nolo-cli 0.1.30 → 0.1.32

package/ai/context/calculateContextUsage.ts

@@ -1,133 +0,0 @@
-// ai/context/calculateContextUsage.ts
-
-/**
- * 计算 Agent 的 Context 使用情况。
- * 
- * 用于：
- * 1. Agent 编辑页面的 Token 预算可视化
- * 2. 运行时的 Context 监控
- * 3. 超标警告
- */
-
-import { estimateTokenCount, CONTEXT_BUDGET, calcTokenUsagePercent } from "./tokenUtils";
-import { getModelContextWindow } from "../llm/getModelContextWindow";
-import type { ReferenceItem } from "../../app/types";
-
-export interface ContextUsageBreakdown {
-    // 各部分 Token 使用
-    referencesTokens: number;
-    spaceContextTokens: number;
-    systemPromptTokens: number;
-    historyTokens: number;
-
-    // 总计
-    totalUsed: number;
-    modelContextWindow: number;
-    availableTokens: number;
-
-    // 百分比
-    usedPercent: number;
-    referencesPercent: number;
-
-    // 状态
-    isWarning: boolean;
-    isCritical: boolean;
-    warningMessage?: string;
-}
-
-export interface ContextUsageInput {
-    modelName: string;
-    references?: ReferenceItem[];
-    referencesContent?: string; // 如果已经拼接好的内容
-    spaceContext?: string | null;
-    systemPrompt?: string;
-    historyContext?: string | null;
-}
-
-/**
- * 计算 Context 使用情况
- */
-export const calculateContextUsage = (input: ContextUsageInput): ContextUsageBreakdown => {
-    const {
-        modelName,
-        references,
-        referencesContent,
-        spaceContext,
-        systemPrompt,
-        historyContext,
-    } = input;
-
-    // 获取模型的 Context Window
-    const modelContextWindow = getModelContextWindow(modelName);
-
-    // 估算各部分 Token 数
-    // References: 如果提供了 content 直接用，否则根据数量估算
-    let referencesTokens = 0;
-    if (referencesContent) {
-        referencesTokens = estimateTokenCount(referencesContent);
-    } else if (references && references.length > 0) {
-        // 粗略估算：每个 reference 平均 2000 tokens（需要实际获取内容才能精确）
-        referencesTokens = references.length * 2000;
-    }
-
-    const spaceContextTokens = estimateTokenCount(spaceContext || "");
-    const systemPromptTokens = estimateTokenCount(systemPrompt || "");
-    const historyTokens = estimateTokenCount(historyContext || "");
-
-    // 计算总使用量（References + Space 是预分配的，其他是动态的）
-    const preAllocatedTokens = referencesTokens + spaceContextTokens;
-    const totalUsed = preAllocatedTokens + systemPromptTokens + historyTokens;
-    const availableTokens = modelContextWindow - totalUsed;
-
-    // 计算百分比
-    const usedPercent = calcTokenUsagePercent(totalUsed, modelContextWindow);
-    const referencesPercent = calcTokenUsagePercent(preAllocatedTokens, modelContextWindow);
-
-    // 判断状态
-    const isWarning = referencesPercent > CONTEXT_BUDGET.REFERENCES_MAX_PERCENT;
-    const isCritical = usedPercent > 80;
-
-    // 生成警告消息
-    let warningMessage: string | undefined;
-    if (isCritical) {
-        warningMessage = `⚠️ 上下文已使用 ${usedPercent}%，可能影响对话质量。建议减少 References 或 Space 内容。`;
-    } else if (isWarning) {
-        warningMessage = `⚡ References 和 Space 已占用 ${referencesPercent}% 上下文，建议控制在 ${CONTEXT_BUDGET.REFERENCES_MAX_PERCENT}% 以内。`;
-    }
-
-    return {
-        referencesTokens,
-        spaceContextTokens,
-        systemPromptTokens,
-        historyTokens,
-        totalUsed,
-        modelContextWindow,
-        availableTokens,
-        usedPercent,
-        referencesPercent,
-        isWarning,
-        isCritical,
-        warningMessage,
-    };
-};
-
-/**
- * 快速检查 References 是否超标
- * 用于 Agent 编辑页面的实时反馈
- */
-export const checkReferencesOverBudget = (
-    modelName: string,
-    referencesTokens: number
-): { isOver: boolean; percent: number; message?: string } => {
-    const modelContextWindow = getModelContextWindow(modelName);
-    const percent = calcTokenUsagePercent(referencesTokens, modelContextWindow);
-    const isOver = percent > CONTEXT_BUDGET.REFERENCES_MAX_PERCENT;
-
-    return {
-        isOver,
-        percent,
-        message: isOver
-            ? `References 占用 ${percent}%，超过建议上限 ${CONTEXT_BUDGET.REFERENCES_MAX_PERCENT}%`
-            : undefined,
-    };
-};

