDataUrl(raw);+ if (!dataUrl) return null;+ return await compressDataImageUrlForModel(dataUrl);+ }++ if (raw.startsWith("data:image")) {+ return await compressDataImageUrlForModel(raw);+ }++ return cleanUrl(raw);+}++export async function normalizeImageUrlsForModel(imageUrls: string[], maxImages: number) {+ const normalized: string[] = [];+ for (const raw of imageUrls) {+ if (normalized.length >= maxImages) break;+ const url = await normalizeImageUrlForModel(raw);+ if (!url || normalized.includes(url)) continue;+ normalized.push(url);+ }+ return normalized;+}diff --git a/src/ai/client.ts b/src/ai/client.tsnew file mode 100644index 0000000..9f885a8--- /dev/null+++ b/src/ai/client.ts@@ -0,0 +1,233 @@+import { t } from "@/lib/i18n";+import { getUiLanguage } from "@/lib/ui-language";+import { getAIConfig, getImageChannel, getTextChannel } from "@/ai/config";+import { applyUiLanguagePolicy } from "@/ai/language-policy";+import { createLimiter } from "@/ai/limiter";+import {+ MAX_REFERENCE_IMAGES,+ cleanUrl,+ normalizeImageUrlForModel,+ normalizeImageUrlsForModel,+} from "@/ai/browser-image";+import type { AIConfig, ChatContentPart, ChatMessage, MultimodalMessage } from "@/ai/types";++export type { AIConfig, ChatMessage } from "@/ai/types";+export { getAIConfig, saveAIConfig } from "@/ai/config";++/**+ * Browser-side facade over the model.+ *+ * Requests are proxied through `/api/ppt-ai` rather than sent straight to the+ * provider: it keeps the key out of cross-origin requests and lets the server+ * inline remote images the browser could not fetch.+ */++const MODEL_CONCURRENCY = 30;+const limitModelCall = createLimiter(MODEL_CONCURRENCY);++export interface ImageGenerationRequest {+ prompt: string;+ referenceImageUrl?: string;+ additionalReferenceImageUrls?: string[];+ maskImageUrl?: string;+}++export type GenerateChatMessageOptions = {+ signal?: AbortSignal;+ timeoutMs?: number;+};++type ProxyRequest =+ | { kind: "chat"; aiConfig: AIConfig; messages: MultimodalMessage[]; model?: string }+ | ({ kind: "image"; aiConfig: AIConfig; model?: string } & ImageGenerationRequest);++async function callProxy<T>(body: ProxyRequest, signal?: AbortSignal): Promise<T> {+ const response = await fetch("/api/ppt-ai", {+ method: "POST",+ headers: { "Content-Type": "application/json" },+ body: JSON.stringify(body),+ signal,+ });++ const text = await response.text().catch(() => "");+ let parsed: any = null;+ try {+ parsed = text ? JSON.parse(text) : null;+ } catch {+ parsed = null;+ }++ if (!response.ok) {+ const detail = String(parsed?.error || text || "").trim();+ throw new Error(detail || `Proxy request failed with status ${response.status}`);+ }+ return parsed as T;+}++function requireTextChannel() {+ const config = getAIConfig();+ const channel = getTextChannel(config);+ if (!channel.apiKey) throw new Error(t(getUiLanguage(), "error.missingApiKey"));+ return { config, channel };+}++function isAbort(error: unknown) {+ const name = (error as any)?.name;+ return name === "AbortError" || name === "APIUserAbortError";+}++/**+ * Run `task` under the concurrency limit, with an optional timeout folded into+ * the caller's own AbortSignal.+ */+async function withCancellation<T>(options: GenerateChatMessageOptions | undefined, task: (signal?: AbortSignal) => Promise<T>) {+ const timeoutMs = typeof options?.timeoutMs === "number" && options.timeoutMs > 0 ? options.timeoutMs : 0;+ const external = options?.signal;+ const controller = typeof AbortController !== "undefined" ? new AbortController() : null;+ const signal = controller?.signal || external;+ let timeoutId: ReturnType<typeof setTimeout> | null = null;++ if (controller && external) {+ if (external.aborted) controller.abort();+ else external.addEventListener("abort", () => controller.abort(), { once: true });+ }+ if (controller && timeoutMs > 0) {+ timeoutId = setTimeout(() => controller.abort(), timeoutMs);+ }++ try {+ return await limitModelCall(() => task(signal), signal);+ } finally {+ if (timeoutId) clearTimeout(timeoutId);+ }+}++export async function generateChatMessage(+ messages: ChatMessage[],+ model?: string,+ options?: GenerateChatMessageOptions,+) {+ return await withCancellation(options, async (signal) => {+ const { config, channel } = requireTextChannel();+ try {+ const data = await callProxy<{ content?: string }>(+ { kind: "chat", aiConfig: config, messages: applyUiLanguagePolicy(messages), model: model || channel.model },+ signal,+ );+ return String(data?.content || "");+ } catch (error) {+ if (isAbort(error)) throw error;+ console.error("Chat Error:", error);+ throw error;+ }+ });+}++/**+ * Same transport as {@link generateChatMessage}, but without the UI-language+ * policy message: callers here own their whole prompt (PPT outline and slide+ * agents write their own language rules).+ */+export async function generatePptProxyChatMessage(+ messages: ChatMessage[],+ model?: string,+ options?: GenerateChatMessageOptions,+) {+ const { config, channel } = requireTextChannel();+ const data = await callProxy<{ content?: string }>(+ { kind: "chat", aiConfig: config, messages, model: model || channel.model },+ options?.signal,+ );+ return String(data?.content || "");+}++export async function generateVisionChatMessage(+ systemPrompt: string,+ userPrompt: string,+ imageUrls: string[],+ model?: string,+ options?: GenerateChatMessageOptions,+) {+ return await withCancellation(options, async (signal) => {+ const { config, channel } = requireTextChannel();+ const normalized = await normalizeImageUrlsForModel(imageUrls, MAX_REFERENCE_IMAGES);+ const userContent: ChatContentPart[] = [+ { type: "text", text: userPrompt },+ ...normalized.map((url) => ({ type: "image_url" as const, image_url: { url } })),+ ];++ try {+ const data = await callProxy<{ content?: string }>(+ {+ kind: "chat",+ aiConfig: config,+ messages: [+ { role: "system", content: systemPrompt },+ { role: "user", content: userContent },+ ],+ model: model || channel.model,+ },+ signal,+ );+ return String(data?.content || "");+ } catch (error) {+ if (isAbort(error)) throw error;+ console.error("Vision Chat Error:", error);+ throw error;+ }+ });+}++/**+ * Non-streaming despite the name: the proxy returns a completed message, so the+ * callback fires exactly once. Kept as the shape the chat panels are written+ * against.+ */+export async function streamChatMessage(+ messages: ChatMessage[],+ onChunk: (content: string) => void,+ model?: string,+ signal?: AbortSignal,+) {+ const content = await generateChatMessage(messages, model, { signal });+ onChunk(content);+ return content;+}++export async function generateImage(request: ImageGenerationRequest, signal?: AbortSignal) {+ return await limitModelCall(async () => {+ const config = getAIConfig();+ const channel = getImageChannel(config);+ if (!channel.apiKey) throw new Error(t(getUiLanguage(), "error.missingApiKey"));++ const referenceImageUrl = request.referenceImageUrl+ ? await normalizeImageUrlForModel(request.referenceImageUrl)+ : null;+ const maskImageUrl = request.maskImageUrl+ ? await normalizeImageUrlForModel(request.maskImageUrl)+ : null;+ const additionalReferenceImageUrls = await normalizeImageUrlsForModel(+ Array.isArray(request.additionalReferenceImageUrls) ? request.additionalReferenceImageUrls : [],+ Math.max(0, MAX_REFERENCE_IMAGES - (referenceImageUrl ? 1 : 0)),+ );++ try {+ const result = await callProxy<{ url?: string }>(+ {+ kind: "image",+ aiConfig: config,+ prompt: request.prompt,+ referenceImageUrl: referenceImageUrl || undefined,+ additionalReferenceImageUrls,+ maskImageUrl: maskImageUrl || undefined,+ model: channel.model,+ },+ signal,+ );+ return cleanUrl(String(result?.url || ""));+ } catch (error) {+ console.error("Image Gen Error:", error);+ throw error;+ }+ }, signal);+}diff --git a/src/ai/config.ts b/src/ai/config.tsnew file mode 100644index 0000000..bc95030--- /dev/null+++ b/src/ai/config.ts@@ -0,0 +1,81 @@+import type { AIChannel, AIConfig } from "./types";++export const DEFAULT_BASE_URL = "https://api.openai.com/v1";++export const DEFAULT_AI_CONFIG: AIConfig = {+ textApiKey: "",+ textBaseUrl: DEFAULT_BASE_URL,+ textModel: "gpt-4o-mini",+ imageApiKey: "",+ imageBaseUrl: DEFAULT_BASE_URL,+ imageModel: "gpt-image-1",+ fileParserApiToken: "",+ systemPrompt: "",+};++const STORAGE_KEY = "unified_ai_workspace_config";++const str = (value: unknown, fallback = "") => {+ const text = String(value ?? "").trim();+ return text || fallback;+};++/**+ * Coerce anything that claims to be an AIConfig into a complete one.+ *+ * Also migrates configs written by the pre-OpenAI-only settings dialog, which+ * stored a single shared `apiKey`/`baseUrl` alongside per-channel overrides+ * and named the models `chatModel` / `imageModelLegacy`.+ */+export function normalizeAIConfig(raw: unknown): AIConfig {+ const input = (raw && typeof raw === "object" ? raw : {}) as Record<string, unknown>;+ const sharedApiKey = str(input.apiKey);+ const sharedBaseUrl = str(input.baseUrl);++ return {+ textApiKey: str(input.textApiKey, sharedApiKey),+ textBaseUrl: str(input.textBaseUrl, sharedBaseUrl || DEFAULT_BASE_URL),+ textModel: str(input.textModel, str(input.chatModel)),+ imageApiKey: str(input.imageApiKey, sharedApiKey),+ imageBaseUrl: str(input.imageBaseUrl, sharedBaseUrl || DEFAULT_BASE_URL),+ imageModel: str(input.imageModel, str(input.imageModelLegacy)),+ fileParserApiToken: str(input.fileParserApiToken),+ systemPrompt: String(input.systemPrompt ?? "").trim(),+ };+}++export function getTextChannel(config: AIConfig): AIChannel {+ return {+ apiKey: config.textApiKey,+ baseUrl: config.textBaseUrl || DEFAULT_BASE_URL,+ model: config.textModel,+ };+}++export function getImageChannel(config: AIConfig): AIChannel {+ return {+ apiKey: config.imageApiKey,+ baseUrl: config.imageBaseUrl || DEFAULT_BASE_URL,+ model: config.imageModel,+ };+}++export function getAIConfig(): AIConfig {+ if (typeof window === "undefined") return DEFAULT_AI_CONFIG;+ const stored = localStorage.getItem(STORAGE_KEY);+ if (!stored) return DEFAULT_AI_CONFIG;+ try {+ return normalizeAIConfig({ ...DEFAULT_AI_CONFIG, ...JSON.parse(stored) });+ } catch {+ try {+ localStorage.removeItem(STORAGE_KEY);+ } catch {+ // A storage that cannot be written to cannot be repaired; fall through.+ }+ return DEFAULT_AI_CONFIG;+ }+}++export function saveAIConfig(config: AIConfig) {+ localStorage.setItem(STORAGE_KEY, JSON.stringify(normalizeAIConfig(config)));+}diff --git a/src/ai/image.ts b/src/ai/image.tsnew file mode 100644index 0000000..4fbe8f0--- /dev/null+++ b/src/ai/image.ts@@ -0,0 +1,110 @@+import { dataUrlToFile, postForm, postJson, toDataUrl } from "./transport";+import type { ImageRequest } from "./types";++/**+ * Image models are reached through three OpenAI routes, tried in order:+ *+ * 1. `/images/edits` — only when a reference image was supplied+ * 2. `/images/generations`+ * 3. `/chat/completions` — many OpenAI-compatible gateways expose their+ * image models as chat models that answer with an+ * image part or a markdown image link+ *+ * The first route that returns an image wins. If every route fails, the error+ * from the first (most specific) attempt is what the caller sees, since that+ * is the route the configuration asked for.+ */+type Route = (req: ImageRequest) => Promise<string>;++function referenceImageUrls(req: ImageRequest): string[] {+ return [+ ...(req.referenceImageUrl ? [req.referenceImageUrl] : []),+ ...(req.additionalReferenceImageUrls || []).filter(Boolean),+ ];+}++/** OpenAI's images routes answer with either base64 or a short-lived URL. */+async function readImagesResponse(result: any): Promise<string> {+ const first = result?.data?.[0];+ if (first?.b64_json) return `data:image/png;base64,${first.b64_json}`;+ if (first?.url) return await toDataUrl(String(first.url));+ throw new Error("Image request succeeded but returned no image data.");+}++/** Chat responses may carry the image as a part, a markdown link, or a bare URL. */+function readChatImage(content: any): string | null {+ const fromText = (text: string): string | null => {+ const markdown = text.match(/!\[.*?\]\((.*?)\)/);+ if (markdown?.[1]) return markdown[1];+ const trimmed = text.trim();+ if (/^https?:\/\//i.test(trimmed) || trimmed.startsWith("data:image")) return trimmed;+ return null;+ };++ if (typeof content === "string") return fromText(content);+ if (!Array.isArray(content)) return null;++ const imagePart = content.find((part) => part?.type === "image_url" && part?.image_url?.url);+ if (imagePart) return String(imagePart.image_url.url);++ const textPart = content.find((part) => part?.type === "text" && typeof part?.text === "string");+ return textPart ? fromText(textPart.text) : null;+}++const editRoute: Route = async (req) => {+ const form = new FormData();+ form.append("model", req.channel.model);+ form.append("prompt", req.prompt);+ form.append("image", await dataUrlToFile(req.referenceImageUrl!, "reference.png"));+ if (req.maskImageUrl) {+ form.append("mask", await dataUrlToFile(req.maskImageUrl, "mask.png"));+ }+ return await readImagesResponse(await postForm(req.channel, "/images/edits", form, req.signal));+};++const generationRoute: Route = async (req) =>+ await readImagesResponse(+ await postJson(req.channel, "/images/generations", { model: req.channel.model, prompt: req.prompt }, req.signal),+ );++const chatRoute: Route = async (req) => {+ const parts: any[] = [{ type: "text", text: req.prompt }];+ if (req.maskImageUrl) {+ parts.push({ type: "image_url", image_url: { url: req.maskImageUrl } });+ }+ for (const url of referenceImageUrls(req)) {+ parts.push({ type: "image_url", image_url: { url } });+ }++ const result = await postJson(+ req.channel,+ "/chat/completions",+ {+ model: req.channel.model,+ messages: [{ role: "user", content: parts.length > 1 ? parts : req.prompt }],+ stream: false,+ },+ req.signal,+ );++ const url = readChatImage(result?.choices?.[0]?.message?.content);+ if (!url) throw new Error("Image request succeeded but returned no image URL.");+ return await toDataUrl(url);+};++export async function generateImage(req: ImageRequest): Promise<string> {+ const routes: Route[] = req.referenceImageUrl+ ? [editRoute, chatRoute]+ : [generationRoute, chatRoute];++ let firstError: unknown = null;+ for (const route of routes) {+ try {+ return await route(req);+ } catch (error) {+ if ((error as any)?.name === "AbortError") throw error;+ if (firstError === null) firstError = error;+ }+ }+ throw firstError instanceof Error ? firstError : new Error("Image generation failed.");+}diff --git a/src/ai/index.ts b/src/ai/index.tsnew file mode 100644index 0000000..c2d8a9c--- /dev/null+++ b/src/ai/index.ts@@ -0,0 +1,4 @@+export * from "./types";+export * from "./config";+export { generateText, contentToText } from "./text";+export { generateImage } from "./image";diff --git a/src/ai/language-policy.ts b/src/ai/language-policy.tsnew file mode 100644index 0000000..bac7289--- /dev/null+++ b/src/ai/language-policy.ts@@ -0,0 +1,31 @@+import { getUiLanguage } from "@/lib/ui-language";+import type { ChatMessage } from "@/ai/types";++/**+ * Agents in this app emit code, XML, and JSON as often as prose, so the output+ * language is pinned by an injected system message rather than left to the+ * model's guess. The prefix lets a re-send replace the previous policy instead+ * of stacking a second one.+ */+const POLICY_PREFIX = "UI_LANG_POLICY:";++const POLICIES: Record<"zh" | "en", string> = {+ en:+ `${POLICY_PREFIX} uiLang=en\n` ++ "All assistant outputs must be in English.\n" ++ "- Do not output Chinese characters.\n" ++ "- If an agent must output code/JSON/XML only, keep the required format and keep any fixed identifiers/schema keys; write any human-readable strings in English unless the schema mandates otherwise.",+ zh:+ `${POLICY_PREFIX} uiLang=zh\n` ++ "All assistant outputs must be in Simplified Chinese.\n" ++ "- Do not output English unless required for code, identifiers, proper nouns, or file paths.\n" ++ "- If an agent must output code/JSON/XML only, keep the required format and keep any fixed identifiers/schema keys; write any human-readable strings in Chinese unless the schema mandates otherwise.",+};++export function applyUiLanguagePolicy(messages: ChatMessage[]): ChatMessage[] {+ const withoutPrevious = (messages || []).filter(+ (message) => !(message.role === "system" && String(message.content || "").startsWith(POLICY_PREFIX)),+ );+ const policy: ChatMessage = { role: "system", content: POLICIES[getUiLanguage() === "en" ? "en" : "zh"] };+ return [policy, ...withoutPrevious];+}diff --git a/src/ai/limiter.ts b/src/ai/limiter.tsnew file mode 100644index 0000000..0c3ae0f--- /dev/null+++ b/src/ai/limiter.ts@@ -0,0 +1,89 @@+/**+ * Bounded-concurrency queue for model calls.+ *+ * The workspaces fan out dozens of per-slide requests at once; without a cap+ * the provider rejects the burst. Queued items honour their AbortSignal both+ * while waiting and while running.+ */+type QueueItem<T> = {+ run: () => Promise<T>;+ resolve: (value: T) => void;+ reject: (reason?: any) => void;+ signal?: AbortSignal;+ onAbort?: () => void;+};++export function createLimiter(max: number) {+ let active = 0;+ const queue: QueueItem<any>[] = [];++ const pump = () => {+ while (active < max && queue.length > 0) {+ const item = queue.shift()!;+ if (item.signal?.aborted) {+ item.reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));+ continue;+ }++ active += 1;+ if (item.signal && item.onAbort) {+ item.signal.removeEventListener("abort", item.onAbort);+ item.onAbort = undefined;+ }+ let finished = false;+ let abortHandler: (() => void) | null = null;++ const finish = () => {+ if (finished) return;+ finished = true;+ if (item.signal && abortHandler) {+ item.signal.removeEventListener("abort", abortHandler as any);+ }+ active -= 1;+ pump();+ };++ if (item.signal) {+ abortHandler = () => {+ const err = Object.assign(new Error("Aborted"), { name: "AbortError" });+ item.reject(err);+ finish();+ };+ item.signal.addEventListener("abort", abortHandler, { once: true });+ }++ item.run().then(+ (value) => {+ item.resolve(value);+ finish();+ },+ (reason) => {+ item.reject(reason);+ finish();+ }+ );+ }+ };++ return function limit<T>(run: () => Promise<T>, signal?: AbortSignal) {+ if (signal?.aborted) {+ return Promise.reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));+ }++ return new Promise<T>((resolve, reject) => {+ const item: QueueItem<T> = { run, resolve, reject, signal };+ if (signal) {+ const onAbort = () => {+ const idx = queue.indexOf(item as any);+ if (idx >= 0) queue.splice(idx, 1);+ reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));+ };+ item.onAbort = onAbort;+ signal.addEventListener("abort", onAbort, { once: true });+ }++ queue.push(item);+ pump();+ });+ };+}diff --git a/src/ai/server-model.ts b/src/ai/server-model.tsnew file mode 100644index 0000000..1473138--- /dev/null+++ b/src/ai/server-model.ts@@ -0,0 +1,29 @@+import { createOpenAI } from "@ai-sdk/openai";+import type { LanguageModel } from "ai";+import type { AIChannel } from "./types";++/**+ * Build the AI SDK model used by the streaming `/api/chat` route.