nolo-cli 0.1.22 → 0.1.26

package/ai/agent/web/useGuidedAgentCreation.ts

@@ -0,0 +1,217 @@
+import React from "react";
+import {
+  buildAgentFormDataFromGuidedDraft,
+  mergeGuidedAgentDraft,
+  validateGuidedAgentDraft,
+} from "ai/agent/guidedCreation/draft";
+import { DEFAULT_MODEL } from "ai/llm/providers";
+import type {
+  GuidedAgentAssistantMessage,
+  GuidedAgentCapabilityId,
+  GuidedAgentDraft,
+} from "ai/agent/guidedCreation/types";
+import { mapCapabilityIdsToToolIds } from "ai/agent/guidedCreation/capabilities";
+
+const EMPTY_DRAFT: GuidedAgentDraft = {
+  name: "",
+  introduction: "",
+  prompt: "",
+  promptSummary: "",
+  provider: DEFAULT_MODEL.provider,
+  model: DEFAULT_MODEL.name,
+  isPublic: false,
+  capabilityIds: [],
+  toolIds: [],
+  references: [],
+  tags: [],
+  unresolved: ["用途", "回答风格", "能力"],
+};
+
+const QUESTIONS = [
+  "你想让这个 AI 主要帮你完成什么？",
+  "它主要服务谁？希望用什么语气和风格回答？",
+  "它需要哪些能力：联网搜索、读写文档、处理表格、调用其他 AI、创建应用或代码？",
+  "有哪些资料、边界或禁区需要它遵守？",
+  "这个 AI 是只给自己用，还是公开给别人使用？",
+];
+
+const textIncludes = (text: string, words: string[]) =>
+  words.some((word) => text.includes(word));
+
+const inferCapabilityIds = (text: string): GuidedAgentCapabilityId[] => {
+  const lower = text.toLowerCase();
+  const ids: GuidedAgentCapabilityId[] = [];
+  if (textIncludes(lower, ["搜索", "联网", "网页", "资料", "最新", "research", "search"])) {
+    ids.push("webSearch");
+  }
+  if (textIncludes(lower, ["文档", "知识库", "页面", "资料", "doc", "docs", "knowledge"])) {
+    ids.push("docs");
+  }
+  if (textIncludes(lower, ["表格", "数据", "记录", "table", "sheet", "csv"])) {
+    ids.push("tables");
+  }
+  if (textIncludes(lower, ["多 agent", "多agent", "评审", "协作", "agent"])) {
+    ids.push("agents");
+  }
+  if (textIncludes(lower, ["应用", "代码", "app", "code", "开发", "生成页面"])) {
+    ids.push("apps");
+  }
+  return Array.from(new Set(ids));
+};
+
+const inferName = (text: string) => {
+  const cleaned = text
+    .replace(/[。！？!?,，；;：:]/g, " ")
+    .replace(/\bAI\b/gi, "")
+    .replace(/帮我创建|创建一个|创建|面向/g, "")
+    .replace(/\s+/g, " ")
+    .trim();
+  if (!cleaned) return "";
+  const purpose = cleaned.slice(0, 18);
+  return `${purpose.replace(/\s+/g, "")}助手`;
+};
+
+const buildPrompt = (messages: GuidedAgentAssistantMessage[], draft: GuidedAgentDraft) => {
+  const userNeeds = messages
+    .filter((message) => message.role === "user")
+    .map((message) => `- ${message.content}`)
+    .join("\n");
+  return [
+    `你是${draft.name || "用户的专属 AI"}。`,
+    "你需要先理解用户目标，再给出可执行、清晰、有边界的帮助。",
+    "根据以下需求工作：",
+    userNeeds || "- 尚未补充具体需求",
+    "如果信息不足，先提出必要澄清问题；不要编造用户没有确认的资料来源。",
+  ].join("\n");
+};
+
+const buildNextQuestion = (draft: GuidedAgentDraft) => {
+  if (!draft.name || !draft.prompt) return QUESTIONS[1];
+  if (draft.capabilityIds.length === 0) return QUESTIONS[2];
+  if (!draft.introduction) return QUESTIONS[3];
+  if (draft.unresolved.includes("公开状态")) return QUESTIONS[4];
+  return "配置草稿已经可创建。你还想补充哪些资料、语气或使用边界？";
+};
+
+const createAssistantDraft = (
+  messages: GuidedAgentAssistantMessage[],
+  current: GuidedAgentDraft
+) => {
+  const userText = messages
+    .filter((message) => message.role === "user")
+    .map((message) => message.content)
+    .join("\n");
+  const capabilityIds = Array.from(
+    new Set([...current.capabilityIds, ...inferCapabilityIds(userText)])
+  );
+  const name = current.name || inferName(userText);