package/ai/context/retention.ts

@@ -1,165 +0,0 @@
-// 文件路径: packages/ai/context/retention.ts
-
-/**
- * 将 contextRetention slider (1–100) 映射为实际保留比例 R ∈ (0, 1)。
- *
- * 分三档线性映射：
- * - 低保留：  1–30  → 15% ~ 30%
- * - 中保留： 31–70 → 30% ~ 60%
- * - 高保留： 71–100 → 60% ~ 85%
- */
-export const mapRetentionSliderToRatio = (slider: number): number => {
-  const s = Math.min(100, Math.max(1, slider));
-
-  if (s <= 30) {
-    // Low retention: 15% ~ 30%
-    return 0.15 + (0.30 - 0.15) * (s / 30);
-  }
-
-  if (s <= 70) {
-    // Medium retention: 30% ~ 60%
-    return 0.30 + (0.60 - 0.30) * ((s - 30) / 40);
-  }
-
-  // High retention: 60% ~ 85%
-  return 0.60 + (0.85 - 0.60) * ((s - 70) / 30);
-};
-
-/**
- * 安全系数：只允许历史对话最多占用 contextWindow 的 SAFE_BUFFER_RATIO 部分，
- * 预留剩余空间给 system prompt / 工具描述等固定开销。
- */
-export const SAFE_BUFFER_RATIO = 0.9;
-
-/**
- * 对话负载分档：
- * - light  ：短句聊天为主
- * - medium ：中等长度混合
- * - heavy  ：长代码 / 长文档为主
- */
-export type ConversationLoad = "light" | "medium" | "heavy";
-
-export interface ContextPlan {
-  /**
-   * 历史总预算（summary + 原始消息），单位：token
-   * 实际不会超过 contextWindow * SAFE_BUFFER_RATIO
-   */
-  historyBudget: number;
-  /**
-   * 历史中分配给「原始消息」的预算，单位：token
-   * 用于 trimMessagesWithSummary 之类的裁剪逻辑。
-   */
-  rawMessageBudget: number;
-  /**
-   * 希望至少保留的「最近尾部消息」的 token 数，用作二次兜底，
-   * 确保短句对话能多保几轮，重内容对话能保住最近几条大块内容。
-   */
-  minTailTokens: number;
-}
-
-const clamp = (v: number, min: number, max: number): number =>
-  Math.min(max, Math.max(min, v));
-
-/**
- * 统一的上下文使用规划函数：
- * - 输入：模型 window、用户 slider、现有 summary 长度、近期对话负载分档
- * - 输出：历史总预算 + 原始消息预算 + 尾部最少保留 token 数
- *
- * 约束与策略：
- * - 历史最多只能占用 contextWindow * SAFE_BUFFER_RATIO
- * - slider 决定「整体偏好多记还是少记」
- * - recentLoad 决定在该偏好下的微调：
- *   - light ：适当放宽，给更多历史空间
- *   - heavy ：适当收紧，为当前大块内容 / 工具描述预留空间
- */
-export const planContextUsage = (params: {
-  contextWindow: number;
-  retentionSlider: number;
-  summaryTokens: number;
-  recentLoad: ConversationLoad;
-}): ContextPlan => {
-  const { contextWindow, retentionSlider, summaryTokens, recentLoad } = params;
-
-  const safeWindowLimit = Math.floor(contextWindow * SAFE_BUFFER_RATIO);
-
-  const s = Math.min(100, Math.max(1, retentionSlider));
-  const iq = s / 100; // 0~1，表示用户偏好的“记忆强度”
