nolo-cli 0.1.23 → 0.1.29

package/integrations/google/models.ts

@@ -0,0 +1,298 @@
+import type { Model } from "../../ai/llm/types";
+import type { ImageSizeKey } from "../../ai/llm/imagePricing";
+
+const GOOGLE_IMAGE_ASPECT_RATIOS: NonNullable<Model["supportedAspectRatios"]> = [
+  "1:1",
+  "2:3",
+  "3:2",
+  "3:4",
+  "4:3",
+  "4:5",
+  "5:4",
+  "9:16",
+  "16:9",
+  "21:9",
+];
+
+const GOOGLE_IMAGE_SIZES: ImageSizeKey[] = ["1K", "2K", "4K"];
+
+const createGoogleImageModel = ({
+  name,
+  displayName,
+  inputPrice,
+  outputPrice,
+  imageTokenPricePerMillion,
+  imageOutputTokenEstimateBySize,
+  maxOutputTokens,
+  contextWindow,
+}: {
+  name: string;
+  displayName: string;
+  inputPrice: number;
+  outputPrice: number;
+  imageTokenPricePerMillion: number;
+  imageOutputTokenEstimateBySize: NonNullable<Model["imageOutputTokenEstimateBySize"]>;
+  maxOutputTokens: number;
+  contextWindow: number;
+}): Model => ({
+  name,
+  displayName,
+  provider: "google",
+  hasVision: true,
+  hasImageOutput: true,
+  supportsImageOutput: true,
+  supportsImageConfig: true,
+  requiresImageModalities: true,
+  defaultModalities: ["image", "text"],
+  supportedAspectRatios: GOOGLE_IMAGE_ASPECT_RATIOS,
+  supportedImageSizes: GOOGLE_IMAGE_SIZES,
+  price: {
+    input: inputPrice,
+    output: outputPrice,
+  },
+  imageTokenPricePerMillion,
+  imageOutputTokenEstimateBySize,
+  maxOutputTokens,
+  contextWindow,
+  supportsTool: false,
+});
+
+export const googleModels: Model[] = [
+  {
+    name: "gemini-2.5-pro",
+    displayName: "Gemini 2.5 Pro",
+    provider: "google",
+    description: "The full version of Gemini 2.5 Pro with all features.",
+    hasVision: true,
+    hasAudio: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 65536,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 1.25 * 8,
+      output: 10 * 8,
+      cachingWrite: 0.2 * 8,
+      cachingRead: 0.2 * 8,
+    },
+    pricingStrategy: {
+      type: "tiered_context",
+      tiers: [
+        {
+          minContext: 200001,
+          price: {
+            input: 2.5 * 8,
+            output: 15 * 8,
+            cachingWrite: 0.25 * 8,
+            cachingRead: 0.25 * 8,
+          },
+        },
+      ],
+    },
+  },
+  {
+    name: "gemini-2.5-flash",
+    displayName: "Gemini 2.5 Flash",
+    provider: "google",
+    description: "The full version of Gemini 2.5 Flash with all features.",
+    hasVision: true,
+    hasAudio: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 8192,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 0.3 * 8,
+      output: 2.5 * 8,
+      cachingWrite: 0.2 * 8,
+      cachingRead: 0.2 * 8,
+    },
+  },
+  {
+    name: "gemini-2.5-flash-lite",
+    displayName: "Gemini 2.5 Flash Lite",
+    provider: "google",
+    description:
+      "Our smallest and most cost-effective model, built for at scale usage.",
+    hasVision: false,
+    hasAudio: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 4096,
+    supportsTool: true,
+    price: {
+      input: 0.1 * 8,
+      output: 0.4 * 8,
+      cachingWrite: 0.2 * 8,
+      cachingRead: 0.2 * 8,
+    },
+  },
+  {
+    name: "gemini-3.5-flash",
+    displayName: "Gemini 3.5 Flash",
+    provider: "google",
+    description:
+      "Gemini 3.5 Flash GA model for fast frontier agentic, coding, and multimodal tasks.",
+    hasVision: true,