+  const isPublic =
+    current.isPublic ||
+    textIncludes(userText.toLowerCase(), ["公开", "分享", "public", "给别人"]);
+  const unresolved = [
+    !name ? "name" : "",
+    !userText.trim() ? "prompt" : "",
+    capabilityIds.length === 0 ? "能力" : "",
+    current.introduction || userText.length > 24 ? "" : "introduction",
+    textIncludes(userText.toLowerCase(), ["公开", "私有", "private", "public"]) ? "" : "公开状态",
+  ].filter(Boolean);
+  const next: GuidedAgentDraft = mergeGuidedAgentDraft(current, {
+    name,
+    introduction:
+      current.introduction ||
+      (userText.trim()
+        ? `面向你的需求，帮助整理、判断并执行：${userText.trim().slice(0, 70)}`
+        : ""),
+    promptSummary: userText.trim()
+      ? `围绕“${userText.trim().slice(0, 46)}”提供结构化帮助，必要时先追问。`
+      : "",
+    capabilityIds,
+    toolIds: mapCapabilityIdsToToolIds(capabilityIds),
+    isPublic,
+    tags: Array.from(new Set([...current.tags, ...capabilityIds])),
+    unresolved,
+  });
+  return mergeGuidedAgentDraft(next, {
+    prompt: buildPrompt(messages, next),
+  });
+};
+
+const createId = () => `${Date.now()}-${Math.random().toString(16).slice(2)}`;
+
+export const useGuidedAgentCreation = (initialDraft?: Partial<GuidedAgentDraft> | null) => {
+  const [draft, setDraft] = React.useState<GuidedAgentDraft>(() =>
+    initialDraft ? mergeGuidedAgentDraft(EMPTY_DRAFT, initialDraft) : EMPTY_DRAFT
+  );
+  const [messages, setMessages] = React.useState<GuidedAgentAssistantMessage[]>([
+    {
+      id: "assistant-initial",
+      role: "assistant",
+      content: initialDraft?.name
+        ? "我已经带入了当前对话整理出的草稿。你可以继续补充要求，或直接预览并创建。"
+        : QUESTIONS[0],
+    },
+  ]);
+  const [input, setInput] = React.useState("");
+  const [isThinking, setIsThinking] = React.useState(false);
+  const [error, setError] = React.useState<string | null>(null);
+
+  const submit = React.useCallback(async () => {
+    const content = input.trim();
+    if (!content || isThinking) return;
+
+    setInput("");
+    setError(null);
+    setIsThinking(true);
+
+    const userMessage: GuidedAgentAssistantMessage = {
+      id: createId(),
+      role: "user",
+      content,
+    };
+
+    try {
+      const nextMessages = [...messages, userMessage];
+      const nextDraft = createAssistantDraft(nextMessages, draft);
+      const assistantMessage: GuidedAgentAssistantMessage = {
+        id: createId(),
+        role: "assistant",
+        content: buildNextQuestion(nextDraft),
+      };
+      setMessages([...nextMessages, assistantMessage]);
+      setDraft(nextDraft);
+    } catch {
+      setError("暂时没能整理这段需求，请换一种说法再试。");
+    } finally {
+      setIsThinking(false);
+    }
+  }, [draft, input, isThinking, messages]);
+
+  const toggleReference = React.useCallback((dbKey: string, selected: boolean) => {
+    setDraft((current) =>
+      mergeGuidedAgentDraft(current, {
+        references: current.references.map((reference) =>
+          reference.dbKey === dbKey ? { ...reference, selected } : reference
+        ),
+      })
+    );
+  }, []);
+
+  const validation = validateGuidedAgentDraft(draft);
+  const formInitialValues = React.useMemo(
+    () => buildAgentFormDataFromGuidedDraft(draft),
+    [draft]
+  );
+
+  return {
+    draft,
+    messages,
+    input,
+    setInput,
+    submit,
+    isThinking,
+    error,
+    validation,
+    formInitialValues,
+    toggleReference,
+  };
+};

