kc-beta 0.1.2 → 0.3.0

package/src/cli/onboard.js

@@ -2,6 +2,8 @@ import fs from "node:fs";
 import path from "node:path";
 import readline from "node:readline";
 import os from "node:os";
+import { getProviders, getProviderById, getProviderLabels, classifyModels, getCuratedModels } from "../providers.js";
+import { LLMClient } from "../agent/llm-client.js";
 
 const CONFIG_DIR = path.join(os.homedir(), ".kc_agent");
 const CONFIG_PATH = path.join(CONFIG_DIR, "config.json");
@@ -23,27 +25,33 @@ const L = {
     langPrompt: "Language",
     langOptions: ["English", "中文"],
     providerPrompt: "LLM Provider",
-    providerLabels: [
-      "SiliconFlow (recommended for China)",
-      "Aliyun Bailian",
-      "Anthropic",
-      "OpenAI",
-      "Custom (enter base URL)",
-    ],
     current: "current",
     choose: "Choose",
     baseUrl: "Base URL",
     baseUrlRequired: "Base URL is required for custom provider.",
     apiKey: "API Key",
     apiKeyRequired: "required",
-    apiKeyKeep: "Enter to keep",
+    apiKeyKeep: "Press Enter to keep",
     apiKeyMissing: "API key is required. Run 'kc-beta onboard' again.",
+    keyType: "Key Type",
+    keyTypeOptions: ["API Key (pay-per-use)", "Coding Plan Key (subscription)"],
     conductorModel: "Conductor Model",
     workerTiers: "Worker LLM Tiers",
-    tierHint: "Enter to accept defaults",
+    vlmTiers: "VLM Tiers (Vision/OCR)",
+    tierHint: "Press Enter to accept defaults",
+    workerConfig: "Worker LLM Provider",
+    workerSameProvider: "Use same provider for worker LLMs?",
+    yesNo: "Y/n",
     accuracy: "Accuracy Threshold",
     saved: "Saved to",
     runHint: "Run {cmd} to start the agent.",
+    discovering: "Discovering available models...",
+    discoveryFailed: "Could not auto-discover models. Using provider defaults.",
+    discoveryFound: "Found {n} models. Suggested tier assignments:",
+    discoveryAccept: "Press Enter to accept, or type model name to override",
+    enterSkip: "Press Enter to skip",
+    enterDefault: "Press Enter to use default",
+    bedrockWarn: "AWS Bedrock is not yet fully supported. Authentication will fail at runtime.",
   },
   zh: {
     title: "KC Agent 配置向导",
@@ -51,13 +59,6 @@ const L = {
     langPrompt: "语言",
     langOptions: ["English", "中文"],
     providerPrompt: "大模型服务商",
-    providerLabels: [
-      "SiliconFlow（国内推荐）",
-      "阿里云百炼",
-      "Anthropic",
-      "OpenAI",
-      "自定义（输入接口地址）",
-    ],
     current: "当前",
     choose: "选择",
     baseUrl: "接口地址",
@@ -66,31 +67,37 @@ const L = {
     apiKeyRequired: "必填",
     apiKeyKeep: "回车保留当前密钥",
     apiKeyMissing: "API 密钥为必填项。请重新运行 'kc-beta onboard'。",
+    keyType: "密钥类型",
+    keyTypeOptions: ["API Key（按量付费）", "Coding Plan Key（包年包月）"],
     conductorModel: "主模型",
     workerTiers: "Worker 模型分层",
+    vlmTiers: "VLM 视觉模型分层（OCR）",
     tierHint: "回车接受默认值",
+    workerConfig: "Worker LLM 服务商",
+    workerSameProvider: "Worker LLM 使用同一服务商？",
+    yesNo: "Y/n",
     accuracy: "准确率阈值",
     saved: "已保存至",
     runHint: "运行 {cmd} 启动 Agent。",
+    discovering: "正在发现可用模型...",
+    discoveryFailed: "无法自动发现模型，使用默认配置。",
+    discoveryFound: "发现 {n} 个模型。建议分层：",
+    discoveryAccept: "回车接受，或输入模型名称覆盖",
+    enterSkip: "回车跳过",
+    enterDefault: "回车使用默认值",
+    bedrockWarn: "AWS Bedrock 尚未完全支持。运行时认证将失败。",
   },
 };
 