+ *+ * Credentials always come from the app's Settings dialog and arrive with the+ * request; `AI_MODEL` / `AI_BASE_URL` only supply defaults for deployments that+ * pin a model server-side.+ */+export function getChatModel(channel: Partial<AIChannel>): { model: LanguageModel; modelId: string } {+ const modelId = String(channel.model || process.env.AI_MODEL || "").trim();+ if (!modelId) {+ throw new Error("No chat model configured. Set one in the app Settings dialog.");+ }++ const apiKey = String(channel.apiKey || "").trim();+ if (!apiKey) {+ throw new Error("API Key is missing. Please set it in the app Settings dialog.");+ }++ const baseURL = String(+ channel.baseUrl || process.env.AI_BASE_URL || process.env.OPENAI_BASE_URL || "",+ ).trim();++ const openai = createOpenAI(baseURL ? { apiKey, baseURL } : { apiKey });+ return { model: openai.chat(modelId), modelId };+}diff --git a/src/ai/text.ts b/src/ai/text.tsnew file mode 100644index 0000000..9d4271c--- /dev/null+++ b/src/ai/text.ts@@ -0,0 +1,30 @@+import { postJson } from "./transport";+import type { ChatContentPart, TextRequest } from "./types";++/** Flatten OpenAI message content (string or part array) to plain text. */+export function contentToText(content: unknown): string {+ if (typeof content === "string") return content;+ if (!Array.isArray(content)) return "";+ return content+ .map((part) => {+ if (typeof part === "string") return part;+ const typed = part as ChatContentPart;+ return typed?.type === "text" && typeof typed.text === "string" ? typed.text : "";+ })+ .filter(Boolean)+ .join("\n");+}++export async function generateText(req: TextRequest): Promise<string> {+ const result = await postJson(+ req.channel,+ "/chat/completions",+ {+ model: req.channel.model,+ messages: req.messages,+ stream: false,+ },+ req.signal,+ );+ return contentToText(result?.choices?.[0]?.message?.content);+}diff --git a/src/ai/transport.ts b/src/ai/transport.tsnew file mode 100644index 0000000..05a373f--- /dev/null+++ b/src/ai/transport.ts@@ -0,0 +1,84 @@+import type { AIChannel } from "./types";++/** Join a configured base URL with an API path, tolerating stray slashes. */+export function joinUrl(baseUrl: string, path: string): string {+ const base = String(baseUrl || "").trim().replace(/\/+$/, "");+ const suffix = String(path || "").trim();+ if (!suffix) return base;+ if (/^https?:\/\//i.test(suffix)) return suffix;+ if (!base) return suffix;+ return `${base}/${suffix.replace(/^\/+/, "")}`;+}++function parseBody(text: string): any {+ try {+ return text ? JSON.parse(text) : null;+ } catch {+ return null;+ }+}++/** Pull the most specific message an OpenAI-shaped error response offers. */+function toRequestError(status: number, text: string): Error {+ const parsed = parseBody(text);+ const message =+ parsed?.error?.message ||+ (typeof parsed?.error === "string" ? parsed.error : "") ||+ parsed?.message ||+ text ||+ `Request failed with status ${status}`;+ return new Error(String(message));+}++async function send(url: string, init: RequestInit): Promise<any> {+ const response = await fetch(url, init);+ const text = await response.text();+ if (!response.ok) throw toRequestError(response.status, text);+ return parseBody(text);+}++export function postJson(channel: AIChannel, path: string, body: unknown, signal?: AbortSignal): Promise<any> {+ return send(joinUrl(channel.baseUrl, path), {+ method: "POST",+ headers: {+ Authorization: `Bearer ${channel.apiKey}`,+ "Content-Type": "application/json",+ },+ body: JSON.stringify(body),+ signal,+ });+}++/** Multipart POST, for `/images/edits`; fetch sets the boundary itself. */+export function postForm(channel: AIChannel, path: string, form: FormData, signal?: AbortSignal): Promise<any> {+ return send(joinUrl(channel.baseUrl, path), {+ method: "POST",+ headers: { Authorization: `Bearer ${channel.apiKey}` },+ body: form,+ signal,+ });+}++/**+ * Inline a remote image so callers never hand out a URL that expires or needs+ * separate credentials. Data URLs pass through untouched.+ */+export async function toDataUrl(url: string): Promise<string> {+ if (url.startsWith("data:image")) return url;+ const response = await fetch(url);+ if (!response.ok) {+ throw new Error(`Failed to fetch remote image with status ${response.status}`);+ }+ const contentType = response.headers.get("content-type") || "image/png";+ const buffer = Buffer.from(await response.arrayBuffer());+ return `data:${contentType};base64,${buffer.toString("base64")}`;+}++export async function dataUrlToFile(dataUrl: string, filename: string): Promise<File> {+ const normalized = await toDataUrl(dataUrl);+ const match = normalized.match(/^data:([^;]+);base64,(.*)$/);+ if (!match) throw new Error("Unsupported image data URL.");+ return new File([Buffer.from(match[2] || "", "base64")], filename, {+ type: match[1] || "image/png",+ });+}diff --git a/src/ai/types.ts b/src/ai/types.tsnew file mode 100644index 0000000..9aec2a2--- /dev/null+++ b/src/ai/types.ts@@ -0,0 +1,63 @@+/**+ * Every model call in CanvasAnvil speaks the OpenAI HTTP protocol:+ * `POST {baseUrl}/chat/completions`, `POST {baseUrl}/images/generations`+ * and `POST {baseUrl}/images/edits`. Any vendor that exposes an+ * OpenAI-compatible endpoint is configured by pointing `baseUrl` at it —+ * there is no per-vendor branching anywhere in this module.+ */++/** One configured endpoint: where to send requests, with what credentials. */+export interface AIChannel {+ apiKey: string;+ baseUrl: string;+ model: string;+}++/**+ * Persisted settings. Text and image are separate channels because the two+ * are commonly served by different endpoints; each is still plain OpenAI.+ */+export interface AIConfig {+ textApiKey: string;+ textBaseUrl: string;+ textModel: string;+ imageApiKey: string;+ imageBaseUrl: string;+ imageModel: string;+ fileParserApiToken: string;+ systemPrompt: string;+}++export type ChatRole = "system" | "user" | "assistant";++export type ChatContentPart =+ | { type: "text"; text: string }+ | { type: "image_url"; image_url: { url: string } };++/** The common case: a text-only turn. */+export interface ChatMessage {+ role: ChatRole;+ content: string;+}++/** A turn that may also carry images, using OpenAI's content-part array. */+export interface MultimodalMessage {+ role: ChatRole;+ content: string | ChatContentPart[];+}++export interface TextRequest {+ channel: AIChannel;+ messages: MultimodalMessage[];+ signal?: AbortSignal;+}++export interface ImageRequest {+ channel: AIChannel;+ prompt: string;+ /** Primary reference image. Its presence selects the `/images/edits` route. */+ referenceImageUrl?: string;+ additionalReferenceImageUrls?: string[];+ maskImageUrl?: string;+ signal?: AbortSignal;+}diff --git a/src/components/SettingsDialog.tsx b/src/components/SettingsDialog.tsxindex 891bb46..90ee9b5 100644--- a/src/components/SettingsDialog.tsx+++ b/src/components/SettingsDialog.tsx@@ -1,54 +1,32 @@-import React, { useEffect, useMemo, useState } from "react";+import React, { useEffect, useState } from "react";import { createPortal } from "react-dom";import { ExternalLink, Eye, EyeOff, Loader2, Save, Settings, X } from "lucide-react";import { toast } from "sonner";-import { getAIConfig, saveAIConfig, type AIConfig } from "@/lib/ai-client";+import { getAIConfig, normalizeAIConfig, saveAIConfig } from "@/ai/config";+import type { AIConfig } from "@/ai/types";import { Button } from "@/components/ui/button";import { getUiLanguage, setUiLanguage, type UiLanguage } from "@/lib/ui-language";import { t } from "@/lib/i18n";-import {- IMAGE_PROVIDER_OPTIONS,- TEXT_PROVIDER_OPTIONS,- getDefaultBaseUrl,- normalizeAIConfig,- type AIProviderId,-} from "@/lib/ai/provider-registry";-import {- Select,- SelectContent,- SelectItem,- SelectTrigger,- SelectValue,-} from "@/workspaces/flow/next/components/ui/select";-type ChannelKind = "text" | "image";-type TestState = "idle" | "testing";+/**+ * Model settings.+ *+ * Every model is reached over the OpenAI HTTP protocol, so a channel is fully+ * described by base URL + key + model name. Pointing the base URL at a+ * vendor's OpenAI-compatible endpoint is all that switching providers takes;+ * there is no provider list to keep in sync.+ */-const TEXT_PROVIDER_LINKS: Partial<Record<AIProviderId, string>> = {- openai: "https://platform.openai.com/api-keys",- ollama: "https://ollama.com/download",- deepseek: "https://platform.deepseek.com/api_keys",- kimi: "https://platform.moonshot.cn/console/api-keys",- aliyun: "https://bailian.console.aliyun.com/?tab=model#/api-key",- tencent: "https://console.cloud.tencent.com/hunyuan/api-key",- bytedance: "https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey",- zhipu: "https://open.bigmodel.cn/usercenter/apikeys",- baidu: "https://console.bce.baidu.com/qianfan/ais/console/applicationConsole/application",- minimax: "https://platform.minimaxi.com/user-center/basic-information/interface-key",- xai: "https://console.x.ai/",- google: "https://aistudio.google.com/app/apikey",- anthropic: "https://console.anthropic.com/settings/keys",-};+const MINERU_LINK = "https://mineru.net/";+const OPENAI_KEYS_LINK = "https://platform.openai.com/api-keys";-const IMAGE_PROVIDER_LINKS: Partial<Record<AIProviderId, string>> = {- openai: "https://platform.openai.com/api-keys",- aliyun: "https://bailian.console.aliyun.com/?tab=model#/api-key",- tencent: "https://console.cloud.tencent.com/hunyuan/api-key",- bytedance: "https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey",- google: "https://console.cloud.google.com/vertex-ai/publishers/google/model-garden/imagen",-};+type ChannelKey = "text" | "image";+type TestState = "idle" | "testing";-const MINERU_LINK = "https://mineru.net/";+const CHANNEL_FIELDS = {+ text: { apiKey: "textApiKey", baseUrl: "textBaseUrl", model: "textModel" },+ image: { apiKey: "imageApiKey", baseUrl: "imageBaseUrl", model: "imageModel" },+} as const satisfies Record<ChannelKey, Record<"apiKey" | "baseUrl" | "model", keyof AIConfig>>;function SectionCard({title,@@ -75,64 +53,30 @@ function SectionCard({);}-function LinkText({- href,- uiLang,-}: {- href?: string;- uiLang: UiLanguage;-}) {- if (!href) return null;- return (- <a- href={href}- target="_blank"- rel="noreferrer"- className="inline-flex items-center gap-1 text-xs text-primary hover:underline"- >- {uiLang === "zh" ? "获取 Key" : "Get Key"}- <ExternalLink className="h-3.5 w-3.5" />- </a>- );-}+const inputClass =+ "flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50";-function ProviderField({+function TextField({label,value,onChange,- options,- docsUrl,- uiLang,+ placeholder,}: {label: string;value: string;- onChange: (value: AIProviderId) => void;- options: Array<{ id: AIProviderId; label: string }>;- docsUrl?: string;- uiLang: UiLanguage;+ onChange: (value: string) => void;+ placeholder: string;}) {return (<div className="space-y-2">- <div className="flex items-center justify-between gap-3">- <label className="text-sm font-medium text-foreground">{label}</label>- <LinkText href={docsUrl} uiLang={uiLang} />- </div>- <Select value={value} onValueChange={(next) => onChange(next as AIProviderId)}>- <SelectTrigger className="h-10 w-full rounded-xl border-border/70 bg-background shadow-sm transition-colors hover:border-border hover:bg-muted/20">- <SelectValue placeholder="Select provider" />- </SelectTrigger>- <SelectContent className="max-h-[260px] overflow-y-auto rounded-xl border-border/70 bg-background/95 p-1 shadow-xl backdrop-blur">- {options.map((option) => (- <SelectItem- key={option.id}- value={option.id}- className="rounded-lg py-2 text-sm data-[highlighted]:bg-muted data-[highlighted]:text-foreground"- >- {option.label}- </SelectItem>- ))}- </SelectContent>- </Select>+ <label className="text-sm font-medium text-foreground">{label}</label>+ <input+ type="text"+ value={value}+ onChange={(e) => onChange(e.target.value)}+ className={inputClass}+ placeholder={placeholder}+ /></div>);}@@ -142,8 +86,6 @@ function SecretInput({value,onChange,placeholder,- visible,- onToggle,docsUrl,uiLang,}: {@@ -151,28 +93,37 @@ function SecretInput({value: string;onChange: (value: string) => void;placeholder: string;- visible: boolean;- onToggle: () => void;docsUrl?: string;uiLang: UiLanguage;}) {+ const [visible, setVisible] = useState(false);return (<div className="space-y-2"><div className="flex items-center justify-between gap-3"><label className="text-sm font-medium text-foreground">{label}</label>- <LinkText href={docsUrl} uiLang={uiLang} />+ {docsUrl ? (+ <a+ href={docsUrl}+ target="_blank"+ rel="noreferrer"+ className="inline-flex items-center gap-1 text-xs text-primary hover:underline"+ >+ {uiLang === "zh" ? "获取 Key" : "Get Key"}+ <ExternalLink className="h-3.5 w-3.5" />+ </a>+ ) : null}</div><div className="relative"><inputtype={visible ? "text" : "password"}value={value}onChange={(e) => onChange(e.target.value)}- className="flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 pr-10 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50"+ className={`${inputClass} pr-10`}placeholder={placeholder}/><buttontype="button"- onClick={onToggle}+ onClick={() => setVisible((prev) => !prev)}className="absolute inset-y-0 right-0 flex w-10 items-center justify-center text-muted-foreground transition-colors hover:text-foreground"aria-label={visible ? "Hide secret" : "Show secret"}>@@ -183,43 +134,84 @@ function SecretInput({);}-function TestButton({- label,- state,- onClick,+function ChannelCard({+ channel,+ title,+ description,+ modelPlaceholder,+ testLabel,+ config,+ onChange,+ onTest,+ testState,+ uiLang,}: {- label: string;- state: TestState;- onClick: () => void;+ channel: ChannelKey;+ title: string;+ description: string;+ modelPlaceholder: string;+ testLabel: string;+ config: AIConfig;+ onChange: (patch: Partial<AIConfig>) => void;+ onTest: () => void;+ testState: TestState;+ uiLang: UiLanguage;}) {- const loading = state === "testing";+ const fields = CHANNEL_FIELDS[channel];return (- <Button type="button" variant="outline" className="h-9 rounded-lg" onClick={onClick} disabled={loading}>- {loading ? <Loader2 className="h-4 w-4 animate-spin" /> : null}- {label}- </Button>+ <SectionCard+ title={title}+ description={description}+ actions={+ <Button+ type="button"+ variant="outline"+ className="h-9 rounded-lg"+ onClick={onTest}+ disabled={testState === "testing"}+ >+ {testState === "testing" ? <Loader2 className="h-4 w-4 animate-spin" /> : null}+ {testLabel}+ </Button>+ }+ >+ <div className="grid grid-cols-1 gap-4 md:grid-cols-2">+ <TextField+ label="Model"+ value={config[fields.model]}+ onChange={(value) => onChange({ [fields.model]: value } as Partial<AIConfig>)}+ placeholder={modelPlaceholder}+ />+ <TextField+ label="Base URL"+ value={config[fields.baseUrl]}+ onChange={(value) => onChange({ [fields.baseUrl]: value } as Partial<AIConfig>)}+ placeholder="https://api.openai.com/v1"+ />+ <SecretInput+ label="API Key"+ value={config[fields.apiKey]}+ onChange={(value) => onChange({ [fields.apiKey]: value } as Partial<AIConfig>)}+ placeholder="sk-..."+ docsUrl={OPENAI_KEYS_LINK}+ uiLang={uiLang}+ />+ </div>+ </SectionCard>);}export function SettingsDialog() {const [isOpen, setIsOpen] = useState(false);- const [config, setConfig] = useState<AIConfig>(normalizeAIConfig(getAIConfig()));+ const [config, setConfig] = useState<AIConfig>(() => getAIConfig());const [uiLang, setUiLangState] = useState<UiLanguage>(() => getUiLanguage());- const [showTextKey, setShowTextKey] = useState(false);- const [showImageKey, setShowImageKey] = useState(false);- const [showMineruKey, setShowMineruKey] = useState(false);- const [textTestState, setTextTestState] = useState<TestState>("idle");- const [imageTestState, setImageTestState] = useState<TestState>("idle");+ const [testState, setTestState] = useState<Record<ChannelKey, TestState>>({ text: "idle", image: "idle" });useEffect(() => {if (!isOpen) return;- setConfig(normalizeAIConfig(getAIConfig()));+ setConfig(getAIConfig());setUiLangState(getUiLanguage());- setShowTextKey(false);- setShowImageKey(false);- setShowMineruKey(false);- setTextTestState("idle");- setImageTestState("idle");+ setTestState({ text: "idle", image: "idle" });}, [isOpen]);useEffect(() => {@@ -228,101 +220,47 @@ export function SettingsDialog() {}, [uiLang, isOpen]);const isZh = uiLang === "zh";+ const patch = (values: Partial<AIConfig>) => setConfig((prev) => ({ ...prev, ...values }));- const textProviderLink = useMemo(- () => TEXT_PROVIDER_LINKS[config.textProvider as AIProviderId],- [config.textProvider],- );- const imageProviderLink = useMemo(- () => IMAGE_PROVIDER_LINKS[config.imageProvider as AIProviderId],- [config.imageProvider],- );-- const handleProviderChange = (kind: ChannelKind, provider: AIProviderId) => {+ /** Send one real request through the same proxy the workspaces use. */+ const testChannel = async (channel: ChannelKey) => {const normalized = normalizeAIConfig(config);- const defaultBaseUrl = getDefaultBaseUrl(provider, kind);-- if (kind === "text") {- setConfig({- ...normalized,- textProvider: provider,- textBaseUrl: defaultBaseUrl || normalized.textBaseUrl,- });- return;- }-- setConfig({- ...normalized,- imageProvider: provider,- imageBaseUrl: defaultBaseUrl || normalized.imageBaseUrl,- });- };-- const handleSave = () => {- saveAIConfig(config);- setUiLanguage(uiLang);- setIsOpen(false);- toast.success(isZh ? "配置已保存" : "Settings saved");- };-- const handleTestText = async () => {- const normalized = normalizeAIConfig(config);- if (!normalized.textApiKey || !normalized.textBaseUrl || !normalized.textModel) {- toast.error(isZh ? "请先填写文本模型的 Key、Base URL 和 Model" : "Please fill text API key, base URL, and model first");+ const fields = CHANNEL_FIELDS[channel];+ if (!normalized[fields.apiKey] || !normalized[fields.baseUrl] || !normalized[fields.model]) {+ toast.error(+ isZh ? "请先填写 Model、Base URL 和 API Key" : "Please fill model, base URL, and API key first",+ );return;}+ setTestState((prev) => ({ ...prev, [channel]: "testing" }));try {- setTextTestState("testing");const response = await fetch("/api/ppt-ai", {method: "POST",headers: { "Content-Type": "application/json" },- body: JSON.stringify({- kind: "chat",- aiConfig: normalized,- messages: [{ role: "user", content: "Reply with OK only." }],- }),+ body: JSON.stringify(+ channel === "text"+ ? { kind: "chat", aiConfig: normalized, messages: [{ role: "user", content: "Reply with OK only." }] }+ : { kind: "image", aiConfig: normalized, prompt: "A simple blue square icon on white background." },+ ),});const payload = await response.json().catch(() => ({}));- if (!response.ok) {- throw new Error(String(payload?.error || "Text model test failed"));+ if (!response.ok || (channel === "image" && !payload?.url)) {+ throw new Error(String(payload?.error || "Model test failed"));}- toast.success(isZh ? "文本模型可用" : "Text model is working");+ toast.success(isZh ? "模型可用" : "Model is working");} catch (error) {- toast.error(error instanceof Error ? error.message : isZh ? "文本模型测试失败" : "Text model test failed");+ toast.error(error instanceof Error ? error.message : isZh ? "模型测试失败" : "Model test failed");} finally {- setTextTestState("idle");+ setTestState((prev) => ({ ...prev, [channel]: "idle" }));}};- const handleTestImage = async () => {- const normalized = normalizeAIConfig(config);- if (!normalized.imageApiKey || !normalized.imageBaseUrl || !normalized.imageModel) {- toast.error(isZh ? "请先填写生图模型的 Key、Base URL 和 Model" : "Please fill image API key, base URL, and model first");- return;- }-- try {- setImageTestState("testing");- const response = await fetch("/api/ppt-ai", {- method: "POST",- headers: { "Content-Type": "application/json" },- body: JSON.stringify({- kind: "image",- aiConfig: normalized,- prompt: "A simple blue square icon on white background.",- }),- });- const payload = await response.json().catch(() => ({}));- if (!response.ok || !payload?.url) {- throw new Error(String(payload?.error || "Image model test failed"));- }- toast.success(isZh ? "生图模型可用" : "Image model is working");- } catch (error) {- toast.error(error instanceof Error ? error.message : isZh ? "生图模型测试失败" : "Image model test failed");- } finally {- setImageTestState("idle");- }+ const handleSave = () => {+ saveAIConfig(config);+ setUiLanguage(uiLang);+ setIsOpen(false);+ toast.success(isZh ? "配置已保存" : "Settings saved");};return (@@ -346,9 +284,7 @@ export function SettingsDialog() {<h2 className="text-lg font-semibold tracking-tight text-foreground">{t(uiLang, "settings.title")}</h2>- <p className="mt-1 text-xs text-muted-foreground">- {t(uiLang, "settings.subtitle")}- </p>+ <p className="mt-1 text-xs text-muted-foreground">{t(uiLang, "settings.subtitle")}</p></div><Buttonvariant="ghost"@@ -383,119 +319,39 @@ export function SettingsDialog() {</div></SectionCard>- <SectionCard+ <ChannelCard+ channel="text"title={isZh ? "文本模型" : "Text Model"}description={isZh- ? "用于聊天、文本生成和图片理解。