package/ai/ai.locale.ts

@@ -27,6 +27,9 @@ export default {
       textOnly: "Text-only",
       details: "Details",
       price: "Price",
+      modelCost: "Model cost",
+      inputCostPerMillionTokens: "Input",
+      outputCostPerMillionTokens: "Output",
       referencePricePerTurn: "Per-turn reference price",
       defaultImageProfileEstimate: "Default profile estimate",
       perMillionTokens: "per 1M tokens",
@@ -251,7 +254,7 @@ export default {
           description: "Create a new Agent (intelligent assistant/app) based on the given configuration and return the complete Agent information.",
           params: {
             name: "The name of the Agent, e.g., 'Meeting Minutes Assistant'.",
-            model: "The model name to use. Logic: Default 'xiaomi/mimo-v2.5-pro'. Complex tasks: 'openai/gpt-5.2'. Cost-saving: 'openai/gpt-5.1'. Aesthetic/Creative: 'gemini-3-pro-preview'. Long Context/Fast: 'x-ai/grok-4.1-fast'. Image Generation: 'gemini-3-pro-image-preview' (Best/Expensive) or 'gemini-2.5-flash-image' (Ordinary).",
+            model: "The model name to use. Logic: Default 'xiaomi/mimo-v2.5-pro'. Complex tasks: 'openai/gpt-5.2'. Cost-saving: 'openai/gpt-5.1'. Aesthetic/Creative: 'gemini-3-pro-preview'. Reliable low-cost summary/reasoning: 'deepseek-v4-pro'. Image Generation: 'gemini-3-pro-image-preview' (Best/Expensive) or 'gemini-2.5-flash-image' (Ordinary).",
             provider: "The model provider identifier, must match the selected model, e.g., 'openrouter', 'deepseek', 'fireworks', 'mistral', etc.",
             prompt: "System prompt/instruction describing the Agent's role, tone, and usage.",
             introduction: "Introduction displayed in the app market or list, explanatory text for end users.",
@@ -318,6 +321,9 @@ export default {
       textOnly: "纯文本",
       details: "详细信息",
       price: "价格",
+      modelCost: "模型成本",
+      inputCostPerMillionTokens: "输入",
+      outputCostPerMillionTokens: "输出",
       referencePricePerTurn: "按次参考价",
       defaultImageProfileEstimate: "默认档参考价",
       perMillionTokens: "每百万Token",
@@ -539,7 +545,7 @@ export default {
           description: "根据给定配置创建一个新的 Agent（智能体 / 应用），并返回完整的 Agent 信息。",
           params: {
             name: "Agent 的名称，例如：'会议纪要助手'。",
-            model: "默认使用 'xiaomi/mimo-v2.5-pro'。复杂任务: 'openai/gpt-5.2'。省成本但复杂: 'openai/gpt-5.1'。追求美观/创意: 'gemini-3-pro-preview'。长上下文/快速: 'x-ai/grok-4.1-fast'。生成图片: 贵/高质量选 'gemini-3-pro-image-preview'，普通选 'gemini-2.5-flash-image'。",
+            model: "默认使用 'xiaomi/mimo-v2.5-pro'。复杂任务: 'openai/gpt-5.2'。省成本但复杂: 'openai/gpt-5.1'。追求美观/创意: 'gemini-3-pro-preview'。低成本摘要/推理: 'deepseek-v4-pro'。生成图片: 贵/高质量选 'gemini-3-pro-image-preview'，普通选 'gemini-2.5-flash-image'。",
             provider: "模型提供方标识，必须与所选模型相匹配，例如 'openrouter'、'deepseek'、'fireworks'、'mistral' 等。",
             prompt: "系统提示词 / 指令，描述这个 Agent 的角色、语气和使用方式。",
             introduction: "在应用市场或列表中展示的简介，面向终端用户的说明文案。",
@@ -606,6 +612,9 @@ export default {
       textOnly: "純文字",
       details: "詳細資訊",
       price: "價格",
+      modelCost: "模型成本",
+      inputCostPerMillionTokens: "輸入",
+      outputCostPerMillionTokens: "輸出",
       referencePricePerTurn: "按次參考價",
       defaultImageProfileEstimate: "預設檔參考價",
       perMillionTokens: "每百萬Token",
@@ -852,6 +861,9 @@ export default {
       textOnly: "テキストのみ",
       details: "詳細",
       price: "価格",
+      modelCost: "モデルコスト",
+      inputCostPerMillionTokens: "入力",
+      outputCostPerMillionTokens: "出力",
       referencePricePerTurn: "回ごとの参考価格",
       defaultImageProfileEstimate: "デフォルト構成の参考価格",
       perMillionTokens: "100万トークンあたり",