-
-  const isBigWindow = contextWindow >= 128000; // 例如 128k / 1M 等大模型
-  const isSmallWindow = contextWindow <= 32000; // 例如 8k / 16k 等小模型
-
-  // 基础历史比例（只考虑 slider，不考虑对话负载）
-  // - 小窗口：历史占 35% ~ 70% 的 safeWindow
-  // - 中窗口：历史占 40% ~ 75%
-  // - 大窗口：历史占 65% ~ 90%（尽量吃满大模型上下文）
-  let baseHistoryRatio: number;
-  if (isBigWindow) {
-    baseHistoryRatio = 0.65 + 0.25 * iq; // 0.65 ~ 0.90
-  } else if (isSmallWindow) {
-    baseHistoryRatio = 0.35 + 0.35 * iq; // 0.35 ~ 0.70
-  } else {
-    baseHistoryRatio = 0.40 + 0.35 * iq; // 0.40 ~ 0.75
-  }
-
-  // 1. 历史比例：在 baseHistoryRatio 基础上，根据对话负载做微调
-  let historyRatio = baseHistoryRatio;
-
-  switch (recentLoad) {
-    case "light": {
-      // 短句对话：可以多给一点历史空间，但仍保留少量余量给其他上下文
-      const maxRatio = isBigWindow ? 0.95 : 0.85;
-      historyRatio = clamp(historyRatio * 1.05, 0.3, maxRatio);
-      break;
-    }
-    case "heavy": {
-      // 重内容对话：适当收紧历史比例，为当前大块内容 / 工具描述等预留更多窗口
-      const maxRatio = isBigWindow ? 0.90 : 0.80;
-      historyRatio = clamp(historyRatio * 0.9, 0.3, maxRatio);
-      break;
-    }
-    case "medium":
-    default: {
-      const maxRatio = isBigWindow ? 0.90 : 0.85;
-      historyRatio = clamp(historyRatio, 0.3, maxRatio);
-      break;
-    }
-  }
-
-  const historyBudget = Math.max(
-    0,
-    Math.floor(safeWindowLimit * historyRatio),
-  );
-
-  // 2. 在历史预算里分配给「原始消息」的预算
-  //    - summary 已经占用一部分
-  //    - 同时保证原始消息至少有一部分空间（heavy 用户更多一点）
-  const minRawRatio = recentLoad === "heavy" ? 0.4 : 0.2;
-
-  const rawMessageBudget = Math.max(
-    Math.floor(historyBudget * minRawRatio),
-    historyBudget - summaryTokens,
-    0,
-  );
-
-  // 3. 希望至少保留的尾部 token 数
-  //    - light ：多保几轮短句（不超过 30% window，且上限 8k）
-  //    - heavy ：至少给几条大块代码空间（>= 16k，或 25% window）
-  //    - medium：取中间值
-  let minTailTokens: number;
-
-  if (recentLoad === "light") {
-    minTailTokens = Math.min(
-      Math.floor(safeWindowLimit * 0.3),
-      8000,
-    );
-  } else if (recentLoad === "heavy") {
-    minTailTokens = Math.max(
-      Math.floor(safeWindowLimit * 0.25),
-      16000,
-    );
-  } else {
-    minTailTokens = Math.floor(safeWindowLimit * 0.2);
-  }
-
-  return { historyBudget, rawMessageBudget, minTailTokens };
-};