+    hasAudio: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 65536,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 1.5 * 8,
+      output: 9 * 8,
+      cachingWrite: 0.15 * 8,
+      cachingRead: 0.15 * 8,
+    },
+    serviceTierPriceMultipliers: {
+      batch: { inputOutput: 0.5, cache: 0.5 },
+      flex: { inputOutput: 0.5, cache: 0.08 / 0.15 },
+      priority: { inputOutput: 1.8, cache: 1.8 },
+    },
+  },
+  {
+    name: "gemini-3.1-flash-lite",
+    displayName: "Gemini 3.1 Flash-Lite",
+    provider: "google",
+    hasVision: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 65500,
+    supportsTool: true,
+    price: {
+      input: 0.25 * 8,
+      output: 1.5 * 8,
+      cachingWrite: 0.025 * 8,
+      cachingRead: 0.025 * 8,
+    },
+    serviceTierPriceMultipliers: {
+      batch: { inputOutput: 0.5, cache: 0.5 },
+      flex: { inputOutput: 0.5, cache: 0.5 },
+      priority: { inputOutput: 1.8, cache: 1.8 },
+    },
+  },
+  {
+    name: "gemini-3-flash-preview",
+    displayName: "Gemini 3 Flash Preview",
+    provider: "google",
+    hasVision: true,
+    contextWindow: 1048576,
+    maxOutputTokens: 65500,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 0.5 * 8,
+      output: 3 * 8,
+      cachingWrite: 0.05 * 8,
+      cachingRead: 0.05 * 8,
+    },
+    serviceTierPriceMultipliers: {
+      priority: { inputOutput: 1.8, cache: 1.8 },
+    },
+  },
+  {
+    name: "gemini-3-pro-preview",
+    displayName: "Gemini 3 Pro (Preview)",
+    provider: "google",
+    description:
+      "Google's most advanced model with enhanced reasoning, vision, and audio.",
+    hasVision: true,
+    hasAudio: true,
+    contextWindow: 2097152,
+    maxOutputTokens: 65536,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 2 * 8,
+      output: 12 * 8,
+      cachingWrite: 0.2 * 8,
+      cachingRead: 0.2 * 8,
+    },
+    pricingStrategy: {
+      type: "tiered_context",
+      tiers: [
+        {
+          minContext: 200001,
+          price: {
+            input: 4 * 8,
+            output: 18 * 8,
+            cachingWrite: 0.4 * 8,
+            cachingRead: 0.4 * 8,
+          },
+        },
+      ],
+    },
+  },
+  {
+    name: "gemini-3.1-pro-preview",
+    displayName: "Gemini 3.1 Pro Preview",
+    provider: "google",
+    hasVision: true,
+    contextWindow: 1050000,
+    maxOutputTokens: 65500,
+    supportsTool: true,
+    supportsReasoningEffort: true,
+    price: {
+      input: 2 * 8,
+      output: 12 * 8,
+      cachingRead: 0.2 * 8,
+      cachingWrite: 0.2 * 8,
+    },
+    serviceTierPriceMultipliers: {
+      priority: { inputOutput: 1.8, cache: 1.8 },
+    },
+    pricingStrategy: {
+      type: "tiered_context",
+      tiers: [
+        {
+          minContext: 200001,
+          price: {
+            input: 4 * 8,
+            output: 18 * 8,
+            cachingRead: 0.4 * 8,
+            cachingWrite: 0.4 * 8,
+          },
+        },
+      ],
+    },
+  },
+  createGoogleImageModel({
+    name: "gemini-3-pro-image-preview",
+    displayName: "Nano Banana Pro (Gemini 3 Pro Image Preview)",
+    inputPrice: 2 * 8,
+    outputPrice: 12 * 8,
+    imageTokenPricePerMillion: 120 * 8,
+    imageOutputTokenEstimateBySize: {
+      "1K": 1120,
+      "2K": 1120,
+      "4K": 2000,
+    },
+    maxOutputTokens: 8192,
+    contextWindow: 65536,
+  }),