package/ai/chat/fetchUtils.native.ts

@@ -42,7 +42,9 @@ const buildProxyPayload = (
         ...bodyData,
         url: api,
         provider,
+        agentKey: agentConfig.dbKey,
         ...(apiSource ? { apiSource } : {}),
+        ...((agentConfig as any).apiKeyHeader ? { apiKeyHeader: (agentConfig as any).apiKeyHeader } : {}),
         KEY: apiKey,
     };
 };

package/ai/chat/fetchUtils.ts

@@ -40,7 +40,9 @@ const buildProxyPayload = (
     ...bodyData,
     url: api,
     provider,
+    agentKey: agentConfig.dbKey,
     ...(apiSource ? { apiSource } : {}),
+    ...((agentConfig as any).apiKeyHeader ? { apiKeyHeader: (agentConfig as any).apiKeyHeader } : {}),
     KEY: apiKey,
   };
 };

package/ai/chat/sendOpenAICompletionsRequest.ts

@@ -160,6 +160,23 @@ export type CompletionMeta = {
   usage?: any;
 };
 
+const STREAM_READ_TIMEOUT_MS = 45_000;
+
+const logQuickChatPerfStage = (
+  startedAt: number | undefined,
+  stage: string,
+  details?: Record<string, unknown>
+) => {
+  if (!startedAt) return;
+  const now = typeof performance !== "undefined" ? performance.now() : Date.now();
+  console.info("[QuickChatPerf]", {
+    stage,
+    elapsedMs: now - startedAt,
+    ...(typeof performance !== "undefined" ? { atMs: now } : {}),
+    ...(details ?? {}),
+  });
+};
+
 /**
  * 初始化流式状态
  */
@@ -398,6 +415,28 @@ async function processAccumulatedToolCalls(
   };
 }
 
+async function readStreamChunkWithTimeout(
+  reader: ReadableStreamDefaultReader<Uint8Array>
+) {
+  let timeoutId: ReturnType<typeof setTimeout> | undefined;
+  try {
+    return await Promise.race([
+      reader.read(),
+      new Promise<ReadableStreamReadResult<Uint8Array>>((_, reject) => {
+        timeoutId = setTimeout(() => {
+          reject(
+            new Error(
+              `模型响应流 ${Math.round(STREAM_READ_TIMEOUT_MS / 1000)} 秒内没有返回新内容`
+            )
+          );
+        }, STREAM_READ_TIMEOUT_MS);
+      }),
+    ]);
+  } finally {
+    if (timeoutId) clearTimeout(timeoutId);
+  }
+}
+
 /**
  * 处理「流结束」场景（done === true）
  */
@@ -455,6 +494,7 @@ export const sendOpenAICompletionsRequest = async ({
   parentMessageId,
   messageMetadata,
   disableToolsForThisRequest = false,
+  quickChatPerfStartedAt,
 }: {
   bodyData: any;
   agentConfig: any;
@@ -463,6 +503,7 @@ export const sendOpenAICompletionsRequest = async ({
   parentMessageId?: string;
   messageMetadata?: Partial<Message>;
   disableToolsForThisRequest?: boolean;
+  quickChatPerfStartedAt?: number;
 }): Promise<CompletionMeta> => {
   const { dispatch, getState, signal: thunkSignal } = thunkApi;
 