package/ai/context/tokenUtils.ts

@@ -1,78 +0,0 @@
-// ai/context/tokenUtils.ts
-
-/**
- * 简单的 Token 估算工具。
- * 
- * 规则：
- * - 中文字符：约 1.5 tokens/字
- * - 英文及其他字符：约 0.25 tokens/字符（即 4 字符 ≈ 1 token）
- * 
- * 这是一个粗略估算，精确值需要使用 tiktoken 等库。
- */
-
-/**
- * 估算文本的 Token 数量
- */
-export const estimateTokenCount = (text: string): number => {
-    if (!text) return 0;
-
-    // 匹配中文字符（包括常用汉字范围）
-    const chineseChars = (text.match(/[\u4e00-\u9fa5\u3400-\u4dbf]/g) || []).length;
-    const otherChars = text.length - chineseChars;
-
-    // 中文 1.5 token/字，其他 0.25 token/字符
-    return Math.ceil(chineseChars * 1.5 + otherChars * 0.25);
-};
-
-/**
- * 估算多个文本的 Token 总数
- */
-export const estimateTotalTokens = (texts: (string | undefined | null)[]): number => {
-    return texts.reduce((sum, text) => sum + estimateTokenCount(text || ""), 0);
-};
-
-/**
- * 格式化 Token 数量为可读字符串
- * 例如：1500 -> "1.5k", 150000 -> "150k"
- */
-export const formatTokenCount = (count: number): string => {
-    if (count < 1000) return String(count);
-    if (count < 10000) return `${(count / 1000).toFixed(1)}k`;
-    return `${Math.round(count / 1000)}k`;
-};
-
-/**
- * 计算 Token 使用百分比
- */
-export const calcTokenUsagePercent = (used: number, total: number): number => {
-    if (total <= 0) return 0;
-    return Math.min(100, Math.round((used / total) * 100));
-};
-
-/**
- * Context 分配建议阈值
- * - References/Space Context 不应超过总量的 40%
- * - 为对话历史和用户输入预留 40%
- * - 为系统提示预留 10%
- * - 为输出预留 10%
- */
-export const CONTEXT_BUDGET = {
-    REFERENCES_MAX_PERCENT: 40, // References + Space Context 最大占比
-    HISTORY_RESERVE_PERCENT: 40, // 为对话历史预留
-    SYSTEM_RESERVE_PERCENT: 10, // 系统提示预留
-    OUTPUT_RESERVE_PERCENT: 10, // 输出预留
-};
-
-/**
- * 判断 Token 使用是否超过警告阈值
- */
-export const isTokenUsageWarning = (usedPercent: number): boolean => {
-    return usedPercent > CONTEXT_BUDGET.REFERENCES_MAX_PERCENT;
-};
-
-/**
- * 判断 Token 使用是否严重超标
- */
-export const isTokenUsageCritical = (usedPercent: number): boolean => {
-    return usedPercent > 60; // 超过 60% 可能影响对话历史
-};

package/ai/llm/agentCapabilities.ts

@@ -1,74 +0,0 @@
-import {
-  findModelConfig,
-  getProviderByModelName,
-} from "./providers";
-
-type AgentCapabilityConfig = {
-  apiSource?: string | null;
-  provider?: string | null;
-  model?: string | null;
-  hasVision?: boolean | null;
-  useServerProxy?: boolean | null;
-};
-
-const isCustomAgent = (agent: AgentCapabilityConfig): boolean => {
-  const apiSource = agent.apiSource?.toLowerCase();
-  const provider = agent.provider?.toLowerCase();
-  return apiSource === "custom" || provider === "custom";
-};
-
-const lookupKnownModelVision = (
-  provider: string | null,
-  model: string,
-): boolean | undefined => {
-  if (!model) return undefined;
-
-  if (provider) {
-    const direct = findModelConfig(provider, model);
-    if (direct) return direct.hasVision;
-  }
-
-  const detected = getProviderByModelName(model);
-  if (detected) {
-    return findModelConfig(detected, model)?.hasVision;
-  }
-
-  if (!provider && model.includes("/")) {
-    const slash = model.indexOf("/");
-    const modelProvider = model.slice(0, slash).toLowerCase();
-    const modelName = model.slice(slash + 1);
-    const nested = findModelConfig(modelProvider, modelName);
-    if (nested) return nested.hasVision;
-  }
-
-  return undefined;
-};
-
-export const resolveAgentImageInputSupport = (
-  agent: AgentCapabilityConfig | null | undefined,
-): boolean => {
-  if (!agent) return true;
-
-  const provider = agent.provider?.trim().toLowerCase() || null;
-  const model = agent.model?.trim() || "";
-  const custom = isCustomAgent(agent);
-  const catalogHasVision = lookupKnownModelVision(
-    custom ? null : provider,
-    model,
-  );
-
-  if (!custom && catalogHasVision !== undefined) {
-    return catalogHasVision;
-  }
-
-  if (custom) {
-    if (catalogHasVision === true) return true;
-    return true;
-  }
-
-  if (typeof agent.hasVision === "boolean") {
-    return agent.hasVision;
-  }
-
-  return true;
-};