上传图片时会直接交给模型处理,由模型真实返回是否支持。"- : "Used for chat, text generation, and image understanding. Image uploads are sent directly to the model, which decides whether they are supported."- }- actions={- <TestButton- label={isZh ? "测试文本模型" : "Test Text Model"}- state={textTestState}- onClick={handleTestText}- />+ ? "用于聊天、文本生成和图片理解。任何 OpenAI 兼容端点都可以,把 Base URL 指过去即可。"+ : "Used for chat, text generation, and image understanding. Any OpenAI-compatible endpoint works — just point the base URL at it."}- >- <div className="grid grid-cols-1 gap-4 md:grid-cols-2">- <ProviderField- label="Provider"- value={config.textProvider}- onChange={(value) => handleProviderChange("text", value)}- options={TEXT_PROVIDER_OPTIONS}- docsUrl={textProviderLink}- uiLang={uiLang}- />- <div className="space-y-2">- <label className="text-sm font-medium text-foreground">Model</label>- <input- type="text"- value={config.textModel}- onChange={(e) => setConfig({ ...config, textModel: e.target.value })}- className="flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50"- placeholder="gpt-4o-mini"- />- </div>- <SecretInput- label="API Key"- value={config.textApiKey}- onChange={(value) => setConfig({ ...config, textApiKey: value })}- placeholder="sk-..."- visible={showTextKey}- onToggle={() => setShowTextKey((prev) => !prev)}- docsUrl={textProviderLink}- uiLang={uiLang}- />- <div className="space-y-2">- <label className="text-sm font-medium text-foreground">Base URL</label>- <input- type="text"- value={config.textBaseUrl}- onChange={(e) => setConfig({ ...config, textBaseUrl: e.target.value })}- className="flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50"- placeholder="https://api.example.com/v1"- />- </div>- </div>- </SectionCard>+ modelPlaceholder="gpt-4o-mini"+ testLabel={isZh ? "测试文本模型" : "Test Text Model"}+ config={config}+ onChange={patch}+ onTest={() => testChannel("text")}+ testState={testState.text}+ uiLang={uiLang}+ />- <SectionCard+ <ChannelCard+ channel="image"title={isZh ? "生图模型" : "Image Model"}description={isZh? "仅用于图片生成,与文本模型完全独立。": "Used only for image generation and fully separate from the text model."}- actions={- <TestButton- label={isZh ? "测试生图模型" : "Test Image Model"}- state={imageTestState}- onClick={handleTestImage}- />- }- >- <div className="grid grid-cols-1 gap-4 md:grid-cols-2">- <ProviderField- label="Provider"- value={config.imageProvider}- onChange={(value) => handleProviderChange("image", value)}- options={IMAGE_PROVIDER_OPTIONS}- docsUrl={imageProviderLink}- uiLang={uiLang}- />- <div className="space-y-2">- <label className="text-sm font-medium text-foreground">Model</label>- <input- type="text"- value={config.imageModel}- onChange={(e) => setConfig({ ...config, imageModel: e.target.value })}- className="flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50"- placeholder="gpt-image-1"- />- </div>- <SecretInput- label="API Key"- value={config.imageApiKey}- onChange={(value) => setConfig({ ...config, imageApiKey: value })}- placeholder="sk-..."- visible={showImageKey}- onToggle={() => setShowImageKey((prev) => !prev)}- docsUrl={imageProviderLink}- uiLang={uiLang}- />- <div className="space-y-2">- <label className="text-sm font-medium text-foreground">Base URL</label>- <input- type="text"- value={config.imageBaseUrl}- onChange={(e) => setConfig({ ...config, imageBaseUrl: e.target.value })}- className="flex h-10 w-full rounded-lg border border-input bg-background px-3 py-2 text-sm shadow-sm transition-all focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/50"- placeholder="https://api.example.com/v1"- />- </div>- </div>- </SectionCard>+ modelPlaceholder="gpt-image-1"+ testLabel={isZh ? "测试生图模型" : "Test Image Model"}+ config={config}+ onChange={patch}+ onTest={() => testChannel("image")}+ testState={testState.image}+ uiLang={uiLang}+ /><SectionCardtitle={isZh ? "其他" : "Other"}@@ -503,11 +359,9 @@ export function SettingsDialog() {><SecretInputlabel="MinerU Token"- value={config.fileParserApiToken || ""}- onChange={(value) => setConfig({ ...config, fileParserApiToken: value })}+ value={config.fileParserApiToken}+ onChange={(value) => patch({ fileParserApiToken: value })}placeholder={isZh ? "留空则使用本地解析" : "Leave empty to use local extraction"}- visible={showMineruKey}- onToggle={() => setShowMineruKey((prev) => !prev)}docsUrl={MINERU_LINK}uiLang={uiLang}/>@@ -515,11 +369,7 @@ export function SettingsDialog() {</div><div className="flex justify-end gap-3 border-t border-border/50 bg-muted/5 p-4 px-6">- <Button- variant="outline"- onClick={() => setIsOpen(false)}- className="rounded-lg hover:bg-muted"- >+ <Button variant="outline" onClick={() => setIsOpen(false)} className="rounded-lg hover:bg-muted">{t(uiLang, "common.cancel")}</Button><Button onClick={handleSave} className="gap-2 rounded-lg shadow-sm">diff --git a/src/lib/ai-client.ts b/src/lib/ai-client.tsdeleted file mode 100644index 6363b69..0000000--- a/src/lib/ai-client.ts+++ /dev/null@@ -1,607 +0,0 @@-import { getUiLanguage } from "@/lib/ui-language";-import { t } from "@/lib/i18n";-import {- type AIConfig,- getImageChannelConfig,- getTextChannelConfig,- normalizeAIConfig,-} from "@/lib/ai/provider-registry";-export type { AIConfig } from "@/lib/ai/provider-registry";--// Default Configuration-const DEFAULT_CONFIG: AIConfig = {- textProvider: "openai",- textApiKey: "",- textBaseUrl: "https://api.openai.com/v1",- textModel: "gpt-4o-mini",- textCustomMapping: "",- imageProvider: "openai",- imageApiKey: "",- imageBaseUrl: "https://api.openai.com/v1",- imageModel: "gpt-image-1",- imageCustomMapping: "",- fileParserApiToken: "",- systemPrompt: "",- apiKey: "",- baseUrl: "https://api.openai.com/v1",- chatModel: "gpt-4o-mini",- imageModelLegacy: "gpt-image-1",-};--const STORAGE_KEY = "unified_ai_workspace_config";-const MODEL_CONCURRENCY = 30;-const IMAGE_MODEL_MAX_REFERENCE_IMAGES = 3;-const IMAGE_MODEL_MAX_DIMENSION = 1536;-const IMAGE_MODEL_MAX_DATA_URL_LENGTH = 1_800_000;-const IMAGE_MODEL_JPEG_QUALITY = 0.86;--type QueueItem<T> = {- run: () => Promise<T>;- resolve: (value: T) => void;- reject: (reason?: any) => void;- signal?: AbortSignal;- onAbort?: () => void;-};--function createLimiter(max: number) {- let active = 0;- const queue: QueueItem<any>[] = [];-- const pump = () => {- while (active < max && queue.length > 0) {- const item = queue.shift()!;- if (item.signal?.aborted) {- item.reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));- continue;- }-- active += 1;- if (item.signal && item.onAbort) {- item.signal.removeEventListener("abort", item.onAbort);- item.onAbort = undefined;- }- let finished = false;- let abortHandler: (() => void) | null = null;-- const finish = () => {- if (finished) return;- finished = true;- if (item.signal && abortHandler) {- item.signal.removeEventListener("abort", abortHandler as any);- }- active -= 1;- pump();- };-- if (item.signal) {- abortHandler = () => {- const err = Object.assign(new Error("Aborted"), { name: "AbortError" });- item.reject(err);- finish();- };- item.signal.addEventListener("abort", abortHandler, { once: true });- }-- item.run().then(- (value) => {- item.resolve(value);- finish();- },- (reason) => {- item.reject(reason);- finish();- }- );- }- };-- return function limit<T>(run: () => Promise<T>, signal?: AbortSignal) {- if (signal?.aborted) {- return Promise.reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));- }-- return new Promise<T>((resolve, reject) => {- const item: QueueItem<T> = { run, resolve, reject, signal };- if (signal) {- const onAbort = () => {- const idx = queue.indexOf(item as any);- if (idx >= 0) queue.splice(idx, 1);- reject(Object.assign(new Error("Aborted"), { name: "AbortError" }));- };- item.onAbort = onAbort;- signal.addEventListener("abort", onAbort, { once: true });- }-- queue.push(item);- pump();- });- };-}--const limitModelCall = createLimiter(MODEL_CONCURRENCY);--// Helper to get config-export function getAIConfig(): AIConfig {- if (typeof window === 'undefined') return DEFAULT_CONFIG;- const stored = localStorage.getItem(STORAGE_KEY);- if (stored) {- try {- return normalizeAIConfig({ ...DEFAULT_CONFIG, ...JSON.parse(stored) });- } catch {- try {- localStorage.removeItem(STORAGE_KEY);- } catch {- }- return DEFAULT_CONFIG;- }- }- return DEFAULT_CONFIG;-}--// Helper to save config-export function saveAIConfig(config: AIConfig) {- localStorage.setItem(STORAGE_KEY, JSON.stringify(normalizeAIConfig(config)));-}--export interface ChatMessage {- role: 'user' | 'assistant' | 'system';- content: string;-}--const UI_LANG_POLICY_PREFIX = "UI_LANG_POLICY:";--function getUiLanguagePolicySystemMessage(): ChatMessage {- const uiLang = getUiLanguage();- if (uiLang === "en") {- return {- role: "system",- content:- `${UI_LANG_POLICY_PREFIX} uiLang=en\n` +- "All assistant outputs must be in English.\n" +- "- Do not output Chinese characters.\n" +- "- If an agent must output code/JSON/XML only, keep the required format and keep any fixed identifiers/schema keys; write any human-readable strings in English unless the schema mandates otherwise.",- };- }- return {- role: "system",- content:- `${UI_LANG_POLICY_PREFIX} uiLang=zh\n` +- "All assistant outputs must be in Simplified Chinese.\n" +- "- Do not output English unless required for code, identifiers, proper nouns, or file paths.\n" +- "- If an agent must output code/JSON/XML only, keep the required format and keep any fixed identifiers/schema keys; write any human-readable strings in Chinese unless the schema mandates otherwise.",- };-}--function applyUiLanguagePolicy(messages: ChatMessage[]): ChatMessage[] {- const filtered = (messages || []).filter(- (m) => !(m.role === "system" && String(m.content || "").startsWith(UI_LANG_POLICY_PREFIX))- );- return [getUiLanguagePolicySystemMessage(), ...filtered];-}--// Stream Chat Message-export async function streamChatMessage(- messages: ChatMessage[],- onChunk: (content: string) => void,- model?: string,- signal?: AbortSignal-) {- const content = await generateChatMessage(messages, model, { signal });- onChunk(content);- return content;-}--// Simple non-stream wrapper-export type GenerateChatMessageOptions = {- signal?: AbortSignal;- timeoutMs?: number;-};--async function normalizeVisionImageUrls(imageUrls: string[]) {- return await normalizeImageUrlsForModel(imageUrls, IMAGE_MODEL_MAX_REFERENCE_IMAGES);-}--export async function generateChatMessage(messages: ChatMessage[], model?: string, options?: GenerateChatMessageOptions) {- const timeoutMs = typeof options?.timeoutMs === "number" && options.timeoutMs > 0 ? options.timeoutMs : 0;- const externalSignal = options?.signal;-- const controller = typeof AbortController !== "undefined" ? new AbortController() : null;- const mergedSignal = controller?.signal || externalSignal;- let timeoutId: any = null;-- if (controller && externalSignal) {- if (externalSignal.aborted) controller.abort();- else externalSignal.addEventListener("abort", () => controller.abort(), { once: true });- }-- if (controller && timeoutMs > 0) {- timeoutId = setTimeout(() => controller.abort(), timeoutMs);- }-- try {- return await limitModelCall(async () => {- const config = getAIConfig();- const textChannel = getTextChannelConfig(config);- if (!textChannel.apiKey) {- throw new Error(t(getUiLanguage(), "error.missingApiKey"));- }-- try {- const data = await callPptProxy<{ content?: string }>(- {- kind: "chat",- aiConfig: config,- messages: applyUiLanguagePolicy(messages),- model: model || textChannel.model,- },- mergedSignal || undefined- );- return String(data?.content || "");- } catch (error) {- if ((error as any)?.name === "AbortError" || (error as any)?.name === "APIUserAbortError") {- throw error;- }- console.error("Chat Error:", error);- throw error;- }- }, mergedSignal);- } finally {- if (timeoutId) clearTimeout(timeoutId);- }-}--export async function generatePptProxyChatMessage(- messages: ChatMessage[],- model?: string,- options?: GenerateChatMessageOptions-) {- const config = getAIConfig();- const textChannel = getTextChannelConfig(config);- if (!textChannel.apiKey) {- throw new Error(t(getUiLanguage(), "error.missingApiKey"));- }-- const data = await callPptProxy<{ content?: string }>(- {- kind: "chat",- aiConfig: config,- messages,- model: model || textChannel.model,- },- options?.signal- );-- return String(data?.content || "");-}--export async function generateVisionChatMessage(- systemPrompt: string,- userPrompt: string,- imageUrls: string[],- model?: string,- options?: GenerateChatMessageOptions-) {- const timeoutMs = typeof options?.timeoutMs === "number" && options.timeoutMs > 0 ? options.timeoutMs : 0;- const externalSignal = options?.signal;-- const controller = typeof AbortController !== "undefined" ? new AbortController() : null;- const mergedSignal = controller?.signal || externalSignal;- let timeoutId: any = null;-- if (controller && externalSignal) {- if (externalSignal.aborted) controller.abort();- else externalSignal.addEventListener("abort", () => controller.abort(), { once: true });- }-- if (controller && timeoutMs > 0) {- timeoutId = setTimeout(() => controller.abort(), timeoutMs);- }-- try {- return await limitModelCall(async () => {- const config = getAIConfig();- const textChannel = getTextChannelConfig(config);-- if (!textChannel.apiKey) {- throw new Error(t(getUiLanguage(), "error.missingApiKey"));- }- const normalizedImageUrls = await normalizeVisionImageUrls(imageUrls);- const userContent: any[] = [{ type: "text", text: userPrompt }];- for (const url of normalizedImageUrls) {- userContent.push({- type: "image_url",- image_url: { url },- });- }-- try {- const data = await callPptProxy<{ content?: string }>(- {- kind: "chat",- aiConfig: config,- messages: [- { role: "system", content: systemPrompt },- { role: "user", content: userContent as any },- ] as any,- model: model || textChannel.model,- },- mergedSignal || undefined- );- return String(data?.content || "");- } catch (error) {- if ((error as any)?.name === "AbortError" || (error as any)?.name === "APIUserAbortError") {- throw error;- }- console.error("Vision Chat Error:", error);- throw error;- }- }, mergedSignal);- } finally {- if (timeoutId) clearTimeout(timeoutId);- }-}--// Legacy non-stream function (kept for compatibility if needed)-export async function sendChatMessage(messages: ChatMessage[]) {- return streamChatMessage(messages, () => {});-}--// Image Generation Client-export interface ImageGenerationRequest {- prompt: string;- referenceImageUrl?: string; // Main reference (kept for backward compatibility)- additionalReferenceImageUrls?: string[]; // Additional references- maskImageUrl?: string;-}--type PptProxyChatRequest = {- kind: "chat";- aiConfig: AIConfig;- messages: ChatMessage[];- model?: string;-};--type PptProxyImageRequest = {- kind: "image";- aiConfig: AIConfig;- prompt: string;- referenceImageUrl?: string;- additionalReferenceImageUrls?: string[];- maskImageUrl?: string;- model?: string;-};--async function callPptProxy<T>(body: PptProxyChatRequest | PptProxyImageRequest, signal?: AbortSignal): Promise<T> {- const response = await fetch("/api/ppt-ai", {- method: "POST",- headers: {- "Content-Type": "application/json",- },- body: JSON.stringify(body),- signal,- });-- const text = await response.text().catch(() => "");- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }-- if (!response.ok) {- const detail = String(parsed?.error || text || "").trim();- throw new Error(detail || `Proxy request failed with status ${response.status}`);- }-- return parsed as T;-}--// Helper to validate and clean URL-function cleanUrl(url: string) {- if (!url) return null;- if (url.startsWith('http')) return url;- // If it looks like a path but not absolute URL, return as is (might be base64 or relative)- if (url.startsWith('data:image')) return url;- return null;-}--async function objectUrlToDataUrl(objectUrl: string) {- if (typeof window === "undefined") return null;- if (!objectUrl || !objectUrl.startsWith("blob:")) return null;- try {- const resp = await fetch(objectUrl);- const blob = await resp.blob();- return await new Promise<string | null>((resolve) => {- const reader = new FileReader();- reader.onerror = () => resolve(null);- reader.onloadend = () => {- const result = reader.result;- resolve(typeof result === "string" ? result : null);- };- reader.readAsDataURL(blob);- });- } catch {- return null;- }-}--function getImageMimeFromDataUrl(dataUrl: string) {- const match = String(dataUrl || "").match(/^data:(image\/[a-zA-Z0-9.+-]+);base64,/i);- return match?.