@@ -549,6 +590,10 @@ export const sendOpenAICompletionsRequest = async ({
 
     const api = getApiEndpoint(agentConfig);
     const token = selectCurrentToken(getState() as RootState);
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-completions-fetch-starting", {
+      api,
+      dialogKey,
+    });
     const response = await performFetchRequest({
       agentConfig,
       api,
@@ -557,6 +602,11 @@ export const sendOpenAICompletionsRequest = async ({
       signal,
       token,
     });
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-completions-fetch-response", {
+      ok: response.ok,
+      status: response.status,
+      dialogKey,
+    });
 
     if (!response.ok) {
       const errorMessage = await parseApiError(response);
@@ -578,9 +628,12 @@ export const sendOpenAICompletionsRequest = async ({
     }
 
     const decoder = new TextDecoder();
+    let loggedFirstStreamChunk = false;
+    let loggedFirstParsedEvent = false;
+    let loggedFirstVisibleDelta = false;
 
     while (true) {
-      const { done, value } = await reader.read();
+      const { done, value } = await readStreamChunkWithTimeout(reader);
 
       if (done) {
         const completion = await handleStreamCompletion(
@@ -602,8 +655,25 @@ export const sendOpenAICompletionsRequest = async ({
         break;
       }
 
+      if (!loggedFirstStreamChunk) {
+        loggedFirstStreamChunk = true;
+        logQuickChatPerfStage(
+          quickChatPerfStartedAt,
+          "openai-completions-first-stream-chunk",
+          { dialogKey, byteLength: value.byteLength }
+        );
+      }
+
       const chunk = decoder.decode(value, { stream: true });
       const parsedResults = parseSSE(chunk);
+      if (parsedResults.length > 0 && !loggedFirstParsedEvent) {
+        loggedFirstParsedEvent = true;
+        logQuickChatPerfStage(
+          quickChatPerfStartedAt,
+          "openai-completions-first-sse-event",
+          { dialogKey, eventCount: parsedResults.length }
+        );
+      }
 
       for (const parsedData of parsedResults) {
         const dataList = Array.isArray(parsedData) ? parsedData : [parsedData];
@@ -641,6 +711,14 @@ export const sendOpenAICompletionsRequest = async ({
             delta
           );
           streamState = updatedState;
+          if (hasNewVisibleContent && !loggedFirstVisibleDelta) {
+            loggedFirstVisibleDelta = true;
+            logQuickChatPerfStage(
+              quickChatPerfStartedAt,
+              "openai-completions-first-visible-delta",
+              { dialogKey }
+            );
+          }
 
           emitStreamingUpdate(hasNewVisibleContent, streamState, {
             dispatch,
@@ -712,6 +790,9 @@ export const sendOpenAICompletionsRequest = async ({
     };
     streamState = await finalizeStream(streamState, finalizeCtx);
   } finally {
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-completions-stream-finished", {
+      dialogKey,
+    });
     dispatch(removeActiveController({ messageId, dialogKey }));
     try {
       await reader?.cancel();

package/ai/chat/sendOpenAIResponseRequest.ts

@@ -42,6 +42,21 @@ const shouldEnableBuiltInImageGeneration = (agentConfig: any): boolean =>
   !getModelInfo(String(agentConfig?.model || ""))?.hasImageOutput &&
   !!agentConfig?.imageConfig?.enabled;
 