package/ai/llm/calculateGeminiImageTokens.ts

@@ -1,57 +0,0 @@
-// src/calcGeminiImageTokens.ts
-
-/**
- * Gemini 图片 token 计算（粗略公式版）
- *
- * 规则：
- * - 若宽高都 <= 384，则直接按 1 个图块：258 tokens
- * - 否则：
- *   - uRaw = floor( min(width, height) / 1.5 )
- *   - u = clamp(uRaw, 256, 768)
- *   - tilesX = ceil(width  / u)
- *   - tilesY = ceil(height / u)
- *   - tiles  = tilesX * tilesY
- *   - tokens = tiles * 258
- */
-
-export const GEMINI_IMAGE_TILE_TOKENS = 258;
-export const GEMINI_SMALL_IMAGE_THRESHOLD = 384;
-export const GEMINI_MIN_TILE_SIZE = 256;
-export const GEMINI_MAX_TILE_SIZE = 768;
-
-const clamp = (value: number, min: number, max: number): number =>
-    Math.min(Math.max(value, min), max);
-
-const ceilDiv = (numerator: number, denominator: number): number =>
-    Math.ceil(numerator / denominator);
-
-/**
- * 计算单张图片的 token 数（适用于 Gemini 1.5 / 2.0 / 2.5 的粗略估算）
- */
-export const calculateGeminiImageTokens = (
-    width: number,
-    height: number,
-): number => {
-    if (!Number.isFinite(width) || !Number.isFinite(height)) {
-        throw new TypeError('width 和 height 必须是有限数字');
-    }
-    if (width <= 0 || height <= 0) {
-        throw new RangeError('width 和 height 必须为正数');
-    }
-
-    // 情况 1：小图（两个维度都 <= 384）
-    if (width <= GEMINI_SMALL_IMAGE_THRESHOLD && height <= GEMINI_SMALL_IMAGE_THRESHOLD) {
-        return GEMINI_IMAGE_TILE_TOKENS;
-    }
-
-    // 情况 2：大图，按粗略公式计算图块数量
-    const minSide = Math.min(width, height);
-    const rawUnit = Math.floor(minSide / 1.5);
-    const unit = clamp(rawUnit, GEMINI_MIN_TILE_SIZE, GEMINI_MAX_TILE_SIZE);
-
-    const tilesX = ceilDiv(width, unit);
-    const tilesY = ceilDiv(height, unit);
-    const tiles = tilesX * tilesY;
-
-    return tiles * GEMINI_IMAGE_TILE_TOKENS;
-};