[1]?.toLowerCase() || "image/png";-}--async function compressDataImageUrlForModel(dataUrl: string) {- if (typeof window === "undefined" || typeof document === "undefined") return dataUrl;- if (!String(dataUrl || "").startsWith("data:image")) return dataUrl;-- return await new Promise<string>((resolve) => {- const img = new Image();- img.onload = () => {- try {- const width = Number(img.naturalWidth || img.width || 0);- const height = Number(img.naturalHeight || img.height || 0);- if (!width || !height) {- resolve(dataUrl);- return;- }-- const longestSide = Math.max(width, height);- const needsResize = longestSide > IMAGE_MODEL_MAX_DIMENSION;- const needsReencode = dataUrl.length > IMAGE_MODEL_MAX_DATA_URL_LENGTH;-- if (!needsResize && !needsReencode) {- resolve(dataUrl);- return;- }-- const scale = needsResize ? IMAGE_MODEL_MAX_DIMENSION / longestSide : 1;- const targetWidth = Math.max(1, Math.round(width * scale));- const targetHeight = Math.max(1, Math.round(height * scale));- const canvas = document.createElement("canvas");- canvas.width = targetWidth;- canvas.height = targetHeight;- const ctx = canvas.getContext("2d");- if (!ctx) {- resolve(dataUrl);- return;- }-- ctx.imageSmoothingEnabled = true;- ctx.imageSmoothingQuality = "high";- ctx.drawImage(img, 0, 0, targetWidth, targetHeight);-- const originalMime = getImageMimeFromDataUrl(dataUrl);- const targetMime =- originalMime === "image/jpeg" || originalMime === "image/jpg" || needsReencode- ? "image/jpeg"- : originalMime;- const compressed =- targetMime === "image/jpeg"- ? canvas.toDataURL(targetMime, IMAGE_MODEL_JPEG_QUALITY)- : canvas.toDataURL(targetMime);-- resolve(compressed.length < dataUrl.length ? compressed : dataUrl);- } catch {- resolve(dataUrl);- }- };- img.onerror = () => resolve(dataUrl);- img.src = dataUrl;- });-}--async function normalizeImageUrlForModel(url: string) {- const raw = String(url || "").trim();- if (!raw) return null;-- if (raw.startsWith("blob:")) {- const dataUrl = await objectUrlToDataUrl(raw);- if (!dataUrl) return null;- return await compressDataImageUrlForModel(dataUrl);- }-- if (raw.startsWith("data:image")) {- return await compressDataImageUrlForModel(raw);- }-- return cleanUrl(raw);-}--async function normalizeImageUrlsForModel(imageUrls: string[], maxImages: number) {- const normalized: string[] = [];- for (const raw of imageUrls) {- if (normalized.length >= maxImages) break;- const url = await normalizeImageUrlForModel(raw);- if (!url || normalized.includes(url)) continue;- normalized.push(url);- }- return normalized;-}--function extractImageUrlFromContent(messageContent: any) {- if (Array.isArray(messageContent)) {- const imagePart = messageContent.find((part: any) => part.type === 'image_url');- if (imagePart) {- return cleanUrl(imagePart.image_url.url);- }- const textPart = messageContent.find((part: any) => part.type === 'text');- if (textPart) {- const text = textPart.text;- const markdownMatch = text.match(/!\[.*?\]\((.*?)\)/);- if (markdownMatch && markdownMatch[1]) {- return cleanUrl(markdownMatch[1]);- }- if (text.trim().startsWith('http') || text.trim().startsWith('data:image')) {- return cleanUrl(text.trim());- }- }- return null;- }-- if (typeof messageContent === 'string') {- const markdownMatch = messageContent.match(/!\[.*?\]\((.*?)\)/);- if (markdownMatch && markdownMatch[1]) {- return cleanUrl(markdownMatch[1]);- }- if (messageContent.trim().startsWith('http')) {- return cleanUrl(messageContent.trim());- }- return null;- }-- return null;-}--function parseAIResponse(result: any) {- if (result.error) {- throw new Error(result.error.message || "API Error");- }-- if (result.choices && result.choices.length > 0) {- const messageContent = result.choices[0].message.content;- return extractImageUrlFromContent(messageContent);- }- return null;-}--export async function generateImage(request: ImageGenerationRequest, signal?: AbortSignal) {- return await limitModelCall(async () => {- const config = getAIConfig();- const imageChannel = getImageChannelConfig(config);- if (!imageChannel.apiKey) {- throw new Error(t(getUiLanguage(), "error.missingApiKey"));- }-- const normalizedReferenceImageUrl = request.referenceImageUrl- ? await normalizeImageUrlForModel(request.referenceImageUrl)- : null;- const normalizedMaskImageUrl = request.maskImageUrl- ? await normalizeImageUrlForModel(request.maskImageUrl)- : null;- const additionalUrls = await normalizeImageUrlsForModel(- Array.isArray(request.additionalReferenceImageUrls) ? request.additionalReferenceImageUrls : [],- Math.max(0, IMAGE_MODEL_MAX_REFERENCE_IMAGES - (normalizedReferenceImageUrl ? 1 : 0))- );-- try {- const result = await callPptProxy<{ url?: string }>(- {- kind: "image",- aiConfig: config,- prompt: request.prompt,- referenceImageUrl: normalizedReferenceImageUrl || undefined,- additionalReferenceImageUrls: additionalUrls,- maskImageUrl: normalizedMaskImageUrl || undefined,- model: imageChannel.model,- },- signal- );- const url = cleanUrl(String(result?.url || ""));- return url;- } catch (error) {- console.error("Image Gen Error:", error);- throw error;- }- }, signal);-}diff --git a/src/lib/ai/gateway.ts b/src/lib/ai/gateway.tsdeleted file mode 100644index 4849ff9..0000000--- a/src/lib/ai/gateway.ts+++ /dev/null@@ -1,893 +0,0 @@-import {- type AIChannelConfig,- type CustomProviderMappingSpec,- parseCustomProviderMapping,- resolveImageCapabilities,- resolveImageRoute,- resolveTextRoute,-} from "./provider-registry";--export type GatewayChatMessage = {- role: "system" | "user" | "assistant";- content: any;-};--export interface TextGatewayRequest {- channel: AIChannelConfig;- messages: GatewayChatMessage[];-}--export interface ImageGatewayRequest {- channel: AIChannelConfig;- prompt: string;- referenceImageUrl?: string;- additionalReferenceImageUrls?: string[];- maskImageUrl?: string;-}--type ImageAdapter = {- supports(request: ImageGatewayRequest): boolean;- generate(request: ImageGatewayRequest): Promise<string>;-};--function shouldUseEditFlow(req: ImageGatewayRequest, capabilities?: { supportsEdits: boolean }) {- return Boolean(req.referenceImageUrl && capabilities?.supportsEdits);-}--function getProvider(req: ImageGatewayRequest) {- return String(req.channel.provider || "").toLowerCase();-}--function joinUrl(baseUrl: string, endpoint: string): string {- const base = String(baseUrl || "").trim().replace(/\/+$/, "");- const path = String(endpoint || "").trim();- if (!path) return base;- if (/^https?:\/\//i.test(path)) return path;- if (!base) return path;- return `${base}/${path.replace(/^\/+/, "")}`;-}--function extractByPath(input: any, path: string) {- const normalized = String(path || "").trim();- if (!normalized) return input;- const tokens = normalized- .replace(/\[(\d+)\]/g, ".$1")- .split(".")- .map((token) => token.trim())- .filter(Boolean);- let current = input;- for (const token of tokens) {- if (current == null) return undefined;- current = current[token];- }- return current;-}--function applyTemplate(value: any, context: Record<string, any>): any {- if (typeof value === "string") {- if (value.startsWith("$")) {- return context[value.slice(1)];- }- return value.replace(/\$([a-zA-Z0-9_]+)/g, (_, key) => {- const resolved = context[key];- return resolved == null ? "" : typeof resolved === "string" ? resolved : JSON.stringify(resolved);- });- }- if (Array.isArray(value)) {- return value.map((item) => applyTemplate(item, context));- }- if (value && typeof value === "object") {- const out: Record<string, any> = {};- for (const [key, item] of Object.entries(value)) {- out[key] = applyTemplate(item, context);- }- return out;- }- return value;-}--async function executeCustomSpec(args: {- spec: CustomProviderMappingSpec;- channel: AIChannelConfig;- context: Record<string, any>;-}) {- const url = joinUrl(args.channel.baseUrl, args.spec.endpoint);- const method = args.spec.method || "POST";- const extraHeaders = applyTemplate(args.spec.headers || {}, args.context);- const bodyValue = applyTemplate(args.spec.body || {}, args.context);- const response = await fetch(url, {- method,- headers: {- Authorization: `Bearer ${args.channel.apiKey}`,- "Content-Type": "application/json",- ...extraHeaders,- },- ...(method === "GET" ? {} : { body: JSON.stringify(bodyValue) }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = text;- }-- if (!response.ok) {- throw new Error(- typeof parsed === "string"- ? parsed || `Request failed with status ${response.status}`- : parsed?.error || parsed?.message || `Request failed with status ${response.status}`,- );- }-- return extractByPath(parsed, args.spec.responsePath);-}--function extractTextFromOpenAIContent(content: any): string {- if (typeof content === "string") return content;- if (Array.isArray(content)) {- return content- .map((part) => {- if (typeof part === "string") return part;- if (part?.type === "text" && typeof part?.text === "string") return part.text;- return "";- })- .filter(Boolean)- .join("\n");- }- return "";-}--function splitSystemMessages(messages: GatewayChatMessage[]) {- const system = messages- .filter((message) => message.role === "system")- .map((message) => extractTextFromOpenAIContent(message.content))- .filter(Boolean)- .join("\n\n");- const nonSystem = messages.filter((message) => message.role !== "system");- return { system, nonSystem };-}--function toGoogleParts(content: any): any[] {- if (Array.isArray(content)) {- return content- .map((part) => {- if (typeof part === "string") return { text: part };- if (part?.type === "text" && typeof part.text === "string") return { text: part.text };- if (part?.type === "image_url" && typeof part?.image_url?.url === "string") {- const match = String(part.image_url.url).match(/^data:([^;]+);base64,(.*)$/);- if (match) {- return {- inlineData: {- mimeType: match[1],- data: match[2],- },- };- }- }- return null;- })- .filter(Boolean);- }- return [{ text: extractTextFromOpenAIContent(content) }];-}--function toAnthropicContent(content: any): any {- if (Array.isArray(content)) {- return content- .map((part) => {- if (typeof part === "string") return { type: "text", text: part };- if (part?.type === "text" && typeof part.text === "string") return { type: "text", text: part.text };- if (part?.type === "image_url" && typeof part?.image_url?.url === "string") {- const match = String(part.image_url.url).match(/^data:([^;]+);base64,(.*)$/);- if (match) {- return {- type: "image",- source: {- type: "base64",- media_type: match[1],- data: match[2],- },- };- }- }- return null;- })- .filter(Boolean);- }- return extractTextFromOpenAIContent(content);-}--function extractTextFromGoogleResponse(parsed: any): string {- return (- parsed?.candidates?.[0]?.content?.parts- ?.map((part: any) => (typeof part?.text === "string" ? part.text : ""))- .filter(Boolean)- .join("\n") || ""- );-}--function extractTextFromAnthropicResponse(parsed: any): string {- return (- parsed?.content- ?.map((part: any) => (part?.type === "text" && typeof part?.text === "string" ? part.text : ""))- .filter(Boolean)- .join("\n") || ""- );-}--function extractImageUrlFromOpenAIContent(content: any): string | null {- if (Array.isArray(content)) {- const imagePart = content.find((part) => part?.type === "image_url" && part?.image_url?.url);- if (imagePart?.image_url?.url) return String(imagePart.image_url.url);- const textPart = content.find((part) => part?.type === "text" && typeof part?.text === "string");- if (textPart?.text) {- const markdownMatch = textPart.text.match(/!\[.*?\]\((.*?)\)/);- if (markdownMatch?.[1]) return markdownMatch[1];- if (/^https?:\/\//i.test(textPart.text.trim()) || textPart.text.trim().startsWith("data:image")) {- return textPart.text.trim();- }- }- return null;- }-- if (typeof content === "string") {- const markdownMatch = content.match(/!\[.*?\]\((.*?)\)/);- if (markdownMatch?.[1]) return markdownMatch[1];- if (/^https?:\/\//i.test(content.trim()) || content.trim().startsWith("data:image")) {- return content.trim();- }- }-- return null;-}--function extractImageUrlFromAliyunContent(content: any): string | null {- if (Array.isArray(content)) {- const imagePart = content.find((part) => part?.image);- if (imagePart?.image) return String(imagePart.image);- const textPart = content.find((part) => part?.text && typeof part.text === "string");- if (textPart?.text) {- const markdownMatch = textPart.text.match(/!\[.*?\]\((.*?)\)/);- if (markdownMatch?.[1]) return markdownMatch[1];- if (/^https?:\/\//i.test(textPart.text.trim()) || textPart.text.trim().startsWith("data:image")) {- return textPart.text.trim();- }- }- }- return null;-}--async function convertRemoteImageToDataUrl(url: string): Promise<string> {- if (url.startsWith("data:image")) return url;- const response = await fetch(url);- if (!response.ok) {- throw new Error(`Failed to fetch remote image with status ${response.status}`);- }- const contentType = response.headers.get("content-type") || "image/png";- const buffer = Buffer.from(await response.arrayBuffer());- return `data:${contentType};base64,${buffer.toString("base64")}`;-}--function extractFirstImageValue(input: any): string | null {- if (!input) return null;- if (typeof input === "string") {- if (/^https?:\/\//i.test(input) || input.startsWith("data:image")) return input;- return null;- }- if (Array.isArray(input)) {- for (const item of input) {- const found = extractFirstImageValue(item);- if (found) return found;- }- return null;- }- if (typeof input === "object") {- for (const key of [- "url",- "image_url",- "imageUrl",- "b64_json",- "base64",- "bytesBase64Encoded",- "image",- "result_image",- "resultImage",- ]) {- const value = input[key];- if (typeof value === "string") {- if (key === "b64_json" || key === "base64" || key === "bytesBase64Encoded") return `data:image/png;base64,${value}`;- if (/^https?:\/\//i.test(value) || value.startsWith("data:image")) return value;- }- }- for (const value of Object.values(input)) {- const found = extractFirstImageValue(value);- if (found) return found;- }- }- return null;-}--async function outputImage(value: string | null, providerName: string): Promise<string> {- if (!value) throw new Error(`${providerName} image request succeeded but returned no image data.`);- return value.startsWith("data:image") ? value : await convertRemoteImageToDataUrl(value);-}--async function dataUrlToFile(dataUrl: string, filename: string): Promise<File> {- const normalized = dataUrl.startsWith("data:")- ? dataUrl- : await convertRemoteImageToDataUrl(dataUrl);- const match = normalized.match(/^data:([^;]+);base64,(.*)$/);- if (!match) {- throw new Error("Unsupported image data URL.");- }- const mime = match[1] || "image/png";- const bytes = Buffer.from(match[2] || "", "base64");- return new File([bytes], filename, { type: mime });-}--async function requestOpenAIImages(req: ImageGatewayRequest): Promise<string> {- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/generations"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- prompt: req.prompt,- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- const first = parsed?.data?.[0];- if (first?.b64_json) {- return `data:image/png;base64,${first.b64_json}`;- }- if (first?.url) {- return await convertRemoteImageToDataUrl(String(first.url));- }- throw new Error("Image request succeeded but returned no image data.");-}--async function requestOpenAIImageEdit(req: ImageGatewayRequest): Promise<string> {- if (!req.referenceImageUrl) {- return await requestOpenAIImages(req);- }-- const form = new FormData();- form.append("model", req.channel.model);- form.append("prompt", req.prompt);- form.append("image", await dataUrlToFile(req.referenceImageUrl, "slide.png"));- if (req.maskImageUrl) {- form.append("mask", await dataUrlToFile(req.maskImageUrl, "mask.png"));- }-- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/edits"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- },- body: form,- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- const first = parsed?.data?.[0];- if (first?.b64_json) {- return `data:image/png;base64,${first.b64_json}`;- }- if (first?.url) {- return await convertRemoteImageToDataUrl(String(first.url));- }- throw new Error("Image edit request succeeded but returned no image data.");-}--async function requestOpenAIChatImage(req: ImageGatewayRequest): Promise<string> {- const content: any[] = [{ type: "text", text: req.prompt }];- if (req.referenceImageUrl) {- content.push({ type: "image_url", image_url: { url: req.referenceImageUrl } });- }- if (req.maskImageUrl) {- content.push({ type: "image_url", image_url: { url: req.maskImageUrl } });- }- for (const url of req.additionalReferenceImageUrls || []) {- if (!url) continue;- content.push({ type: "image_url", image_url: { url } });- }-- const response = await fetch(joinUrl(req.channel.baseUrl, "/chat/completions"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- messages: [{ role: "user", content: content.length > 1 ? content : req.prompt }],- stream: false,- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- const url = extractImageUrlFromOpenAIContent(parsed?.choices?.[0]?.message?.content);- if (!url) {- throw new Error("Image request succeeded but returned no image URL.");- }- return await convertRemoteImageToDataUrl(url);-}--async function requestAliyunImage(req: ImageGatewayRequest): Promise<string> {- const content: any[] = [];- for (const url of [- ...(req.referenceImageUrl ? [req.referenceImageUrl] : []),- ...(req.additionalReferenceImageUrls || []),- ]) {- content.push({ image: url });- }- content.push({ text: req.prompt });-- const response = await fetch(joinUrl(req.channel.baseUrl, "/services/aigc/multimodal-generation/generation"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- input: {- messages: [- {- role: "user",- content,- },- ],- },- parameters: {- n: 1,- watermark: false,- },- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.message || parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- const url = extractImageUrlFromAliyunContent(parsed?.output?.choices?.[0]?.message?.content);- if (!url) {- throw new Error("Aliyun image request succeeded but returned no image URL.");- }- return await convertRemoteImageToDataUrl(url);-}--async function requestAliyunImageEdit(req: ImageGatewayRequest): Promise<string> {- if (!req.referenceImageUrl) return await requestAliyunImage(req);- return await requestAliyunImage(req);-}--async function requestOpenAICompatibleImage(req: ImageGatewayRequest, providerName: string): Promise<string> {- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/generations"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- prompt: req.prompt,- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return await outputImage(extractFirstImageValue(parsed?.data?.[0] || parsed), providerName);-}--async function requestBytedanceImage(req: ImageGatewayRequest): Promise<string> {- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/generations"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- prompt: req.prompt,- response_format: "url",- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return await outputImage(extractFirstImageValue(parsed?.data?.