+const logQuickChatPerfStage = (
+  startedAt: number | undefined,
+  stage: string,
+  details?: Record<string, unknown>
+) => {
+  if (!startedAt) return;
+  const now = typeof performance !== "undefined" ? performance.now() : Date.now();
+  console.info("[QuickChatPerf]", {
+    stage,
+    elapsedMs: now - startedAt,
+    ...(typeof performance !== "undefined" ? { atMs: now } : {}),
+    ...(details ?? {}),
+  });
+};
+
 type StreamState = {
   content: string;
   contentBuffer: Array<
@@ -140,6 +155,7 @@ export const sendOpenAIResponseRequest = async ({
   dialogKey,
   parentMessageId,
   messageMetadata,
+  quickChatPerfStartedAt,
 }: {
   bodyData: any;
   agentConfig: any;
@@ -147,6 +163,7 @@ export const sendOpenAIResponseRequest = async ({
   dialogKey: string;
   parentMessageId?: string;
   messageMetadata?: Partial<Message>;
+  quickChatPerfStartedAt?: number;
 }): Promise<CompletionMeta> => {
   const { dispatch, getState, signal: thunkSignal } = thunkApi;
   const dialogId = extractCustomId(dialogKey);
@@ -354,6 +371,10 @@ export const sendOpenAIResponseRequest = async ({
 
     const api = getApiEndpoint(agentConfig);
     const token = selectCurrentToken(getState() as RootState);
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-response-fetch-starting", {
+      api,
+      dialogKey,
+    });
     const response = await performFetchRequest({
       agentConfig,
       api,
@@ -362,6 +383,11 @@ export const sendOpenAIResponseRequest = async ({
       signal,
       token,
     });
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-response-fetch-response", {
+      ok: response.ok,
+      status: response.status,
+      dialogKey,
+    });
 
     if (!response.ok) {
       const errorMessage = await parseApiError(response);
@@ -379,13 +405,33 @@ export const sendOpenAIResponseRequest = async ({
     const parseSSE = createSSEParser();
     const decoder = new TextDecoder();
     let finishReason: string | null = null;
+    let loggedFirstStreamChunk = false;
+    let loggedFirstParsedEvent = false;
+    let loggedFirstVisibleDelta = false;
 
     while (true) {
       const { done, value } = await reader.read();
       if (done) break;
 
+      if (!loggedFirstStreamChunk) {
+        loggedFirstStreamChunk = true;
+        logQuickChatPerfStage(
+          quickChatPerfStartedAt,
+          "openai-response-first-stream-chunk",
+          { dialogKey, byteLength: value.byteLength }
+        );
+      }
+
       const chunk = decoder.decode(value, { stream: true });
       const events = parseSSE(chunk);
+      if (events.length > 0 && !loggedFirstParsedEvent) {
+        loggedFirstParsedEvent = true;
+        logQuickChatPerfStage(
+          quickChatPerfStartedAt,
+          "openai-response-first-sse-event",
+          { dialogKey, eventCount: events.length }
+        );
+      }
       const eventList = Array.isArray(events) ? events : [events];
 
       for (const event of eventList) {
@@ -404,6 +450,14 @@ export const sendOpenAIResponseRequest = async ({
             if (event.delta) {
               state.content += event.delta;
               state.contentBuffer = seg(state.content);
+              if (!loggedFirstVisibleDelta) {
+                loggedFirstVisibleDelta = true;
+                logQuickChatPerfStage(
+                  quickChatPerfStartedAt,
+                  "openai-response-first-visible-delta",
+                  { dialogKey }
+                );
+              }
               flush();
             }
             break;
@@ -427,6 +481,14 @@ export const sendOpenAIResponseRequest = async ({
               if (itemText) {
                 state.content = itemText;
                 state.contentBuffer = seg(state.content);
+                if (!loggedFirstVisibleDelta) {
+                  loggedFirstVisibleDelta = true;
+                  logQuickChatPerfStage(
+                    quickChatPerfStartedAt,
+                    "openai-response-first-visible-delta",
+                    { dialogKey }
+                  );
+                }
                 flush();
               }
             }
@@ -506,6 +568,9 @@ export const sendOpenAIResponseRequest = async ({
     await finalize();
     return buildMeta(false, false, "error");
   } finally {
+    logQuickChatPerfStage(quickChatPerfStartedAt, "openai-response-stream-finished", {
+      dialogKey,
+    });
     dispatch(removeActiveController({ messageId, dialogKey }));
     await safeCancel(reader);
   }

package/ai/llm/getPricing.ts

@@ -209,6 +209,20 @@ export const formatPriceAmount = (amount: number): string => {
   return amount.toFixed(6);
 };
 