+  createGoogleImageModel({
+    name: "gemini-3.1-flash-image-preview",
+    displayName: "Nano Banana 2 (Gemini 3.1 Flash Image Preview)",
+    inputPrice: 0.5 * 8,
+    outputPrice: 3 * 8,
+    imageTokenPricePerMillion: 60 * 8,
+    imageOutputTokenEstimateBySize: {
+      "1K": 1120,
+      "2K": 1680,
+      "4K": 2520,
+    },
+    maxOutputTokens: 65536,
+    contextWindow: 65536,
+  }),
+  createGoogleImageModel({
+    name: "gemini-2.5-flash-image",
+    displayName: "Gemini 2.5 Flash Image (Nano Banana)",
+    inputPrice: 0.3 * 8,
+    outputPrice: 2.5 * 8,
+    imageTokenPricePerMillion: 30 * 8,
+    imageOutputTokenEstimateBySize: {
+      "1K": 1290,
+    },
+    maxOutputTokens: 4096,
+    contextWindow: 32768,
+  }),
+];

package/integrations/openai/chatCompletionStreamMode.ts

@@ -0,0 +1,10 @@
+export const applyChatCompletionsStreamMode = (
+  body: Record<string, any>,
+  stream: boolean
+): Record<string, any> => {
+  const nextBody = { ...body, stream };
+  if (!stream) {
+    delete nextBody.stream_options;
+  }
+  return nextBody;
+};

package/integrations/openai/filterAndCleanMessages.ts

@@ -0,0 +1,284 @@
+// packages/integrations/openai/filterAndCleanMessages.ts
+
+export type MessagePart =
+  | { type: "text"; text: string }
+  | { type: "image_url"; image_url: { url: string; detail?: "low" | "high" | "auto" } };
+
+export interface ToolCallFunction {
+  name: string;
+  arguments: string;
+}
+
+export interface ToolCall {
+  id: string;
+  type: "function";
+  function: ToolCallFunction;
+}
+
+export interface OpenAIMessage {
+  id?: string;
+  role: "system" | "user" | "assistant" | "tool";
+  content: string | MessagePart[];
+  name?: string;
+  tool_calls?: ToolCall[];
+  tool_call_id?: string;
+}
+
+export interface InternalMessage extends Omit<Partial<OpenAIMessage>, "content"> {
+  role: OpenAIMessage["role"];
+  content?: unknown;
+  toolName?: string;
+  toolCalls?: ToolCall[];
+  toolCallId?: string;
+  toolPayload?: {
+    toolName?: string;
+    rawToolCall?: ToolCall & { id: string };
+  };
+  [key: string]: any;
+}
+
+const TOOL_CONTEXT_MAX_CHARS = 4000;
+
+const truncateText = (text: string, maxChars: number = TOOL_CONTEXT_MAX_CHARS): string => {
+  if (text.length <= maxChars) return text;
+  return `${text.slice(0, maxChars)}\n…[截断，原始长度 ${text.length} 字符]`;
+};
+
+export const isValidMessagePart = (part: any): part is MessagePart => {
+  if (!part || typeof part !== "object") return false;
+  if (part.type === "text") return typeof part.text === "string";
+  if (part.type === "image_url") {
+    const img = part.image_url;
+    return img && typeof img === "object" && typeof img.url === "string" && img.url.length > 0;
+  }
+  return false;
+};
+
+const extractContent = (msg: InternalMessage): string | MessagePart[] | null => {
+  if (msg.role === "tool") {
+    const llmContext = typeof msg.toolPayload?.llmContext === "string"
+      ? msg.toolPayload.llmContext.trim()
+      : "";
+    if (llmContext) return truncateText(llmContext);
+
+    const summary = typeof msg.toolPayload?.summary === "string"
+      ? msg.toolPayload.summary.trim()
+      : "";
+    if (summary) return truncateText(summary);
+
+    if (typeof msg.content === "string") return truncateText(msg.content);
+    if (msg.content != null) return truncateText(JSON.stringify(msg.content));
+    return null;
+  }
+