[0] || parsed), "Bytedance Seedream");-}--async function requestBytedanceImageEdit(req: ImageGatewayRequest): Promise<string> {- if (!req.referenceImageUrl) return await requestBytedanceImage(req);- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/generations"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- prompt: req.prompt,- image: req.referenceImageUrl,- images: [- req.referenceImageUrl,- ...(req.additionalReferenceImageUrls || []),- ].filter(Boolean),- response_format: "url",- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return await outputImage(extractFirstImageValue(parsed?.data?.[0] || parsed), "Bytedance Seedream edit");-}--async function requestTencentImage(req: ImageGatewayRequest): Promise<string> {- return await requestOpenAICompatibleImage(req, "Tencent Hunyuan Image");-}--async function requestTencentImageEdit(req: ImageGatewayRequest): Promise<string> {- if (!req.referenceImageUrl) return await requestTencentImage(req);- const response = await fetch(joinUrl(req.channel.baseUrl, "/images/edits"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- prompt: req.prompt,- image: req.referenceImageUrl,- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return await outputImage(extractFirstImageValue(parsed?.data?.[0] || parsed), "Tencent Hunyuan Image edit");-}--async function requestGoogleImagen(req: ImageGatewayRequest, mode: "generation" | "edit"): Promise<string> {- const model = req.channel.model || "imagen-4.0-generate-001";- const baseUrl = req.channel.baseUrl.includes(":predict")- ? req.channel.baseUrl- : joinUrl(req.channel.baseUrl, `${model}:predict`);- const instance: Record<string, any> = {- prompt: req.prompt,- };- if (mode === "edit" && req.referenceImageUrl) {- const dataUrl = req.referenceImageUrl.startsWith("data:")- ? req.referenceImageUrl- : await convertRemoteImageToDataUrl(req.referenceImageUrl);- instance.image = { bytesBase64Encoded: dataUrl.split(",").pop() || dataUrl };- }-- const response = await fetch(baseUrl, {- method: "POST",- headers: {- "Content-Type": "application/json",- "x-goog-api-key": req.channel.apiKey,- },- body: JSON.stringify({- instances: [instance],- parameters: {- sampleCount: 1,- },- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return await outputImage(extractFirstImageValue(parsed?.predictions?.[0] || parsed), "Google Imagen");-}--const openaiImageAdapter: ImageAdapter = {- supports: (req) => String(req.channel.provider || "").toLowerCase() === "openai",- generate: async (req) => {- const capabilities = resolveImageCapabilities(req.channel);- if (shouldUseEditFlow(req, capabilities)) {- return await requestOpenAIImageEdit(req);- }- return await requestOpenAIImages(req);- },-};--const aliyunImageAdapter: ImageAdapter = {- supports: (req) => getProvider(req) === "aliyun",- generate: async (req) => {- return shouldUseEditFlow(req, resolveImageCapabilities(req.channel))- ? await requestAliyunImageEdit(req)- : await requestAliyunImage(req);- },-};--const tencentImageAdapter: ImageAdapter = {- supports: (req) => getProvider(req) === "tencent",- generate: async (req) => {- return shouldUseEditFlow(req, resolveImageCapabilities(req.channel))- ? await requestTencentImageEdit(req)- : await requestTencentImage(req);- },-};--const bytedanceImageAdapter: ImageAdapter = {- supports: (req) => getProvider(req) === "bytedance",- generate: async (req) => {- return shouldUseEditFlow(req, resolveImageCapabilities(req.channel))- ? await requestBytedanceImageEdit(req)- : await requestBytedanceImage(req);- },-};--const googleImageAdapter: ImageAdapter = {- supports: (req) => getProvider(req) === "google",- generate: async (req) => {- return shouldUseEditFlow(req, resolveImageCapabilities(req.channel))- ? await requestGoogleImagen(req, "edit")- : await requestGoogleImagen(req, "generation");- },-};--const imageAdapters: ImageAdapter[] = [- openaiImageAdapter,- aliyunImageAdapter,- tencentImageAdapter,- bytedanceImageAdapter,- googleImageAdapter,-];--function findImageAdapter(req: ImageGatewayRequest): ImageAdapter | null {- return imageAdapters.find((adapter) => adapter.supports(req)) || null;-}--export async function generateTextThroughGateway(req: TextGatewayRequest): Promise<string> {- const route = resolveTextRoute(req.channel);- if (route.protocol === "custom") {- const spec = parseCustomProviderMapping(req.channel.customMapping);- if (!spec) throw new Error("Custom text mapping is required.");- const result = await executeCustomSpec({- spec,- channel: req.channel,- context: {- apiKey: req.channel.apiKey,- baseUrl: req.channel.baseUrl,- model: req.channel.model,- messages: req.messages,- prompt: extractTextFromOpenAIContent(req.messages[req.messages.length - 1]?.content),- },- });- return typeof result === "string" ? result : JSON.stringify(result ?? "");- }-- if (route.protocol === "google-gemini") {- const { system, nonSystem } = splitSystemMessages(req.messages);- const response = await fetch(joinUrl(req.channel.baseUrl, `/models/${req.channel.model}:generateContent`), {- method: "POST",- headers: {- "Content-Type": "application/json",- "x-goog-api-key": req.channel.apiKey,- },- body: JSON.stringify({- ...(system ? { systemInstruction: { parts: [{ text: system }] } } : {}),- contents: nonSystem.map((message) => ({- role: message.role === "assistant" ? "model" : "user",- parts: toGoogleParts(message.content),- })),- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }- return extractTextFromGoogleResponse(parsed);- }-- if (route.protocol === "anthropic-messages") {- const { system, nonSystem } = splitSystemMessages(req.messages);- const response = await fetch(joinUrl(req.channel.baseUrl, "/messages"), {- method: "POST",- headers: {- "Content-Type": "application/json",- "x-api-key": req.channel.apiKey,- "anthropic-version": "2023-06-01",- },- body: JSON.stringify({- model: req.channel.model,- max_tokens: 4096,- ...(system ? { system } : {}),- messages: nonSystem.map((message) => ({- role: message.role === "assistant" ? "assistant" : "user",- content: toAnthropicContent(message.content),- })),- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }- return extractTextFromAnthropicResponse(parsed);- }-- const response = await fetch(joinUrl(req.channel.baseUrl, "/chat/completions"), {- method: "POST",- headers: {- Authorization: `Bearer ${req.channel.apiKey}`,- "Content-Type": "application/json",- },- body: JSON.stringify({- model: req.channel.model,- messages: req.messages,- stream: false,- }),- });-- const text = await response.text();- let parsed: any = null;- try {- parsed = text ? JSON.parse(text) : null;- } catch {- parsed = null;- }-- if (!response.ok) {- throw new Error(parsed?.error?.message || parsed?.error || text || `Request failed with status ${response.status}`);- }-- return extractTextFromOpenAIContent(parsed?.choices?.[0]?.message?.content);-}--export async function generateImageThroughGateway(req: ImageGatewayRequest): Promise<string> {- const route = resolveImageRoute(req.channel);- const capabilities = resolveImageCapabilities(req.channel);- if (route.protocol === "custom") {- const spec = parseCustomProviderMapping(req.channel.customMapping);- if (!spec) throw new Error("Custom image mapping is required.");- const result = await executeCustomSpec({- spec,- channel: req.channel,- context: {- apiKey: req.channel.apiKey,- baseUrl: req.channel.baseUrl,- model: req.channel.model,- prompt: req.prompt,- referenceImageUrl: req.referenceImageUrl || "",- maskImageUrl: req.maskImageUrl || "",- additionalReferenceImageUrls: req.additionalReferenceImageUrls || [],- referenceImages: [- ...(req.referenceImageUrl ? [req.referenceImageUrl] : []),- ...(req.maskImageUrl ? [req.maskImageUrl] : []),- ...(req.additionalReferenceImageUrls || []),- ],- },- });- const url = typeof result === "string" ? result : String(result || "");- return await convertRemoteImageToDataUrl(url);- }-- const adapter = findImageAdapter(req);- if (adapter) {- return await adapter.generate(req);- }-- if (shouldUseEditFlow(req, capabilities)) {- return await requestOpenAIImageEdit(req);- }-- if (route.protocol === "openai-images" || route.protocol === "openai-images-fallback-chat-image") {- try {- return await requestOpenAIImages(req);- } catch (imagesError) {- if (route.protocol === "openai-images-fallback-chat-image") {- try {- return await requestOpenAIChatImage(req);- } catch {- throw imagesError;- }- }- throw imagesError;- }- }-- return await requestOpenAIChatImage(req);-}diff --git a/src/lib/ai/provider-registry.ts b/src/lib/ai/provider-registry.tsdeleted file mode 100644index 8688a52..0000000--- a/src/lib/ai/provider-registry.ts+++ /dev/null@@ -1,418 +0,0 @@-export type AIProviderId =- | "openai"- | "ollama"- | "deepseek"- | "kimi"- | "aliyun"- | "tencent"- | "bytedance"- | "zhipu"- | "baidu"- | "minimax"- | "xai"- | "google"- | "anthropic"- | "custom";--export type TextProtocol = "openai-chat" | "google-gemini" | "anthropic-messages" | "custom";-export type ImageProtocol = "openai-images" | "openai-chat-image" | "openai-images-fallback-chat-image" | "custom";--export interface CustomProviderMappingSpec {- endpoint: string;- method?: "GET" | "POST" | "PUT" | "PATCH";- headers?: Record<string, string>;- body?: unknown;- responsePath: string;- supportsVision?: boolean;- supportsReferenceImages?: boolean;-}--export interface AIChannelConfig {- provider: AIProviderId | string;- apiKey: string;- baseUrl: string;- model: string;- customMapping?: string;-}--export interface AIConfig {- textProvider: AIProviderId | string;- textApiKey: string;- textBaseUrl: string;- textModel: string;- textCustomMapping?: string;- imageProvider: AIProviderId | string;- imageApiKey: string;- imageBaseUrl: string;- imageModel: string;- imageCustomMapping?: string;- fileParserApiToken?: string;- systemPrompt?: string;- apiKey?: string;- baseUrl?: string;- chatModel?: string;- imageModelLegacy?: string;-}--export interface ProviderOption {- id: AIProviderId;- label: string;- defaultBaseUrl?: string;-}--export interface TextRoute {- protocol: TextProtocol;- supportsVision: boolean;-}--export interface ImageRoute {- protocol: ImageProtocol;- supportsReferenceImages: boolean;-}--export interface ImageProviderCapabilities {- supportsGeneration: boolean;- supportsEdits: boolean;- supportsMask: boolean;- supportsReferenceImages: boolean;- supportsMultiReferenceImages: boolean;-}--const TEXT_VISION_RULES: Array<{ provider: AIProviderId; patterns: RegExp[] }> = [- {- provider: "openai",- patterns: [/gpt-4o/i, /gpt-4\.1/i, /gpt-5/i, /o1/i, /o3/i],- },- {- provider: "ollama",- patterns: [/llava/i, /bakllava/i, /minicpm-v/i, /qwen2\.5-vl/i, /qwen2-vl/i, /vision/i, /vl/i, /gemma3/i],- },- {- provider: "kimi",- patterns: [/vision/i, /vl/i, /moonshot/i, /kimi/i],- },- {- provider: "aliyun",- patterns: [/vl/i, /vision/i, /qwen2\.5-vl/i, /qvq/i],- },- {- provider: "zhipu",- patterns: [/glm-4v/i, /glm-4\.1v/i, /glm-4\.5v/i, /cogvlm/i],- },- {- provider: "baidu",- patterns: [/vision/i, /vl/i, /ernie-4\.5-vl/i],- },- {- provider: "tencent",- patterns: [/vision/i, /vl/i, /hunyuan-vision/i],- },- {- provider: "bytedance",- patterns: [/vision/i, /vl/i, /doubao-vision/i],- },- {- provider: "google",- patterns: [/gemini/i],- },- {- provider: "anthropic",- patterns: [/claude/i],- },- {- provider: "minimax",- patterns: [/vision/i, /vl/i, /minimax-vl/i],- },- {- provider: "xai",- patterns: [/vision/i, /grok/i],- },-];--const IMAGE_OPENAI_IMAGES_RULES: Array<{ provider: AIProviderId; patterns: RegExp[] }> = [- {- provider: "openai",- patterns: [/gpt-image/i, /dall-e/i],- },-];--const IMAGE_REFERENCE_RULES: Array<{ provider: AIProviderId; patterns: RegExp[] }> = [- {- provider: "openai",- patterns: [/gpt-image/i],- },- {- provider: "aliyun",- patterns: [/wan/i, /qwen-image/i],- },- {- provider: "bytedance",- patterns: [/seedream/i],- },- {- provider: "tencent",- patterns: [/hunyuan/i],- },-];--const IMAGE_CAPABILITY_RULES: Array<{- provider: AIProviderId;- patterns: RegExp[];- capabilities: Partial<ImageProviderCapabilities>;-}> = [- {- provider: "openai",- patterns: [/gpt-image/i, /dall-e/i],- capabilities: {- supportsGeneration: true,- supportsEdits: true,- supportsMask: true,- supportsReferenceImages: true,- supportsMultiReferenceImages: false,- },- },- {- provider: "aliyun",- patterns: [/wan/i, /qwen-image/i],- capabilities: {- supportsGeneration: true,- supportsEdits: true,- supportsMask: false,- supportsReferenceImages: true,- supportsMultiReferenceImages: true,- },- },- {- provider: "bytedance",- patterns: [/seedream/i],- capabilities: {- supportsGeneration: true,- supportsEdits: true,- supportsMask: false,- supportsReferenceImages: true,- supportsMultiReferenceImages: true,- },- },- {- provider: "tencent",- patterns: [/hunyuan/i],- capabilities: {- supportsGeneration: true,- supportsEdits: true,- supportsMask: false,- supportsReferenceImages: true,- supportsMultiReferenceImages: false,- },- },- {- provider: "google",- patterns: [/imagen/i],- capabilities: {- supportsGeneration: true,- supportsEdits: true,- supportsMask: true,- supportsReferenceImages: true,- supportsMultiReferenceImages: false,- },- },-];--export const TEXT_PROVIDER_OPTIONS: ProviderOption[] = [- { id: "openai", label: "OpenAI", defaultBaseUrl: "https://api.openai.com/v1" },- { id: "ollama", label: "Ollama", defaultBaseUrl: "http://localhost:11434/v1" },- { id: "deepseek", label: "DeepSeek", defaultBaseUrl: "https://api.deepseek.com/v1" },- { id: "kimi", label: "Kimi", defaultBaseUrl: "https://api.moonshot.cn/v1" },- { id: "aliyun", label: "Alibaba Bailian", defaultBaseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1" },- { id: "tencent", label: "Tencent Hunyuan", defaultBaseUrl: "https://api.hunyuan.cloud.tencent.com/v1" },- { id: "bytedance", label: "Bytedance Ark", defaultBaseUrl: "https://ark.cn-beijing.volces.com/api/v3" },- { id: "zhipu", label: "Zhipu", defaultBaseUrl: "https://open.bigmodel.cn/api/paas/v4" },- { id: "baidu", label: "Baidu Qianfan", defaultBaseUrl: "https://qianfan.baidubce.com/v2" },- { id: "minimax", label: "MiniMax", defaultBaseUrl: "https://api.minimax.io/v1" },- { id: "xai", label: "xAI", defaultBaseUrl: "https://api.x.ai/v1" },- { id: "google", label: "Google", defaultBaseUrl: "https://generativelanguage.googleapis.com/v1beta" },- { id: "anthropic", label: "Anthropic", defaultBaseUrl: "https://api.anthropic.com/v1" },-];--export const IMAGE_PROVIDER_OPTIONS: ProviderOption[] = [- { id: "openai", label: "OpenAI", defaultBaseUrl: "https://api.openai.com/v1" },- { id: "aliyun", label: "Alibaba Wanx", defaultBaseUrl: "https://dashscope.aliyuncs.com/api/v1" },- { id: "tencent", label: "Tencent Hunyuan Image", defaultBaseUrl: "https://tokenhub.tencentmaas.com/v1/api/image/lite" },- { id: "bytedance", label: "Bytedance Seedream", defaultBaseUrl: "https://ark.cn-beijing.volces.com/api/v3" },- { id: "google", label: "Google Imagen", defaultBaseUrl: "https://generativelanguage.googleapis.com/v1beta" },-];--function matchesRule(provider: string, model: string, rules: Array<{ provider: AIProviderId; patterns: RegExp[] }>) {- const normalizedProvider = String(provider || "").toLowerCase();- const normalizedModel = String(model || "");- return rules.some(- (rule) =>- rule.provider === normalizedProvider &&- rule.patterns.some((pattern) => pattern.test(normalizedModel)),- );-}--export function getDefaultBaseUrl(provider: string, kind: "text" | "image"): string {- const options = kind === "text" ? TEXT_PROVIDER_OPTIONS : IMAGE_PROVIDER_OPTIONS;- return options.find((item) => item.id === provider)?.defaultBaseUrl || "";-}--export function parseCustomProviderMapping(raw: string | undefined): CustomProviderMappingSpec | null {- const text = String(raw || "").trim();- if (!text) return null;- const parsed = JSON.parse(text) as CustomProviderMappingSpec;- if (!parsed || typeof parsed !