package/ai/llm/deepinfra.ts

@@ -1,79 +0,0 @@
-// ai/llm/deepinfra.ts
-import { DEEPINFRA_KIMI_FALLBACK_MODEL } from "./kimi";
-
-export const deepinfraModels = [
-  {
-    name: DEEPINFRA_KIMI_FALLBACK_MODEL,
-    displayName: "MoonshotAI: Kimi K2.6 (DeepInfra)",
-    hasVision: true,
-    price: {
-      input: 0.6 * 8,
-      output: 2.5 * 8,
-      inputCacheHit: 0.06 * 8,
-    },
-    maxOutputTokens: 262144,
-    contextWindow: 262144,
-    supportsTool: true,
-  },
-  {
-    name: "moonshotai/Kimi-K2.5",
-    displayName: "MoonshotAI: Kimi K2.5 (DeepInfra)",
-    hasVision: true,
-    price: {
-      input: 0.45 * 8,
-      output: 2.25 * 8,
-      inputCacheHit: 0.07 * 8,
-    },
-    maxOutputTokens: 262144,
-    contextWindow: 262144,
-    supportsTool: true,
-  },
-  {
-    name: "zai-org/GLM-5.1",
-    displayName: "Z.AI: GLM 5.1 (DeepInfra)",
-    hasVision: false,
-    price: {
-      input: 1.26 * 8,
-      output: 3.96 * 8,
-      inputCacheHit: 0.26 * 8,
-    },
-    contextWindow: 202752,
-    supportsTool: true,
-  },
-  {
-    name: "anthropic/claude-haiku-4-5",
-    displayName: "Anthropic: Claude Haiku 4.5 (DeepInfra)",
-    hasVision: true,
-    price: {
-      input: 1 * 9,
-      output: 5 * 9,
-    },
-    contextWindow: 195000,
-    maxOutputTokens: 4092,
-    supportsTool: false,
-  },
-  {
-    name: "anthropic/claude-sonnet-4-6",
-    displayName: "Anthropic: Claude Sonnet 4.6 (DeepInfra)",
-    hasVision: true,
-    price: {
-      input: 3 * 9,
-      output: 15 * 9,
-    },
-    contextWindow: 976000,
-    maxOutputTokens: 4092,
-    supportsTool: false,
-  },
-  {
-    name: "anthropic/claude-opus-4-7",
-    displayName: "Anthropic: Claude Opus 4.7 (DeepInfra)",
-    hasVision: true,
-    price: {
-      input: 5 * 9,
-      output: 25 * 9,
-    },
-    contextWindow: 976000,
-    maxOutputTokens: 4092,
-    supportsTool: false,
-  },
-];

package/ai/llm/fireworks.ts

@@ -1,68 +0,0 @@
-// ai/llm/fireworks.ts
-import {
-    FIREWORKS_KIMI_CURRENT_MODEL,
-    FIREWORKS_KIMI_LATEST_MODEL,
-} from "./kimi";
-
-export const fireworksModels = [
-    // --- MoonshotAI Models (Kimi) ---
-    {
-        name: FIREWORKS_KIMI_LATEST_MODEL,
-        displayName: "MoonshotAI: Kimi Latest",
-        hasVision: true,
-        price: {
-            input: 0.95 * 8,
-            output: 4.0 * 8,
-            cachingRead: 0.16 * 8,
-        },
-        maxOutputTokens: 262144,
-        contextWindow: 262144,
-        supportsTool: true,
-    },
-    {
-        name: FIREWORKS_KIMI_CURRENT_MODEL,
-        displayName: "MoonshotAI: Kimi K2.6",
-        hasVision: true,
-        price: {
-            input: 0.95 * 8,
-            output: 4.0 * 8,
-            cachingRead: 0.16 * 8,
-        },
-        maxOutputTokens: 262144,
-        contextWindow: 262144,
-        supportsTool: true,
-    },
-    {
-        name: "accounts/fireworks/models/minimax-m2p7",
-        displayName: "MiniMax: MiniMax M2.7",
-        hasVision: true,
-        price: {
-            input: 0.3 * 8,
-            output: 1.2 * 8,
-            cachingRead: 0.06 * 8,
-        },
-        supportsTool: true,
-    },
-    {
-        name: "accounts/fireworks/models/qwen3p6-plus",
-        displayName: "Qwen: Qwen 3.6 Plus",
-        hasVision: true,
-        price: {
-            input: 0.5 * 8,
-            output: 3.0 * 8,
-            cachingRead: 0.1 * 8,
-        },
-        supportsTool: true,
-    },
-    {
-        name: "accounts/fireworks/models/glm-5p1",
-        displayName: "Z.AI: GLM 5.1",
-        hasVision: true,
-        price: {
-            input: 1.0 * 8,
-            output: 3.2 * 8,
-            cachingRead: 0.2 * 8,
-        },
-        supportsTool: true,
-    },
-];