+  if (typeof msg.content === "string") return msg.content;
+  if (Array.isArray(msg.content)) {
+    const parts = msg.content.filter(isValidMessagePart);
+    return parts.length > 0 ? parts : null;
+  }
+  if (msg.content && typeof msg.content === "object" && isValidMessagePart(msg.content)) {
+    return [msg.content as MessagePart];
+  }
+  return null;
+};
+
+const extractToolCalls = (msg: InternalMessage): ToolCall[] | undefined => {
+  const calls = msg.tool_calls ?? msg.toolCalls;
+  return Array.isArray(calls) && calls.length > 0 ? calls : undefined;
+};
+
+const extractToolCallId = (msg: InternalMessage): string | undefined =>
+  msg.tool_call_id ?? msg.toolCallId ?? msg.toolPayload?.rawToolCall?.id;
+
+const extractName = (msg: InternalMessage): string | undefined => {
+  if (typeof msg.name === "string") return msg.name;
+  if (typeof msg.toolName === "string") return msg.toolName;
+  if (typeof msg.toolPayload?.toolName === "string") return msg.toolPayload.toolName;
+  return undefined;
+};
+
+const getToolName = (msg: any): string | undefined =>
+  msg?.toolName ?? msg?.toolPayload?.toolName;
+
+const parseJsonObject = (value: unknown): Record<string, any> | undefined => {
+  if (typeof value !== "string") return undefined;
+  try {
+    const parsed = JSON.parse(value);
+    return parsed && typeof parsed === "object" && !Array.isArray(parsed)
+      ? parsed
+      : undefined;
+  } catch {
+    return undefined;
+  }
+};
+
+const getStreamParallelReturnMode = (msg: any): string | undefined =>
+  msg?.toolPayload?.input?.returnMode ?? parseJsonObject(msg?.content)?.returnMode;
+
+const isHandoffToolMessage = (msg: any): boolean => {
+  const toolName = getToolName(msg);
+  if (toolName === "runStreamingAgent") return true;
+  if (toolName !== "streamParallelAgents") return false;
+  return getStreamParallelReturnMode(msg) !== "continue";
+};
+
+const getToolMessageCallId = (msg: any): string | undefined =>
+  msg?.tool_call_id ?? msg?.toolCallId ?? msg?.toolPayload?.rawToolCall?.id;
+
+const collectHandoffToolCallIds = (msgs: any[]): Set<string> => {
+  const ids = new Set<string>();
+  for (const msg of msgs) {
+    if (msg?.role !== "tool" || !isHandoffToolMessage(msg)) continue;
+    const id = getToolMessageCallId(msg);
+    if (id) ids.add(id);
+  }
+  return ids;
+};
+
+const stripHandoffToolMessage = (msg: any): any | null =>
+  msg?.role === "tool" && isHandoffToolMessage(msg) ? null : msg;
+
+const removeToolCallsById = (calls: any[] | undefined, ids: Set<string>) =>
+  calls?.filter((call) => !ids.has(call.id));
+
+const stripHandoffToolCallsFromAssistant = (msg: any, handoffIds: Set<string>): any => {
+  if (msg?.role !== "assistant") return msg;
+
+  const calls: any[] | undefined = msg.tool_calls ?? msg.toolCalls;
+  if (!calls?.some((call) => handoffIds.has(call.id))) return msg;
+
+  const kept = removeToolCallsById(calls, handoffIds);
+  return {
+    ...msg,
+    tool_calls: kept?.length ? kept : undefined,
+    toolCalls: kept?.length ? kept : undefined,
+  };
+};
+
+export const stripHandoffToolMessages = (msgs: any[]): any[] => {
+  const handoffIds = collectHandoffToolCallIds(msgs);
+  if (handoffIds.size === 0) return msgs;
+
+  return msgs
+    .map(stripHandoffToolMessage)
+    .filter(Boolean)
+    .map((msg) => stripHandoffToolCallsFromAssistant(msg, handoffIds));
+};