== "object" || !parsed.endpoint || !parsed.responsePath) {- throw new Error("Invalid custom mapping: endpoint and responsePath are required.");- }- return parsed;-}--export function resolveTextRoute(channel: AIChannelConfig): TextRoute {- const provider = String(channel.provider || "").toLowerCase();- if (provider === "custom" && String(channel.customMapping || "").trim()) {- const spec = parseCustomProviderMapping(channel.customMapping);- return {- protocol: "custom",- supportsVision: spec?.supportsVision === true,- };- }-- if (provider === "google") {- return {- protocol: "google-gemini",- supportsVision: true,- };- }-- if (provider === "anthropic") {- return {- protocol: "anthropic-messages",- supportsVision: true,- };- }-- return {- protocol: "openai-chat",- supportsVision: matchesRule(provider, channel.model, TEXT_VISION_RULES),- };-}--export function resolveImageRoute(channel: AIChannelConfig): ImageRoute {- const provider = String(channel.provider || "").toLowerCase();- if (provider === "custom" && String(channel.customMapping || "").trim()) {- const spec = parseCustomProviderMapping(channel.customMapping);- return {- protocol: "custom",- supportsReferenceImages: spec?.supportsReferenceImages === true,- };- }-- if (provider === "custom") {- return {- protocol: "openai-images-fallback-chat-image",- supportsReferenceImages: false,- };- }-- return {- protocol: matchesRule(provider, channel.model, IMAGE_OPENAI_IMAGES_RULES)- ? "openai-images"- : "openai-chat-image",- supportsReferenceImages: matchesRule(provider, channel.model, IMAGE_REFERENCE_RULES),- };-}--export function resolveImageCapabilities(channel: AIChannelConfig): ImageProviderCapabilities {- const provider = String(channel.provider || "").toLowerCase();- const model = String(channel.model || "");-- const matched = IMAGE_CAPABILITY_RULES.find(- (rule) => rule.provider === provider && rule.patterns.some((pattern) => pattern.test(model)),- );-- if (matched) {- return {- supportsGeneration: true,- supportsEdits: true,- supportsMask: false,- supportsReferenceImages: false,- supportsMultiReferenceImages: false,- ...matched.capabilities,- };- }-- if (provider === "custom") {- const spec = String(channel.customMapping || "").trim() ? parseCustomProviderMapping(channel.customMapping) : null;- return {- supportsGeneration: true,- supportsEdits: Boolean(spec?.supportsReferenceImages),- supportsMask: false,- supportsReferenceImages: Boolean(spec?.supportsReferenceImages),- supportsMultiReferenceImages: Boolean(spec?.supportsReferenceImages),- };- }-- return {- supportsGeneration: true,- supportsEdits: false,- supportsMask: false,- supportsReferenceImages: false,- supportsMultiReferenceImages: false,- };-}--export function getTextChannelConfig(config: AIConfig): AIChannelConfig {- return {- provider: String(config.textProvider || "openai"),- apiKey: String(config.textApiKey || config.apiKey || "").trim(),- baseUrl: String(config.textBaseUrl || config.baseUrl || getDefaultBaseUrl(String(config.textProvider || "openai"), "text") || "").trim(),- model: String(config.textModel || config.chatModel || "").trim(),- customMapping: String(config.textCustomMapping || "").trim(),- };-}--export function getImageChannelConfig(config: AIConfig): AIChannelConfig {- return {- provider: String(config.imageProvider || "openai"),- apiKey: String(config.imageApiKey || config.apiKey || "").trim(),- baseUrl: String(config.imageBaseUrl || config.baseUrl || getDefaultBaseUrl(String(config.imageProvider || "openai"), "image") || "").trim(),- model: String(config.imageModel || config.imageModelLegacy || "").trim(),- customMapping: String(config.imageCustomMapping || "").trim(),- };-}--export function normalizeAIConfig(rawConfig: Partial<AIConfig> | null | undefined): AIConfig {- const raw = rawConfig || {};- const requestedTextProvider = String((raw.textProvider || "openai") as string).trim() || "openai";- const requestedImageProvider = String((raw.imageProvider || "openai") as string).trim() || "openai";- const textProvider = requestedTextProvider === "custom" ? "openai" : requestedTextProvider;- const imageProvider = IMAGE_PROVIDER_OPTIONS.some((option) => option.id === requestedImageProvider)- ? requestedImageProvider- : "openai";- const textApiKey = String((raw.textApiKey || raw.apiKey || "") as string).trim();- const imageApiKey = String((raw.imageApiKey || raw.apiKey || "") as string).trim();- const textBaseUrl =- String((raw.textBaseUrl || raw.baseUrl || getDefaultBaseUrl(textProvider, "text") || "") as string).trim();- const imageBaseUrl =- String((raw.imageBaseUrl || raw.baseUrl || getDefaultBaseUrl(imageProvider, "image") || "") as string).trim();- const textModel = String((raw.textModel || raw.chatModel || "") as string).trim();- const imageModel = String((raw.imageModel || raw.imageModelLegacy || "") as string).trim();-- return {- textProvider,- textApiKey,- textBaseUrl,- textModel,- textCustomMapping: "",- imageProvider,- imageApiKey,- imageBaseUrl,- imageModel,- imageCustomMapping: "",- fileParserApiToken: String(raw.fileParserApiToken || "").trim(),- systemPrompt: String(raw.systemPrompt || "").trim(),- apiKey: textApiKey,- baseUrl: textBaseUrl,- chatModel: textModel,- imageModelLegacy: imageModel,- };-}diff --git a/src/lib/cad-tasks.ts b/src/lib/cad-tasks.tsindex 27f5479..e4e74c3 100644--- a/src/lib/cad-tasks.ts+++ b/src/lib/cad-tasks.ts@@ -1,4 +1,4 @@-import type { ChatMessage } from "@/lib/ai-client";+import type { ChatMessage } from "@/ai/client";import cadBomPromptTemplate from "../../agent/cad/bom-prompt.md?raw";import cadImagesMasterRenovationSchemeTemplate from "../../agent/cad/images-agents/master-renovation-scheme.md?raw";diff --git a/src/lib/ppt-service.ts b/src/lib/ppt-service.tsindex b7bac7b..fb443a9 100644--- a/src/lib/ppt-service.ts+++ b/src/lib/ppt-service.ts@@ -1,6 +1,6 @@import PptxGenJS from "pptxgenjs"import { PDFDocument, rgb } from "pdf-lib"-import { generateChatMessage, generateImage, generateVisionChatMessage } from "./ai-client"+import { generateChatMessage, generateImage, generateVisionChatMessage } from "@/ai/client"import { buildRisenPrompt } from "./risen-prompt"import pptOutlineSystem from "../../agent/ppt/outline.md?raw"import pptSlidesGenerateSystem from "../../agent/ppt/slides-generate.md?raw"diff --git a/src/server/routes/chat.ts b/src/server/routes/chat.tsindex 3e33196..6e556c2 100644--- a/src/server/routes/chat.ts+++ b/src/server/routes/chat.ts@@ -13,16 +13,10 @@ import path from "path"import { z } from "zod"import { extractText, getDocumentProxy } from "unpdf"import mammoth from "mammoth"-import { generateImageThroughGateway } from "../../lib/ai/gateway"-import {- getImageChannelConfig,- getTextChannelConfig,- normalizeAIConfig,-} from "../../lib/ai/provider-registry"-import {- getAIModel,- supportsPromptCaching,-} from "../../workspaces/flow/next/lib/ai-providers"+import { generateImage } from "../../ai/image"+import { getImageChannel, getTextChannel, normalizeAIConfig } from "../../ai/config"+import type { AIChannel } from "../../ai/types"+import { getChatModel } from "../../ai/server-model"import { formatAvailableShapeLibraries } from "../../workspaces/flow/next/lib/shape-library"import {getTelemetryConfig,@@ -163,16 +157,9 @@ async function extractUploadedFileText(file: UploadedFilePayload): Promise<strinreturn providedText || cleanExtractedText(buffer.toString("utf8"))}-async function summarizeChunk(- model: any,- providerOptions: any,- headers: any,- chunk: string,-): Promise<string> {+async function summarizeChunk(model: any, chunk: string): Promise<string> {return generateSummaryWithRetry({model,- providerOptions,- headers,system:"R - Role\nYou are a strict summarizer.\n\nI - Instructions\nOutput only direct summary text.\n\nE - End Goal\nProduce a concise faithful summary.\n\nN - Narrowing\nNo questions. No instructions. No extra framing.",user: `任务:仅根据下述文本输出摘要(<=300字)。禁止提问、禁止让用户补充内容、禁止输出模板化客套。\n\n文本:\n${chunk}`,@@ -238,8 +225,6 @@ function isBadSummaryText(text: string): boolean {async function generateSummaryWithRetry(args: {model: any- providerOptions: any- headers: anysystem: stringuser: stringmaxOutputTokens: number@@ -253,8 +238,6 @@ async function generateSummaryWithRetry(args: {{ role: "system" as const, content: args.system },{ role: "user" as const, content: args.user },],- ...(args.providerOptions && { providerOptions: args.providerOptions }),- ...(args.headers && { headers: args.headers }),maxOutputTokens: args.maxOutputTokens,})const text = sanitizeSummaryText(@@ -267,23 +250,16 @@ async function generateSummaryWithRetry(args: {return extractiveFallbackSummary(args.fallbackSource, args.maxChars)}-async function summarizeBlockMethod(args: {- model: any- providerOptions: any- headers: any- text: string-}): Promise<string> {+async function summarizeBlockMethod(args: { model: any; text: string }): Promise<string> {const chunks = splitByApproxTokens(args.text, CHUNK_TOKEN_TARGET)const partial = await runWithConcurrency(chunks,SUMMARY_CONCURRENCY,- (chunk) => summarizeChunk(args.model, args.providerOptions, args.headers, chunk),+ (chunk) => summarizeChunk(args.model, chunk),)const merged = partial.filter(Boolean).join("\n")return generateSummaryWithRetry({model: args.model,- providerOptions: args.providerOptions,- headers: args.headers,system:"R - Role\nYou merge chunk summaries.\n\nI - Instructions\nOutput only concise faithful summary text.\n\nE - End Goal\nProduce one merged summary.\n\nN - Narrowing\nNo questions. No instructions. No extra framing.",user: `任务:将以下分块摘要合并为完整摘要(<=1000字)。禁止提问、禁止请求补充材料,仅输出摘要正文。\n\n${merged}`,@@ -293,17 +269,12 @@ async function summarizeBlockMethod(args: {})}-async function summarizeRecursiveMethod(args: {- model: any- providerOptions: any- headers: any- text: string-}): Promise<string> {+async function summarizeRecursiveMethod(args: { model: any; text: string }): Promise<string> {const baseChunks = splitByApproxTokens(args.text, 3000)let level = await runWithConcurrency(baseChunks,SUMMARY_CONCURRENCY,- (chunk) => summarizeChunk(args.model, args.providerOptions, args.headers, chunk),+ (chunk) => summarizeChunk(args.model, chunk),)level = level.filter(Boolean)@@ -318,8 +289,6 @@ async function summarizeRecursiveMethod(args: {async (group) =>generateSummaryWithRetry({model: args.model,- providerOptions: args.providerOptions,- headers: args.headers,system:"R - Role\nYou merge summaries into a higher-level summary.\n\nI - Instructions\nOutput only summary text.\n\nE - End Goal\nProduce one higher-level merged summary.\n\nN - Narrowing\nNo questions. No instructions. No extra framing.",user: `任务:把这组摘要提炼成更高层摘要(<=350字)。禁止提问、禁止让用户补充,只输出摘要。\n\n${group}`,@@ -333,8 +302,6 @@ async function summarizeRecursiveMethod(args: {return generateSummaryWithRetry({model: args.model,- providerOptions: args.providerOptions,- headers: args.headers,system:"R - Role\nYou produce final executive summaries.\n\nI - Instructions\nKeep hierarchical fidelity and avoid hallucination. Output only summary text.\n\nE - End Goal\nProduce one final executive summary.\n\nN - Narrowing\nNo questions. No requests for more info. No extra framing.",user: `任务:输出最终摘要(<=1000字)。禁止提问、禁止请求更多信息,仅输出摘要正文。\n\n${level.join("\n")}`,@@ -556,13 +523,7 @@ async function generateDeepThinkingDiagramImage(args: {globalConstraints: stringprocessedFilesContext: stringimageAttachments: ImageAttachment[]- channel: {- provider: string- apiKey: string- baseUrl: string- model: string- customMapping?: string- }+ channel: AIChannel}): Promise<string | null> {let promptText = ""try {@@ -598,7 +559,7 @@ async function generateDeepThinkingDiagramImage(args: {].join("\n\n")}- return await generateImageThroughGateway({+ return await generateImage({channel: args.channel,prompt: promptText,referenceImageUrl: args.imageAttachments[0]?.url,@@ -815,51 +776,21 @@ async function handleChatRequest(req: Request): Promise<Response> {const normalizedClientConfig = normalizeAIConfig(typeof bodyAIConfig === "object" && bodyAIConfig ? bodyAIConfig : {},)- const textChannel = getTextChannelConfig(normalizedClientConfig)- const imageChannel = getImageChannelConfig(normalizedClientConfig)-- // Read client AI provider overrides from headers first, then body fallback.- const clientOverrides = {- provider:- req.headers.get("x-ai-provider") ||- (typeof bodyAIConfig.provider === "string"- ? bodyAIConfig.provider- : textChannel.provider || null),- baseUrl:- req.headers.get("x-ai-base-url") ||- (typeof bodyAIConfig.baseUrl === "string"- ? bodyAIConfig.baseUrl- : textChannel.baseUrl || null),- apiKey:- req.headers.get("x-ai-api-key") ||- (typeof bodyAIConfig.apiKey === "string"- ? bodyAIConfig.apiKey- : textChannel.apiKey || null),- modelId:- req.headers.get("x-ai-model") ||- req.headers.get("x-ai-chat-model") ||- (typeof bodyAIConfig.chatModel === "string"- ? bodyAIConfig.chatModel- : typeof bodyAIConfig.modelId === "string"- ? bodyAIConfig.modelId- : textChannel.model || null),+ // Settings arrive in the request body; headers stay supported so an+ // embedding host can override them without rewriting the payload.+ const header = (name: string) => req.headers.get(name) || ""+ const textChannel = {+ ...getTextChannel(normalizedClientConfig),+ ...(header("x-ai-base-url") ? { baseUrl: header("x-ai-base-url") } : {}),+ ...(header("x-ai-api-key") ? { apiKey: header("x-ai-api-key") } : {}),}- const imageModelId =- req.headers.get("x-ai-image-model") ||- (typeof bodyAIConfig.imageModel === "string"- ? bodyAIConfig.imageModel- : imageChannel.model || null)+ const headerModel = header("x-ai-model") || header("x-ai-chat-model")+ if (headerModel) textChannel.model = headerModel- // Get AI model with optional client overrides- const { model, providerOptions, headers, modelId } =- getAIModel(clientOverrides)+ const imageChannel = getImageChannel(normalizedClientConfig)+ const imageModelId = header("x-ai-image-model") || imageChannel.model || null-- // Check if model supports prompt caching- const shouldCache = supportsPromptCaching(modelId)- console.log(- `[Prompt Caching] ${shouldCache ? "ENABLED" : "DISABLED"} for model: ${modelId}`,- )+ const { model, modelId } = getChatModel(textChannel)let parsedFilesContext = ""if (uploadedFiles.length > 0) {@@ -873,18 +804,8 @@ async function handleChatRequest(req: Request): Promise<Response> {tokens >= RECURSIVE_THRESHOLD_TOKENS ? "recursive" : "chunk"const summary =method === "recursive"- ? await summarizeRecursiveMethod({- model,- providerOptions,- headers,- text: extracted,- })- : await summarizeBlockMethod({- model,- providerOptions,- headers,- text: extracted,- })+ ? await summarizeRecursiveMethod({ model, text: extracted })+ : await summarizeBlockMethod({ model, text: extracted })return { name: file.name, tokens, method, summary }},)@@ -1269,47 +1190,17 @@ ${parsedFilesContext ? `\n\nParsed file summaries:\n"""md\n${parsedFilesContext}}}- // Add cache point to the last assistant message in conversation history- // This caches the entire conversation prefix for subsequent requests- // Strategy: system (cached) + history with last assistant (cached) + new user message- if (shouldCache && enhancedMessages.length >= 2) {- // Find the last assistant message (should be second-to-last, before current user message)- for (let i = enhancedMessages.length - 2; i >= 0; i--) {- if (enhancedMessages[i].role === "assistant") {- enhancedMessages[i] = {- ...enhancedMessages[i],- providerOptions: {- bedrock: { cachePoint: { type: "default" } },- },- }- break // Only cache the last assistant message- }- }- }-- // System messages with multiple cache breakpoints for optimal caching:- // - Breakpoint 1: Static instructions (~1500 tokens) - rarely changes- // - Breakpoint 2: Current XML context - changes per diagram, but constant within a conversation turn- // This allows: if only user message changes, both system caches are reused- // if XML changes, instruction cache is still reused+ // Two system messages rather than one: the instructions are stable across a+ // conversation while the diagram XML changes every turn, so keeping them+ // apart lets the provider reuse the cached instruction prefix.const systemMessages = [{role: "system" as const,content: systemMessage,- ...(shouldCache && {- providerOptions: {- bedrock: { cachePoint: { type: "default" } },- },- }),},{role: "system" as const,content: `${previousXml ? `Previous diagram XML (before user's last message):\n"""xml\n${previousXml}\n"""\n\n` : ""}Current diagram XML (AUTHORITATIVE - the source of truth):\n"""xml\n${xml || ""}\n"""\n\nIMPORTANT: The "Current diagram XML" is the SINGLE SOURCE OF TRUTH for what's on the canvas right now. The user can manually add, delete, or modify shapes directly in draw.io. Always count and describe elements based on the CURRENT XML, not on what you previously generated. If both previous and current XML are shown, compare them to understand what the user changed. When using edit_diagram, COPY search patterns exactly from the CURRENT XML - attribute order matters!`,- ...(shouldCache && {- providerOptions: {- bedrock: { cachePoint: { type: "default" } },- },- }),},]@@ -1352,8 +1243,6 @@ ${parsedFilesContext ? `\n\nParsed file summaries:\n"""md\n${parsedFilesContext}stopWhen: stepCountIs(5),messages: allMessages,...(safeMaxTokens && { maxTokens: safeMaxTokens }),- ...(providerOptions && { providerOptions }),- ...(headers && { headers }),...(getTelemetryConfig({ sessionId: validSessionId, userId }) && {experimental_telemetry: getTelemetryConfig({sessionId: validSessionId,diff --git a/src/server/routes/ppt-ai.ts b/src/server/routes/ppt-ai.tsindex 321be44..1a655e6 100644--- a/src/server/routes/ppt-ai.ts+++ b/src/server/routes/ppt-ai.ts@@ -1,27 +1,18 @@-import {- generateImageThroughGateway,- generateTextThroughGateway,-} from "../../lib/ai/gateway"-import {- getImageChannelConfig,- getTextChannelConfig,- normalizeAIConfig,- type AIConfig,-} from "../../lib/ai/provider-registry"+import { generateImage } from "../../ai/image"+import { generateText } from "../../ai/text"+import { getImageChannel, getTextChannel, normalizeAIConfig } from "../../ai/config"+import type { AIConfig, MultimodalMessage } from "../../ai/types"-type ProxyChatMessage = {- role: "system" | "user" | "assistant"- content: any-}+/**+ * Non-streaming model proxy for the CAD and PPT workspaces.+ *+ * The browser never talks to the provider directly, so the API key stays out+ * of cross-origin requests and remote images can be inlined server-side.+ */-type PptAIRequestBody =+type RequestBody =+ | { kind: "chat"; aiConfig?: Partial<AIConfig>; messages?: MultimodalMessage[]; model?: string }| {- kind: "chat"- aiConfig?: Partial<AIConfig>- messages?: ProxyChatMessage[]- model?: string- }- | {kind: "image"aiConfig?: Partial<AIConfig>prompt?: string@@ -33,75 +24,53 @@ type PptAIRequestBody =const jsonHeaders = { "Content-Type": "application/json" }-function getErrorMessage(error: unknown) {- return error instanceof Error ? error.message : "Unknown error"-}+const badRequest = (error: string) =>+ Response.json({ error }, { status: 400, headers: jsonHeaders })++const optionalString = (value: unknown) =>+ typeof value === "string" && value ? value : undefinedexport async function POST(req: Request) {try {- const body = (await req.json()) as PptAIRequestBody+ const body = (await req.json()) as RequestBodyconst config = normalizeAIConfig(body?.aiConfig)if (body.kind === "chat") {const messages = Array.isArray(body.messages) ? body.messages : []- const textChannel = getTextChannelConfig(config)- textChannel.model = String(body.model || textChannel.model || "").trim()+ const channel = { ...getTextChannel(config), model: String(body.model || config.textModel || "").trim() }- if (!textChannel.apiKey || !textChannel.model || messages.length === 0) {- return Response.json(- { error: "Missing text model, API key, or messages." },- { status: 400, headers: jsonHeaders },- )+ if (!channel.apiKey || !channel.model || messages.length === 0) {+ return badRequest("Missing text model, API key, or messages.")}- const content = await generateTextThroughGateway({- channel: textChannel,- messages,- })+ const content = await generateText({ channel, messages })return Response.json({ content }, { headers: jsonHeaders })}if (body.kind === "image") {const prompt = String(body.prompt || "").trim()- const imageChannel = getImageChannelConfig(config)- imageChannel.model = String(body.model || imageChannel.model || "").trim()+ const channel = { ...getImageChannel(config), model: String(body.model || config.imageModel || "").trim() }- if (!imageChannel.apiKey || !imageChannel.model || !prompt) {- return Response.json(- { error: "Missing image model, API key, or prompt." },- { status: 400, headers: jsonHeaders },- )+ if (!channel.apiKey || !channel.model || !prompt) {+ return badRequest("Missing image model, API key, or prompt.")}- const url = await generateImageThroughGateway({- channel: imageChannel,+ const url = await generateImage({+ channel,prompt,- referenceImageUrl:- typeof body.referenceImageUrl === "string"- ? body.referenceImageUrl- : undefined,- maskImageUrl:- typeof body.maskImageUrl === "string"- ? body.maskImageUrl- : undefined,- additionalReferenceImageUrls: Array.isArray(- body.additionalReferenceImageUrls,- )- ? body.additionalReferenceImageUrls- .map((x) => String(x || ""))- .filter(Boolean)+ referenceImageUrl: optionalString(body.referenceImageUrl),+ maskImageUrl: optionalString(body.maskImageUrl),+ additionalReferenceImageUrls: Array.isArray(body.additionalReferenceImageUrls)+ ? body.additionalReferenceImageUrls.map((x) => String(x || "")).filter(Boolean): [],})return Response.json({ url }, { headers: jsonHeaders })}- return Response.json(- { error: "Unsupported PPT AI request kind." },- { status: 400, headers: jsonHeaders },- )+ return badRequest("Unsupported PPT AI request kind.")