+export const formatModelCostPerMillion = (
+  value?: number,
+  creditsUnit = "积分"
+): string => {
+  if (typeof value !== "number" || !Number.isFinite(value)) return "未知";
+  if (value === 0) return "免费";
+  const decimals = Math.abs(value) < 1 ? 3 : 2;
+  const amount = value
+    .toFixed(decimals)
+    .replace(/\.0+$/, "")
+    .replace(/(\.\d*?[1-9])0+$/, "$1");
+  return `${amount} ${creditsUnit}`;
+};
+
 export const formatCompactTurnPrice = (
   amount: number
 ): CompactTurnPriceDisplay => {

package/ai/llm/kimi.ts

@@ -2,11 +2,14 @@ export const FIREWORKS_KIMI_LATEST_MODEL = "accounts/fireworks/models/kimi-lates
 export const FIREWORKS_KIMI_CURRENT_MODEL = "accounts/fireworks/models/kimi-k2p6";
 export const DEEPINFRA_KIMI_FALLBACK_MODEL = "moonshotai/Kimi-K2.6";
 export const OPENROUTER_KIMI_FALLBACK_MODEL = "moonshotai/kimi-k2.6";
-export const KIMI_PLATFORM_FALLBACK_STATUSES = [402, 429, 500, 502, 503, 504];
+export const KIMI_PLATFORM_FALLBACK_STATUSES = [401, 402, 429, 500, 502, 503, 504];
 
 export const isFireworksKimiModel = (model?: string | null): boolean =>
   model === FIREWORKS_KIMI_LATEST_MODEL || model === FIREWORKS_KIMI_CURRENT_MODEL;
 
+export const isDeepInfraKimiModel = (model?: string | null): boolean =>
+  model === DEEPINFRA_KIMI_FALLBACK_MODEL;
+
 export const resolveFireworksKimiModel = (model?: string | null): string => {
   if (model === FIREWORKS_KIMI_LATEST_MODEL) {
     return FIREWORKS_KIMI_CURRENT_MODEL;

package/ai/llm/modelAvatar.ts

@@ -393,6 +393,7 @@ export async function getModelAvatarComponent(
   if (cliProviderKey === "copilot") return get("copilot");
   if (cliProviderKey === "codex") return get("openai");
   if (cliProviderKey === "claude") return get("claude");
+  if (cliProviderKey === "agy") return get("gemini");
 
   // ── 按 model 名称匹配（优先，处理 openrouter/deepinfra 等聚合场景） ──
   if (m.includes("claude")) return get("claude");

package/ai/llm/providers.ts

@@ -267,7 +267,7 @@ export function getApiEndpoint(agent: Agent): string {
   }
 
   /* custom provider / apiSource=custom 但未给 URL 的兜底
-   * 注意：apiSource="custom" 的 agent（如 kimi-k2.5）不需要 provider 字段，
+   * 注意：apiSource="custom" 的 agent 不需要 provider 字段，
    * 使用 customProviderUrl + 自己的 apiKey，上面已经 return 了。
    * 只有 customProviderUrl 为空时才会走到这里报错。
    */
@@ -276,6 +276,9 @@ export function getApiEndpoint(agent: Agent): string {
     effectiveProvider.toLowerCase() === "custom" ||
     (agent as any).apiSource === "custom"
   ) {
+    if (agent.useServerProxy) {
+      return "";
+    }
     throw new Error(
       "Custom provider URL is required when apiSource is 'custom'."
     );

package/ai/llm/reasoningModels.ts

@@ -4,6 +4,7 @@ const REASONING_MODEL_NAMES = new Set([
   "deepseek-reasoner",
   "gemini-2.5-pro",
   "gemini-2.5-flash",
+  "gemini-3.5-flash",
   "gemini-3-flash-preview",
   "gemini-3-pro-preview",
   "gemini-3.1-pro-preview",