+
+const toOpenAIMessage = (rawMsg: any): OpenAIMessage | null => {
+  const msg = rawMsg as InternalMessage;
+  const { role, id } = msg;
+
+  if (!role || !["system", "user", "assistant", "tool"].includes(role)) return null;
+
+  const toolCalls = extractToolCalls(msg);
+  let content = extractContent(msg);
+
+  // assistant with tool_calls 允许 content 为空
+  if (content === null) {
+    if (role === "assistant" && toolCalls) content = "";
+    else return null;
+  }
+
+  const cleaned: OpenAIMessage = { role, content };
+
+  if (id) cleaned.id = id;
+  const name = extractName(msg);
+  if (name) cleaned.name = name;
+
+  if (role === "assistant" && toolCalls) {
+    cleaned.tool_calls = toolCalls;
+  }
+
+  if (role === "tool") {
+    const toolCallId = extractToolCallId(msg);
+    if (toolCallId) cleaned.tool_call_id = toolCallId;
+    if (!cleaned.tool_call_id) {
+      console.warn("[filterAndCleanMessages] 丢弃无效 tool 消息：缺少 tool_call_id", msg);
+      return null;
+    }
+  }
+
+  return cleaned;
+};
+
+/**
+ * 清除孤立的 tool_result / assistant stub 对。
+ *
+ * 一条 tool 消息若在其前面最近的 assistant 消息的 tool_calls 中找不到对应 id，
+ * 则视为孤立——连同对应的空 assistant stub 一起丢弃。
+ */
+const removeOrphanedToolPairs = (msgs: OpenAIMessage[]): OpenAIMessage[] => {
+  const dropIndexes = new Set<number>();
+
+  // 为每条 tool 消息找到对应的 assistant 索引
+  const toolToAssistant = new Map<number, number>();
+  for (let i = 0; i < msgs.length; i++) {
+    if (msgs[i].role !== "tool") continue;
+    for (let j = i - 1; j >= 0; j--) {
+      if (msgs[j].role === "assistant") {
+        toolToAssistant.set(i, j);
+        break;
+      }
+    }
+  }
+
+  // 标记孤立的 tool 消息
+  for (let i = 0; i < msgs.length; i++) {
+    if (msgs[i].role === "tool" && !toolToAssistant.has(i)) {
+      dropIndexes.add(i);
+      console.warn("[filterAndCleanMessages] 丢弃孤立 tool 消息：缺少前置 assistant", msgs[i].tool_call_id);
+    }
+  }
+
+  for (const [toolIdx, assistantIdx] of toolToAssistant) {
+    const toolMsg = msgs[toolIdx];
+    if (!toolMsg.tool_call_id) continue;
+
+    const knownIds = new Set((msgs[assistantIdx].tool_calls ?? []).map((tc) => tc.id));
+    if (!knownIds.has(toolMsg.tool_call_id)) {
+      dropIndexes.add(toolIdx);
+      console.warn("[filterAndCleanMessages] 丢弃孤立 tool 消息:", toolMsg.tool_call_id);
+    }
+  }
+
+  // 检查 assistant stub：若其所有 tool_calls 都已被孤立且自身无文本，一并丢弃
+  for (const assistantIdx of new Set(toolToAssistant.values())) {
+    if (dropIndexes.has(assistantIdx)) continue;
+
+    const stub = msgs[assistantIdx];
+    const content = typeof stub.content === "string" ? stub.content.trim() : stub.content;
+    const isEmpty = content === "" || (Array.isArray(content) && content.length === 0);
+    if (!isEmpty) continue;
+
+    const allToolsDropped = (stub.tool_calls ?? []).every((tc) => {
+      // 找所有引用该 id 的 tool 消息，看是否全被 drop
+      return msgs.every((m, idx) =>
+        m.role !== "tool" || m.tool_call_id !== tc.id || dropIndexes.has(idx)
+      );
+    });
+
+    if (allToolsDropped) {
+      dropIndexes.add(assistantIdx);
+      console.warn("[filterAndCleanMessages] 一并丢弃孤立 assistant stub（index:", assistantIdx, ")");
+    }
+  }
+
+  if (dropIndexes.size === 0) return msgs;
+  return msgs.filter((_, idx) => !dropIndexes.has(idx));
+};
+
+/**
+ * 将内部消息数组清洗为严格的 OpenAIMessage[]。
+ *
+ * 调用顺序建议：
+ *   selectAllMsgs → buildAgentViewMessages → filterAndCleanMessages → generateRequestBody