} catch (error) {return Response.json(- { error: getErrorMessage(error) },+ { error: error instanceof Error ? error.message : "Unknown error" },{ status: 500, headers: jsonHeaders },)}diff --git a/src/workspaces/cad/CadWorkspaceShell.tsx b/src/workspaces/cad/CadWorkspaceShell.tsxindex 600e28c..20dbbd3 100644--- a/src/workspaces/cad/CadWorkspaceShell.tsx+++ b/src/workspaces/cad/CadWorkspaceShell.tsx@@ -16,8 +16,8 @@ import {type PanelImperativeHandle,} from "@/workspaces/cad/ui/resizable";import { CAD_SYSTEM_PROMPT } from "@/lib/system-prompts";-import type { ChatMessage } from "@/lib/ai-client";-import { generateImage } from "@/lib/ai-client";+import type { ChatMessage } from "@/ai/client";+import { generateImage } from "@/ai/client";import { CadWorkspace } from "@/workspaces/cad/workspace/CadWorkspace";import { ChatPanel as CadChatPanel } from "@/workspaces/cad/chat/ChatPanel";diff --git a/src/workspaces/cad/chat/ChatPanel.tsx b/src/workspaces/cad/chat/ChatPanel.tsxindex eef72e9..59575a3 100644--- a/src/workspaces/cad/chat/ChatPanel.tsx+++ b/src/workspaces/cad/chat/ChatPanel.tsx@@ -1,7 +1,7 @@import React, { useState, useRef, useEffect } from 'react';import { PanelRightClose, PanelRightOpen } from 'lucide-react';import { cn } from '@/lib/utils';-import { streamChatMessage, generateChatMessage, ChatMessage, getAIConfig } from '@/lib/ai-client';+import { streamChatMessage, generateChatMessage, ChatMessage, getAIConfig } from '@/ai/client';import { DRAWIO_SYSTEM_PROMPT } from '@/lib/system-prompts';import { ButtonWithTooltip } from '@/workspaces/cad/chat/components/button-with-tooltip';import { ChatInput } from '@/workspaces/cad/chat/ChatInput';diff --git a/src/workspaces/flow/next/components/chat-panel.tsx b/src/workspaces/flow/next/components/chat-panel.tsxindex e72beb7..234a15a 100644--- a/src/workspaces/flow/next/components/chat-panel.tsx+++ b/src/workspaces/flow/next/components/chat-panel.tsx@@ -972,11 +972,9 @@ Please fix the XML issues. Ensure cell_id values exist in the current XML and th// Reset auto-retry count on user-initiated messageautoRetryCountRef.current = 0- const config = getAIConfig()- const modelToUse = String(config.aiModel || defaultModel || "").trim()- const imageModelToUse = String(- (config as { aiImageModel?: string }).aiImageModel || "",- ).trim()+ const { accessCode, ai } = getAIConfig()+ const modelToUse = String(ai.textModel || defaultModel || "").trim()+ const imageModelToUse = String(ai.imageModel || "").trim()const globalConstraints = localStorage.getItem(STORAGE_GLOBAL_CONSTRAINTS_KEY) || ""// Safe parts access@@ -1039,38 +1037,17 @@ Please fix the XML issues. Ensure cell_id values exist in the current XML and thuploadedFiles,deepThinkingEnabled,aiConfig: {- provider: config.aiProvider,- baseUrl: config.aiBaseUrl,- apiKey: config.aiApiKey,- chatModel: modelToUse,- imageModel: imageModelToUse,- textProvider: config.aiProvider,- textBaseUrl: config.aiBaseUrl,- textApiKey: config.aiApiKey,+ ...ai,textModel: modelToUse,- imageProvider:- (config as { aiImageProvider?: string }).aiImageProvider ||- config.aiProvider,- imageBaseUrl:- (config as { aiImageBaseUrl?: string }).aiImageBaseUrl ||- config.aiBaseUrl,- imageApiKey:- (config as { aiImageApiKey?: string }).aiImageApiKey ||- config.aiApiKey,+ imageModel: imageModelToUse,},aiConstraints: globalConstraints,},headers: {- "x-access-code": config.accessCode,- ...(config.aiProvider && {- "x-ai-provider": config.aiProvider,- }),- ...(config.aiBaseUrl && {- "x-ai-base-url": config.aiBaseUrl,- }),- ...(config.aiApiKey && { "x-ai-api-key": config.aiApiKey }),+ "x-access-code": accessCode,+ ...(ai.textBaseUrl && { "x-ai-base-url": ai.textBaseUrl }),+ ...(ai.textApiKey && { "x-ai-api-key": ai.textApiKey }),...(modelToUse && { "x-ai-model": modelToUse }),- ...(modelToUse && { "x-ai-chat-model": modelToUse }),...(imageModelToUse && {"x-ai-image-model": imageModelToUse,}),diff --git a/src/workspaces/flow/next/components/settings-dialog.tsx b/src/workspaces/flow/next/components/settings-dialog.tsxindex 8b216b3..0224b52 100644--- a/src/workspaces/flow/next/components/settings-dialog.tsx+++ b/src/workspaces/flow/next/components/settings-dialog.tsx@@ -2,7 +2,7 @@import { Moon, Sun } from "lucide-react"import { useEffect, useState } from "react"-import { getAIConfig as getWorkspaceAIConfig } from "@/lib/ai-client"+import { getAIConfig as getWorkspaceAIConfig } from "@/ai/client"import { Button } from "@/workspaces/flow/next/components/ui/button"import {Dialog,@@ -110,15 +110,9 @@ export function SettingsDialog({)setCloseProtection(storedCloseProtection !== "false")- const workspaceConfig = getWorkspaceAIConfig()- setTopSettings({- textApiKey: String(workspaceConfig.textApiKey || workspaceConfig.apiKey || ""),- textBaseUrl: String(workspaceConfig.textBaseUrl || workspaceConfig.baseUrl || ""),- textModel: String(workspaceConfig.textModel || workspaceConfig.chatModel || ""),- imageApiKey: String(workspaceConfig.imageApiKey || workspaceConfig.apiKey || ""),- imageBaseUrl: String(workspaceConfig.imageBaseUrl || workspaceConfig.baseUrl || ""),- imageModel: String(workspaceConfig.imageModel || workspaceConfig.imageModelLegacy || ""),- })+ const { textApiKey, textBaseUrl, textModel, imageApiKey, imageBaseUrl, imageModel } =+ getWorkspaceAIConfig()+ setTopSettings({ textApiKey, textBaseUrl, textModel, imageApiKey, imageBaseUrl, imageModel })setError("")}diff --git a/src/workspaces/flow/next/lib/ai-config.ts b/src/workspaces/flow/next/lib/ai-config.tsindex 8cf2bf1..e0de7ec 100644--- a/src/workspaces/flow/next/lib/ai-config.ts+++ b/src/workspaces/flow/next/lib/ai-config.ts@@ -1,43 +1,23 @@-import { STORAGE_KEYS } from "./storage"-import { getAIConfig as getWorkspaceAIConfig } from "@/lib/ai-client"+import { STORAGE_KEYS } from "@/workspaces/flow/next/lib/storage"+import { DEFAULT_AI_CONFIG, getAIConfig as getWorkspaceAIConfig } from "@/ai/config"+import type { AIConfig } from "@/ai/types"/**- * Get AI configuration from localStorage.- * Returns API keys and settings for custom AI providers.- * Uses top-bar workspace settings as the primary source.+ * The Flow workspace sends its model settings to `/api/chat` alongside an+ * optional deployment access code. The model settings themselves are the+ * shared workspace config; only the access code is Flow-specific.*/-export function getAIConfig() {+export interface FlowRequestConfig {+ accessCode: string+ ai: AIConfig+}++export function getAIConfig(): FlowRequestConfig {if (typeof window === "undefined") {- return {- accessCode: "",- aiProvider: "openai",- aiBaseUrl: "",- aiApiKey: "",- aiModel: "",- aiImageModel: "",- aiImageProvider: "openai",- aiImageBaseUrl: "",- aiImageApiKey: "",- }+ return { accessCode: "", ai: DEFAULT_AI_CONFIG }}-- const topConfig = getWorkspaceAIConfig()- const topApiKey = String(topConfig.textApiKey || topConfig.apiKey || "").trim()- const topBaseUrl = String(topConfig.textBaseUrl || topConfig.baseUrl || "").trim()- const topChatModel = String(topConfig.textModel || topConfig.chatModel || "").trim()- const topImageModel = String(topConfig.imageModel || topConfig.imageModelLegacy || "").trim()- const topImageApiKey = String(topConfig.imageApiKey || topConfig.apiKey || "").trim()- const topImageBaseUrl = String(topConfig.imageBaseUrl || topConfig.baseUrl || "").trim()-return {accessCode: localStorage.getItem(STORAGE_KEYS.accessCode) || "",- aiProvider: String(topConfig.textProvider || "openai"),- aiBaseUrl: topBaseUrl,- aiApiKey: topApiKey,- aiModel: topChatModel,- aiImageModel: topImageModel,- aiImageProvider: String(topConfig.imageProvider || "openai"),- aiImageBaseUrl: topImageBaseUrl,- aiImageApiKey: topImageApiKey,+ ai: getWorkspaceAIConfig(),}}diff --git a/src/workspaces/flow/next/lib/ai-providers.ts b/src/workspaces/flow/next/lib/ai-providers.tsdeleted file mode 100644index 1af537c..0000000--- a/src/workspaces/flow/next/lib/ai-providers.ts+++ /dev/null@@ -1,472 +0,0 @@-import { createOpenAI } from "@ai-sdk/openai"--export type ProviderName =- | "bedrock"- | "openai"- | "anthropic"- | "google"- | "azure"- | "ollama"- | "openrouter"- | "deepseek"- | "siliconflow"--interface ModelConfig {- model: any- providerOptions?: any- headers?: Record<string, string>- modelId: string-}--export interface ClientOverrides {- provider?: string | null- baseUrl?: string | null- apiKey?: string | null- modelId?: string | null-}--// Providers that can be used with client-provided API keys-const _ALLOWED_CLIENT_PROVIDERS: ProviderName[] = [- "openai",- "anthropic",- "google",- "azure",- "openrouter",- "deepseek",- "siliconflow",-]--// Bedrock provider options for Anthropic beta features-const _BEDROCK_ANTHROPIC_BETA = {- bedrock: {- anthropicBeta: ["fine-grained-tool-streaming-2025-05-14"],- },-}--// Direct Anthropic API headers for beta features-const _ANTHROPIC_BETA_HEADERS = {- "anthropic-beta": "fine-grained-tool-streaming-2025-05-14",-}--/**- * Safely parse integer from environment variable with validation- */-function parseIntSafe(- value: string | undefined,- varName: string,- min?: number,- max?: number,-): number | undefined {- if (!value) return undefined- const parsed = Number.parseInt(value, 10)- if (Number.isNaN(parsed)) {- throw new Error(`${varName} must be a valid integer, got: ${value}`)- }- if (min !== undefined && parsed < min) {- throw new Error(`${varName} must be >= ${min}, got: ${parsed}`)- }- if (max !== undefined && parsed > max) {- throw new Error(`${varName} must be <= ${max}, got: ${parsed}`)- }- return parsed-}--/**- * Build provider-specific options from environment variables- * Supports various AI SDK providers with their unique configuration options- *- * Environment variables:- * - OPENAI_REASONING_EFFORT: OpenAI reasoning effort level (minimal/low/medium/high) - for o1/o3/gpt-5- * - OPENAI_REASONING_SUMMARY: OpenAI reasoning summary (none/brief/detailed) - auto-enabled for o1/o3/gpt-5- * - ANTHROPIC_THINKING_BUDGET_TOKENS: Anthropic thinking budget in tokens (1024-64000)- * - ANTHROPIC_THINKING_TYPE: Anthropic thinking type (enabled)- * - GOOGLE_THINKING_BUDGET: Google Gemini 2.5 thinking budget in tokens (1024-100000)- * - GOOGLE_THINKING_LEVEL: Google Gemini 3 thinking level (low/high)- * - AZURE_REASONING_EFFORT: Azure/OpenAI reasoning effort (low/medium/high)- * - AZURE_REASONING_SUMMARY: Azure reasoning summary (none/brief/detailed)- * - BEDROCK_REASONING_BUDGET_TOKENS: Bedrock Claude reasoning budget in tokens (1024-64000)- * - BEDROCK_REASONING_EFFORT: Bedrock Nova reasoning effort (low/medium/high)- * - OLLAMA_ENABLE_THINKING: Enable Ollama thinking mode (set to "true")- */-function _buildProviderOptions(- provider: ProviderName,- modelId?: string,-): Record<string, any> | undefined {- const options: Record<string, any> = {}-- switch (provider) {- case "openai": {- const reasoningEffort = process.env.OPENAI_REASONING_EFFORT- const reasoningSummary = process.env.OPENAI_REASONING_SUMMARY-- // OpenAI reasoning models (o1, o3, gpt-5) need reasoningSummary to return thoughts- if (- modelId &&- (modelId.includes("o1") ||- modelId.includes("o3") ||- modelId.includes("gpt-5"))- ) {- options.openai = {- // Auto-enable reasoning summary for reasoning models (default: detailed)- reasoningSummary:- (reasoningSummary as "none" | "brief" | "detailed") ||- "detailed",- }-- // Optionally configure reasoning effort- if (reasoningEffort) {- options.openai.reasoningEffort = reasoningEffort as- | "minimal"- | "low"- | "medium"- | "high"- }- } else if (reasoningEffort || reasoningSummary) {- // Non-reasoning models: only apply if explicitly configured- options.openai = {}- if (reasoningEffort) {- options.openai.reasoningEffort = reasoningEffort as- | "minimal"- | "low"- | "medium"- | "high"- }- if (reasoningSummary) {- options.openai.reasoningSummary = reasoningSummary as- | "none"- | "brief"- | "detailed"- }- }- break- }-- case "anthropic": {- const thinkingBudget = parseIntSafe(- process.env.ANTHROPIC_THINKING_BUDGET_TOKENS,- "ANTHROPIC_THINKING_BUDGET_TOKENS",- 1024,- 64000,- )- const thinkingType =- process.env.ANTHROPIC_THINKING_TYPE || "enabled"-- if (thinkingBudget) {- options.anthropic = {- thinking: {- type: thinkingType,- budgetTokens: thinkingBudget,- },- }- }- break- }-- case "google": {- const reasoningEffort = process.env.GOOGLE_REASONING_EFFORT- const thinkingBudgetVal = parseIntSafe(- process.env.GOOGLE_THINKING_BUDGET,- "GOOGLE_THINKING_BUDGET",- 1024,- 100000,- )- const thinkingLevel = process.env.GOOGLE_THINKING_LEVEL- const maxOutputTokens = parseIntSafe(- process.env.GOOGLE_MAX_OUTPUT_TOKENS,- "GOOGLE_MAX_OUTPUT_TOKENS",- 256,- 65535,- )-- if (- modelId &&- (modelId.includes("gemini-2") ||- modelId.includes("gemini-3") ||- modelId.includes("gemini2") ||- modelId.includes("gemini3"))- ) {- const thinkingConfig: Record<string, any> = {- includeThoughts: true,- }-- if (- thinkingBudgetVal &&- (modelId.includes("2.5") || modelId.includes("2-5"))- ) {- thinkingConfig.thinkingBudget = thinkingBudgetVal- } else if (- thinkingLevel &&- (modelId.includes("gemini-3") ||- modelId.includes("gemini3"))- ) {- thinkingConfig.thinkingLevel = thinkingLevel as- | "low"- | "high"- }-- options.google = { thinkingConfig }- } else if (reasoningEffort) {- options.google = {- reasoningEffort: reasoningEffort as- | "low"- | "medium"- | "high",- }- }-- const effectiveMaxOutputTokens = maxOutputTokens || 8192-- options.google = {- ...options.google,- maxOutputTokens: effectiveMaxOutputTokens,- responseModalities: ["TEXT"],- }-- const options_obj: Record<string, any> = {}- const candidateCount = parseIntSafe(- process.env.GOOGLE_CANDIDATE_COUNT,- "GOOGLE_CANDIDATE_COUNT",- 1,- 8,- )- if (candidateCount) {- options_obj.candidateCount = candidateCount- }- const topK = parseIntSafe(- process.env.GOOGLE_TOP_K,- "GOOGLE_TOP_K",- 1,- 100,- )- if (topK) {- options_obj.topK = topK- }- if (process.env.GOOGLE_TOP_P) {- const topP = Number.parseFloat(process.env.GOOGLE_TOP_P)- if (Number.isNaN(topP) || topP < 0 || topP > 1) {- throw new Error(- `GOOGLE_TOP_P must be a number between 0 and 1, got: ${process.env.GOOGLE_TOP_P}`,- )- }- options_obj.topP = topP- }-- if (Object.keys(options_obj).length > 0) {- options.google = { ...options.google, ...options_obj }- }-- break- }-- case "azure": {- const reasoningEffort = process.env.AZURE_REASONING_EFFORT- const reasoningSummary = process.env.AZURE_REASONING_SUMMARY-- if (reasoningEffort || reasoningSummary) {- options.azure = {}- if (reasoningEffort) {- options.azure.reasoningEffort = reasoningEffort as- | "low"- | "medium"- | "high"- }- if (reasoningSummary) {- options.azure.reasoningSummary = reasoningSummary as- | "none"- | "brief"- | "detailed"- }- }- break- }-- case "bedrock": {- const budgetTokens = parseIntSafe(- process.env.BEDROCK_REASONING_BUDGET_TOKENS,- "BEDROCK_REASONING_BUDGET_TOKENS",- 1024,- 64000,- )- const reasoningEffort = process.env.BEDROCK_REASONING_EFFORT-- // Bedrock reasoning ONLY for Claude and Nova models- // Other models (MiniMax, etc.) don't support reasoningConfig- if (- modelId &&- (budgetTokens || reasoningEffort) &&- (modelId.includes("claude") ||- modelId.includes("anthropic") ||- modelId.includes("nova") ||- modelId.includes("amazon"))- ) {- const reasoningConfig: Record<string, any> = { type: "enabled" }-- // Claude models: use budgetTokens (1024-64000)- if (- budgetTokens &&- (modelId.includes("claude") ||- modelId.includes("anthropic"))- ) {- reasoningConfig.budgetTokens = budgetTokens- }- // Nova models: use maxReasoningEffort (low/medium/high)- else if (- reasoningEffort &&- (modelId.includes("nova") || modelId.includes("amazon"))- ) {- reasoningConfig.maxReasoningEffort = reasoningEffort as- | "low"- | "medium"- | "high"- }-- options.bedrock = { reasoningConfig }- }- break- }-- case "ollama": {- const enableThinking = process.env.OLLAMA_ENABLE_THINKING- // Ollama supports reasoning with think: true for models like qwen3- if (enableThinking === "true") {- options.ollama = { think: true }- }- break- }-- case "deepseek":- case "openrouter":- case "siliconflow": {- // These providers don't have reasoning configs in AI SDK yet- break- }-- default:- break- }-- return Object.keys(options).length > 0 ? options : undefined-}--// Map of provider to required environment variable-const PROVIDER_ENV_VARS: Record<ProviderName, string | null> = {- bedrock: null, // AWS SDK auto-uses IAM role on AWS, or env vars locally- openai: "OPENAI_API_KEY",- anthropic: "ANTHROPIC_API_KEY",- google: "GOOGLE_GENERATIVE_AI_API_KEY",- azure: "AZURE_API_KEY",- ollama: null, // No credentials needed for local Ollama- openrouter: "OPENROUTER_API_KEY",- deepseek: "DEEPSEEK_API_KEY",- siliconflow: "SILICONFLOW_API_KEY",-}--/**- * Auto-detect provider based on available API keys- * Returns the provider if exactly one is configured, otherwise null- */-function _detectProvider(): ProviderName | null {- const configuredProviders: ProviderName[] = []-- for (const [provider, envVar] of Object.entries(PROVIDER_ENV_VARS)) {- if (envVar === null) {- // Skip ollama - it doesn't require credentials- continue- }- if (process.env[envVar]) {- configuredProviders.push(provider as ProviderName)- }- }-- if (configuredProviders.length === 1) {- return configuredProviders[0]- }-- return null-}--/**- * Validate that required API keys are present for the selected provider- */-function _validateProviderCredentials(provider: ProviderName): void {- const requiredVar = PROVIDER_ENV_VARS[provider]- if (requiredVar && !process.env[requiredVar]) {- throw new Error(- `${requiredVar} environment variable is required for ${provider} provider. ` +- `Please set it in your .env.local file.`,- )- }-}--/**- * Get the AI model based on environment variables- *- * Environment variables:- * - AI_MODEL: The model ID (e.g. gpt-4, claude-3, gemini-pro)- * - AI_API_KEY: Universal API key (or specific provider key)- * - AI_BASE_URL: Universal API base URL (optional)- */-export function getAIModel(overrides?: ClientOverrides): ModelConfig {- // 1. Determine Model ID- const modelId = String(- overrides?.modelId || process.env.AI_MODEL || "",- ).trim()-- if (!modelId) {- throw new Error(- `AI_MODEL environment variable is required. Example: AI_MODEL=claude-3-5-sonnet-20240620`,- )- }-- // 2. Determine API Key- // Use the key supplied by the app Settings dialog as the authoritative source.- const apiKey = String(overrides?.apiKey || "").trim()-- if (!apiKey) {- throw new Error(- "API Key is missing. Please set it in the app Settings dialog.",- )- }-- // 3. Determine Base URL- const baseURL = String(- overrides?.baseUrl ||- process.env.AI_BASE_URL ||- process.env.OPENAI_BASE_URL ||- "",- ).trim()-- // 4. Initialize Universal Provider- const openAIConfig: { apiKey: string; baseURL?: string } = {- apiKey,- }- if (baseURL) {- openAIConfig.baseURL = baseURL- }- const universalOpenAI = createOpenAI(openAIConfig)-- const model = universalOpenAI.chat(modelId)- const providerOptions = {}- const headers: Record<string, string> | undefined = undefined-- console.log(- `[AI Provider] Initializing Universal Provider with model: ${modelId}${baseURL ? ` at ${baseURL}` : ""}`,- )-- return { model, providerOptions, headers, modelId }-}--/**- * Check if a model supports prompt caching.- * Currently only Claude models on Bedrock support prompt caching.- */-export function supportsPromptCaching(modelId: string): boolean {- // Bedrock prompt caching is supported for Claude models- return (- modelId.includes("claude") ||- modelId.includes("anthropic") ||- modelId.startsWith("us.anthropic") ||- modelId.startsWith("eu.anthropic")- )-}-diff --git a/src/workspaces/flow/next/lib/use-file-processor.tsx b/src/workspaces/flow/next/lib/use-file-processor.tsxindex c052c53..ddd106d 100644--- a/src/workspaces/flow/next/lib/use-file-processor.tsx+++ b/src/workspaces/flow/next/lib/use-file-processor.tsx@@ -21,7 +21,7 @@ import {isTarGzFile,MAX_EXTRACTED_CHARS,} from "@/workspaces/flow/next/lib/pdf-utils"-import { getAIConfig } from "@/lib/ai-client"+import { getAIConfig } from "@/ai/client"export interface FileData {text: stringdiff --git a/src/workspaces/ppt/PptWorkspaceShell.tsx b/src/workspaces/ppt/PptWorkspaceShell.tsxindex 2884871..1288a3d 100644--- a/src/workspaces/ppt/PptWorkspaceShell.tsx+++ b/src/workspaces/ppt/PptWorkspaceShell.tsx@@ -13,7 +13,7 @@ import {PPT_SLIDES_EDIT_SYSTEM_PROMPT,} from "@/lib/system-prompts";import { clearPersistedPptWorkspaceState } from "@/lib/ppt-persistence";-import type { ChatMessage } from "@/lib/ai-client";+import type { ChatMessage } from "@/ai/client";import { PptWorkspace } from "@/workspaces/ppt/workspace/PptWorkspace";import { ChatPanel as PptChatPanel } from "@/workspaces/ppt/chat/ChatPanel";import type { HistoryItem } from "@/workspaces/ppt/chat/history-dialog";diff --git a/src/workspaces/ppt/chat/ChatPanel.tsx b/src/workspaces/ppt/chat/ChatPanel.tsxindex b0edbdf..b7fa100 100644--- a/src/workspaces/ppt/chat/ChatPanel.tsx+++ b/src/workspaces/ppt/chat/ChatPanel.tsx@@ -1,7 +1,7 @@import React, { useState, useRef, useEffect } from 'react';import { PanelRightClose, PanelRightOpen } from 'lucide-react';import { cn } from '@/lib/utils';-import { streamChatMessage, generateChatMessage, generatePptProxyChatMessage, ChatMessage, getAIConfig } from '@/lib/ai-client';+import { streamChatMessage, generateChatMessage, generatePptProxyChatMessage, ChatMessage, getAIConfig } from '@/ai/client';import { DRAWIO_SYSTEM_PROMPT } from '@/lib/system-prompts';import { ButtonWithTooltip } from '@/workspaces/ppt/chat/components/button-with-tooltip';import { ChatInput } from '@/workspaces/ppt/chat/ChatInput';diff --git a/src/workspaces/ppt/workspace/PptWorkspace.tsx b/src/workspaces/ppt/workspace/PptWorkspace.tsxindex d9e3097..d2aefcd 100644--- a/src/workspaces/ppt/workspace/PptWorkspace.tsx+++ b/src/workspaces/ppt/workspace/PptWorkspace.tsx@@ -1,6 +1,6 @@import React, { useState, useEffect, useLayoutEffect, useRef } from 'react';import { motion } from "framer-motion";-import { generateImage, generateChatMessage } from '@/lib/ai-client';+import { generateImage, generateChatMessage } from '@/ai/client';import { pptService, PptPage, type PptElement, type PptTextBlock, type SlideEditRoutingItem, resolveTextBlockFontSize, PPT_REFERENCE_SLIDE_HEIGHT, PPT_REFERENCE_SLIDE_WIDTH, textBlocksToPptElements } from '@/lib/ppt-service';import { getTemplateGenerationPrompt } from '@/lib/ppt-prompts';import {diff --git a/tests/ai-config.test.ts b/tests/ai-config.test.tsnew file mode 100644index 0000000..6e41dd5--- /dev/null+++ b/tests/ai-config.test.ts@@ -0,0 +1,78 @@+import assert from "node:assert/strict";+import { test } from "node:test";+import {+ DEFAULT_BASE_URL,+ getImageChannel,+ getTextChannel,+ normalizeAIConfig,+} from "../src/ai/config";++test("fills both channels from an empty config", () => {+ const config = normalizeAIConfig({});+ assert.equal(config.textBaseUrl, DEFAULT_BASE_URL);+ assert.equal(config.imageBaseUrl, DEFAULT_BASE_URL);+ assert.equal(config.textApiKey, "");+ assert.equal(config.textModel, "");+});++test("tolerates null and non-object input", () => {+ assert.equal(normalizeAIConfig(null).textBaseUrl, DEFAULT_BASE_URL);+ assert.equal(normalizeAIConfig(undefined).imageBaseUrl, DEFAULT_BASE_URL);+ assert.equal(normalizeAIConfig("nonsense" as unknown).textModel, "");+});++test("migrates the legacy shared-credential config", () => {+ // Settings written before the OpenAI-only rewrite stored one shared key and+ // base URL, and named the models chatModel / imageModelLegacy.+ const config = normalizeAIConfig({+ apiKey: "sk-shared",+ baseUrl: "https://legacy.test/v1",+ chatModel: "gpt-4o",+ imageModelLegacy: "dall-e-3",+ });++ assert.equal(config.textApiKey, "sk-shared");+ assert.equal(config.imageApiKey, "sk-shared");+ assert.equal(config.textBaseUrl, "https://legacy.test/v1");+ assert.equal(config.imageBaseUrl, "https://legacy.test/v1");+ assert.equal(config.textModel, "gpt-4o");+ assert.equal(config.imageModel, "dall-e-3");+});++test("per-channel values win over the shared legacy ones", () => {+ const config = normalizeAIConfig({+ apiKey: "sk-shared",+ baseUrl: "https://legacy.test/v1",+ imageApiKey: "sk-image",+ imageBaseUrl: "https://images.test/v1",+ });++ assert.equal(config.textApiKey, "sk-shared");+ assert.equal(config.imageApiKey, "sk-image");+ assert.equal(config.imageBaseUrl, "https://images.test/v1");+});++test("drops the provider field the old config carried", () => {+ const config = normalizeAIConfig({ textProvider: "anthropic", imageProvider: "aliyun" });+ assert.equal("textProvider" in config, false);+ assert.equal("imageProvider" in config, false);+});++test("channels fall back to the default base URL when one is blank", () => {+ const config = normalizeAIConfig({ textBaseUrl: " ", imageBaseUrl: "" });+ assert.equal(getTextChannel(config).baseUrl, DEFAULT_BASE_URL);+ assert.equal(getImageChannel(config).baseUrl, DEFAULT_BASE_URL);+});++test("channels expose exactly the three fields a request needs", () => {+ const config = normalizeAIConfig({+ textApiKey: "sk-text",+ textBaseUrl: "https://text.test/v1",+ textModel: "gpt-4o-mini",+ });+ assert.deepEqual(getTextChannel(config), {+ apiKey: "sk-text",+ baseUrl: "https://text.test/v1",+ model: "gpt-4o-mini",+ });+});diff --git a/tests/ai-gateway-request-shape.test.ts b/tests/ai-gateway-request-shape.test.tsdeleted file mode 100644index 5d51c5d..0000000--- a/tests/ai-gateway-request-shape.test.ts+++ /dev/null@@ -1,131 +0,0 @@-import assert from "node:assert/strict";-import { generateImageThroughGateway, generateTextThroughGateway } from "../src/lib/ai/gateway";--type CapturedRequest = {- url: string;- init: RequestInit;- body: any;-};--const captured: CapturedRequest[] = [];-const originalFetch = globalThis.fetch;--function mockJsonResponse(payload: any, ok = true) {- return {- ok,- status: ok ? 200 : 400,- headers: new Headers({ "content-type": "application/json" }),- text: async () => JSON.stringify(payload),- arrayBuffer: async () => new ArrayBuffer(0),- } as Response;-}--globalThis.fetch = (async (url: any, init?: RequestInit) => {- const bodyText = typeof init?.body === "string" ? init.body : "";- const body = bodyText ? JSON.parse(bodyText) : null;- captured.push({ url: String(url), init: init || {}, body });-- if (String(url).includes(":generateContent")) {- return mockJsonResponse({ candidates: [{ content: { parts: [{ text: "ok" }] } }] });- }- if (String(url).endsWith("/messages")) {- return mockJsonResponse({ content: [{ type: "text", text: "ok" }] });- }- if (String(url).includes(":predict")) {- return mockJsonResponse({ predictions: [{ bytesBase64Encoded: "aGVsbG8=" }] });- }- if (String(url).includes("/services/aigc/multimodal-generation/generation")) {- return mockJsonResponse({- output: {- choices: [- {- message: {- content: [{ image: "data:image/png;base64,aGVsbG8=" }],- },- },- ],- },- });- }- return mockJsonResponse({ data: [{ b64_json: "aGVsbG8=" }] });-}) as typeof fetch;--function lastRequest() {- const request = captured[captured.length - 1];- assert.ok(request, "expected a captured request");- return request;-}--async function assertImageRequest(provider: string, expectedPath: string, referenceImageUrl?: string) {- const model =- provider === "openai"- ? "gpt-image-1"- : provider === "aliyun"- ? "wan2.2-image"- : provider === "tencent"- ? "hunyuan-image"- : provider === "bytedance"- ? "doubao-seedream-4-0"- : "imagen-4.0-generate-001";-- captured.length = 0;- await generateImageThroughGateway({- channel: {- provider,- apiKey: "test-key",- baseUrl: provider === "google" ? "https://generativelanguage.googleapis.com/v1beta/models" : "https://example.test/v1",- model,- },- prompt: "draw a square",- referenceImageUrl,- });-- const request = lastRequest();- assert.ok(request.url.includes(expectedPath), `${provider} should call ${expectedPath}, got ${request.url}`);- if (provider === "google") {- assert.equal((request.init.headers as Record<string, string>)["x-goog-api-key"], "test-key");- } else {- assert.equal((request.init.headers as Record<string, string>).Authorization, "Bearer test-key");- }-}--await assertImageRequest("openai", "/images/generations");-await assertImageRequest("openai", "/images/edits", "data:image/png;base64,aGVsbG8=");-await assertImageRequest("aliyun", "/services/aigc/multimodal-generation/generation");-await assertImageRequest("aliyun", "/services/aigc/multimodal-generation/generation", "data:image/png;base64,aGVsbG8=");-await assertImageRequest("tencent", "/images/generations");-await assertImageRequest("tencent", "/images/edits", "data:image/png;base64,aGVsbG8=");-await assertImageRequest("bytedance", "/images/generations");-await assertImageRequest("bytedance", "/images/generations", "data:image/png;base64,aGVsbG8=");-await assertImageRequest("google", ":predict");-await assertImageRequest("google", ":predict", "data:image/png;base64,aGVsbG8=");--captured.length = 0;-await generateTextThroughGateway({- channel: {- provider: "google",- apiKey: "google-key",- baseUrl: "https://generativelanguage.googleapis.com/v1beta",- model: "gemini-2.5-flash",- },- messages: [{ role: "user", content: "hello" }],-});-assert.ok(lastRequest().url.endsWith("/models/gemini-2.5-flash:generateContent"));-assert.equal((lastRequest().init.headers as Record<string, string>)["x-goog-api-key"], "google-key");--captured.length = 0;-await generateTextThroughGateway({- channel: {- provider: "anthropic",- apiKey: "anthropic-key",- baseUrl: "https://api.anthropic.com/v1",- model: "claude-sonnet-4-5",- },- messages: [{ role: "user", content: "hello" }],-});-assert.ok(lastRequest().url.endsWith("/messages"));-assert.equal((lastRequest().init.headers as Record<string, string>)["x-api-key"], "anthropic-key");--globalThis.fetch = originalFetch;--console.log("ai-gateway request-shape tests passed");diff --git a/tests/ai-provider-registry.test.ts b/tests/ai-provider-registry.test.tsdeleted file mode 100644index ce450d8..0000000--- a/tests/ai-provider-registry.test.ts+++ /dev/null@@ -1,61 +0,0 @@-import assert from "node:assert/strict";-import {- IMAGE_PROVIDER_OPTIONS,- TEXT_PROVIDER_OPTIONS,- getDefaultBaseUrl,- normalizeAIConfig,- resolveImageCapabilities,- resolveTextRoute,-} from "../src/lib/ai/provider-registry";--const imageProviderIds = IMAGE_PROVIDER_OPTIONS.map((option) => option.id);--assert.deepEqual(imageProviderIds, [- "openai",- "aliyun",- "tencent",- "bytedance",- "google",-]);--assert.equal(normalizeAIConfig({ imageProvider: "zhipu" }).imageProvider, "openai");-assert.equal(normalizeAIConfig({ imageProvider: "bfl" }).imageProvider, "openai");--assert.ok(TEXT_PROVIDER_OPTIONS.length > 0, "text providers should remain available");--assert.equal(resolveTextRoute({- provider: "google",- apiKey: "test",- baseUrl: getDefaultBaseUrl("google", "text"),- model: "gemini-2.5-flash",-}).protocol, "google-gemini");--assert.equal(resolveTextRoute({- provider: "anthropic",- apiKey: "test",- baseUrl: getDefaultBaseUrl("anthropic", "text"),- model: "claude-sonnet-4-5",-}).protocol, "anthropic-messages");--for (const provider of imageProviderIds) {- const capabilities = resolveImageCapabilities({- provider,- apiKey: "test",- baseUrl: "https://example.com",- model:- provider === "openai"- ? "gpt-image-1"- : provider === "aliyun"- ? "wan2.7-image"- : provider === "tencent"- ? "hunyuan-image"- : provider === "bytedance"- ? "doubao-seedream-4.0"- : "imagen-4.0-generate-001",- });-- assert.equal(capabilities.supportsGeneration, true, `${provider} should support generation`);- assert.equal(capabilities.supportsEdits, true, `${provider} should expose an edit branch`);-}--console.log("ai-provider-registry tests passed");diff --git a/tests/ai-request-shape.test.ts b/tests/ai-request-shape.test.tsnew file mode 100644index 0000000..e82f93d--- /dev/null+++ b/tests/ai-request-shape.test.ts@@ -0,0 +1,163 @@+import assert from "node:assert/strict";+import { afterEach, beforeEach, test } from "node:test";+import { generateImage } from "../src/ai/image";+import { generateText } from "../src/ai/text";+import type { AIChannel } from "../src/ai/types";++/**+ * These lock in the wire format: every model call must be one of the three+ * OpenAI routes, authenticated with a bearer token, against the configured+ * base URL. They are the guard against per-vendor branching creeping back in.+ */++type Captured = { url: string; init: RequestInit; body: any };++const channel: AIChannel = {+ apiKey: "test-key",+ baseUrl: "https://example.test/v1",+ model: "test-model",+};++const PNG_DATA_URL = "data:image/png;base64,aGVsbG8=";++let captured: Captured[] = [];+let respond: (url: string) => any;+const originalFetch = globalThis.fetch;++function jsonResponse(payload: any, ok = true) {+ return {+ ok,+ status: ok ? 200 : 400,+ headers: new Headers({ "content-type": "application/json" }),+ text: async () => JSON.stringify(payload),+ arrayBuffer: async () => new ArrayBuffer(0),+ } as Response;+}++beforeEach(() => {+ captured = [];+ respond = () => ({ data: [{ b64_json: "aGVsbG8=" }] });+ globalThis.fetch = (async (url: any, init?: RequestInit) => {+ const bodyText = typeof init?.body === "string" ? init.body : "";+ captured.push({ url: String(url), init: init || {}, body: bodyText ? JSON.parse(bodyText) : null });+ return jsonResponse(respond(String(url)));+ }) as typeof fetch;+});++afterEach(() => {+ globalThis.fetch = originalFetch;+});++const last = () => {+ const request = captured[captured.length - 1];+ assert.ok(request, "expected a captured request");+ return request;+};++const authHeader = (request: Captured) => (request.init.headers as Record<string, string>).Authorization;++test("text goes to /chat/completions with a bearer token", async () => {+ respond = () => ({ choices: [{ message: { content: "hi" } }] });++ const result = await generateText({ channel, messages: [{ role: "user", content: "hello" }] });++ assert.equal(result, "hi");+ assert.equal(last().url, "https://example.test/v1/chat/completions");+ assert.equal(authHeader(last()), "Bearer test-key");+ assert.equal(last().body.model, "test-model");+ assert.equal(last().body.stream, false);+});++test("text flattens a multimodal reply to its text parts", async () => {+ respond = () => ({+ choices: [{ message: { content: [{ type: "text", text: "one" }, { type: "text", text: "two" }] } }],+ });++ assert.equal(await generateText({ channel, messages: [{ role: "user", content: "hello" }] }), "one\ntwo");+});++test("a prompt with no reference image goes to /images/generations", async () => {+ const result = await generateImage({ channel, prompt: "draw a square" });++ assert.equal(result, "data:image/png;base64,aGVsbG8=");+ assert.equal(last().url, "https://example.test/v1/images/generations");+ assert.equal(authHeader(last()), "Bearer test-key");+});++test("a reference image switches to /images/edits as multipart", async () => {+ await generateImage({ channel, prompt: "make it blue", referenceImageUrl: PNG_DATA_URL });++ assert.equal(last().url, "https://example.test/v1/images/edits");+ assert.ok(last().init.body instanceof FormData, "edits must be sent as multipart");+ assert.equal(authHeader(last()), "Bearer test-key");+});++test("falls back to /chat/completions when the images route fails", async () => {+ respond = (url) => {+ if (url.includes("/images/")) throw new Error("no images route here");+ return { choices: [{ message: { content: [{ type: "image_url", image_url: { url: PNG_DATA_URL } }] } }] };+ };++ const result = await generateImage({ channel, prompt: "draw a square" });++ assert.equal(result, PNG_DATA_URL);+ assert.deepEqual(+ captured.map((request) => new URL(request.url).pathname),+ ["/v1/images/generations", "/v1/chat/completions"],+ );+});++test("the chat fallback also reads a markdown image link", async () => {+ respond = (url) => {+ if (url.includes("/images/")) throw new Error("no images route here");+ return { choices: [{ message: { content: `here you go ` } }] };+ };++ assert.equal(await generateImage({ channel, prompt: "draw a square" }), PNG_DATA_URL);+});++test("reference images ride along on the chat fallback", async () => {+ respond = (url) => {+ if (url.includes("/images/")) throw new Error("no images route here");+ return { choices: [{ message: { content: [{ type: "image_url", image_url: { url: PNG_DATA_URL } }] } }] };+ };++ await generateImage({+ channel,+ prompt: "combine these",+ referenceImageUrl: PNG_DATA_URL,+ additionalReferenceImageUrls: [PNG_DATA_URL],+ });++ const parts = last().body.messages[0].content;+ assert.equal(parts.filter((part: any) => part.type === "image_url").length, 2);+});++test("reports the first route's error when every route fails", async () => {+ respond = (url) => {+ throw new Error(url.includes("/images/") ? "images route said no" : "chat route said no");+ };++ await assert.rejects(generateImage({ channel, prompt: "draw a square" }), /images route said no/);+});++test("a trailing slash on the base URL does not double up", async () => {+ respond = () => ({ choices: [{ message: { content: "hi" } }] });++ await generateText({+ channel: { ...channel, baseUrl: "https://example.test/v1/" },+ messages: [{ role: "user", content: "hello" }],+ });++ assert.equal(last().url, "https://example.test/v1/chat/completions");+});++test("surfaces the provider's error message verbatim", async () => {+ globalThis.fetch = (async () =>+ jsonResponse({ error: { message: "model not found" } }, false)) as typeof fetch;++ await assert.rejects(+ generateText({ channel, messages: [{ role: "user", content: "hello" }] }),+ /model not found/,+ );+});