+ */
+export const filterAndCleanMessages = (msgs: any[]): OpenAIMessage[] => {
+  if (!Array.isArray(msgs)) return [];
+
+  const flat = msgs.flat();
+  const preprocessed = stripHandoffToolMessages(flat);
+  const cleaned = preprocessed.map(toOpenAIMessage).filter((m): m is OpenAIMessage => m !== null);
+  return removeOrphanedToolPairs(cleaned);
+};

package/integrations/openai/flexTier.ts

@@ -0,0 +1,102 @@
+export type OpenAIServiceTier = "flex" | "default";
+
+type OpenAIModelProfile = {
+  upstreamModel: string;
+  defaultServiceTier?: OpenAIServiceTier;
+  preAdjustedFlexPrice?: boolean;
+};
+
+const OPENAI_MODEL_PROFILES: Record<string, OpenAIModelProfile> = {
+  "gpt-5.5": {
+    upstreamModel: "gpt-5.5",
+    defaultServiceTier: "default",
+  },
+  "gpt-5.5-flex": {
+    upstreamModel: "gpt-5.5",
+    defaultServiceTier: "flex",
+    preAdjustedFlexPrice: true,
+  },
+  "gpt-5.5-pro-flex": {
+    upstreamModel: "gpt-5.5-pro",
+    defaultServiceTier: "flex",
+    preAdjustedFlexPrice: true,
+  },
+  "gpt-5.4": {
+    upstreamModel: "gpt-5.4",
+    defaultServiceTier: "default",
+  },
+  "gpt-5.4-flex": {
+    upstreamModel: "gpt-5.4",
+    defaultServiceTier: "flex",
+    preAdjustedFlexPrice: true,
+  },
+  "gpt-5.4-pro-flex": {
+    upstreamModel: "gpt-5.4-pro",
+    defaultServiceTier: "flex",
+    preAdjustedFlexPrice: true,
+  },
+};
+
+const getOpenAIModelProfile = (model?: string | null): OpenAIModelProfile | undefined => {
+  if (!model) return undefined;
+  return OPENAI_MODEL_PROFILES[model];
+};
+
+const isExplicitFlexModel = (model?: string | null): boolean => {
+  return getOpenAIModelProfile(model)?.defaultServiceTier === "flex";
+};
+
+export const resolveOpenAIUpstreamModel = ({
+  providerName,
+  model,
+}: {
+  providerName?: string | null;
+  model?: string | null;
+}): string | undefined => {
+  if (!model) return undefined;
+  if (providerName !== "openai") return model;
+  return getOpenAIModelProfile(model)?.upstreamModel ?? model;
+};
+
+export const isOpenAIFlexPricedModel = (model?: string | null): boolean => {
+  return getOpenAIModelProfile(model)?.preAdjustedFlexPrice === true;
+};
+
+export const resolveOpenAIServiceTier = ({
+  providerName,
+  model,
+  requestedServiceTier,
+}: {
+  providerName?: string | null;
+  model?: string | null;
+  requestedServiceTier?: string | null;
+}): OpenAIServiceTier | undefined => {
+  if (providerName !== "openai" || !model) return undefined;
+  const normalizedRequestedServiceTier =
+    requestedServiceTier === "flex" || requestedServiceTier === "default"
+      ? requestedServiceTier
+      : undefined;
+
+  if (normalizedRequestedServiceTier === "flex") {
+    return isExplicitFlexModel(model) ? "flex" : undefined;
+  }
+  if (normalizedRequestedServiceTier === "default") return "default";
+  return getOpenAIModelProfile(model)?.defaultServiceTier;
+};
+
+export const resolveOpenAIRequestTarget = ({
+  providerName,
+  model,
+  requestedServiceTier,
+}: {
+  providerName?: string | null;
+  model?: string | null;
+  requestedServiceTier?: string | null;
+}): { model?: string; serviceTier?: OpenAIServiceTier } => ({
+  model: resolveOpenAIUpstreamModel({ providerName, model }),
+  serviceTier: resolveOpenAIServiceTier({
+    providerName,
+    model,
+    requestedServiceTier,
+  }),
+});