-const PROVIDERS = [
-  { name: "SiliconFlow", base_url: "https://api.siliconflow.cn/v1", model: "Pro/zai-org/GLM-5",
-    tiers: { tier1: "Pro/zai-org/GLM-5, Pro/moonshotai/Kimi-K2.5", tier2: "Pro/deepseek-ai/DeepSeek-V3.2, Pro/MiniMaxAI/MiniMax-M2.5", tier3: "Qwen/Qwen3.5-122B-A10B", tier4: "Qwen/Qwen3.5-35B-A3B" } },
-  { name: "Aliyun", base_url: "https://coding.dashscope.aliyuncs.com/v1", model: "glm-5",
-    tiers: { tier1: "glm-5", tier2: "deepseek-v3", tier3: "qwen-plus", tier4: "qwen-turbo" } },
-  { name: "Anthropic", base_url: "https://api.anthropic.com/v1", model: "claude-sonnet-4-20250514",
-    tiers: { tier1: "claude-sonnet-4-20250514", tier2: "claude-sonnet-4-20250514", tier3: "claude-haiku-4-5-20251001", tier4: "claude-haiku-4-5-20251001" } },
-  { name: "OpenAI", base_url: "https://api.openai.com/v1", model: "gpt-4o",
-    tiers: { tier1: "gpt-4o", tier2: "gpt-4o-mini", tier3: "gpt-4o-mini", tier4: "gpt-4o-mini" } },
-  { name: "Custom", base_url: "", model: "", tiers: { tier1: "", tier2: "", tier3: "", tier4: "" } },
-];
-
-function ask(rl, question, defaultValue = "") {
+function ask(rl, question, defaultValue = "", hint = "") {
   const suffix = defaultValue ? ` ${DIM}[${defaultValue}]${RESET}` : "";
+  const hintText = hint
+    ? ` ${GRAY}(${hint})${RESET}`
+    : defaultValue
+      ? ` ${GRAY}(Press Enter to keep)${RESET}`
+      : "";
   return new Promise((resolve) => {
-    rl.question(`${question}${suffix}: `, (answer) => resolve(answer.trim() || defaultValue));
+    rl.question(`${question}${suffix}${hintText}: `, (answer) => resolve(answer.trim() || defaultValue));
   });
 }
 
@@ -101,11 +108,14 @@ export async function onboard() {
   if (fs.existsSync(CONFIG_PATH)) {
     try { existing = JSON.parse(fs.readFileSync(CONFIG_PATH, "utf-8")); } catch { /* ignore */ }
   }
+  const isUpdate = Object.keys(existing).length > 0;
 
   console.log();
   console.log(`  ${BOLD}KC Agent Setup / KC Agent 配置向导${RESET}`);
   console.log(`  ${GRAY}${"─".repeat(40)}${RESET}`);
   console.log();
+
+  // --- Language ---
   console.log(`  ${CYAN}Language / 语言:${RESET}`);
   console.log(`    1. English`);
   console.log(`    2. 中文`);
@@ -115,69 +125,218 @@ export async function onboard() {
   const t = L[lang];
   console.log();
 
-  if (Object.keys(existing).length > 0) {
+  if (isUpdate) {
     console.log(`  ${DIM}${t.existingConfig}${RESET}`);
     console.log();
   }
 
+  // --- Provider ---
+  const providers = getProviders();
+  const labels = getProviderLabels(lang);
   console.log(`  ${CYAN}${t.providerPrompt}:${RESET}`);
-  for (let i = 0; i < PROVIDERS.length; i++) {
-    const marker = PROVIDERS[i].name.toLowerCase() === existing.provider ? ` ${GREEN}(${t.current})${RESET}` : "";
-    console.log(`    ${i + 1}. ${t.providerLabels[i]}${marker}`);
+  for (let i = 0; i < labels.length; i++) {
+    const marker = providers[i].id === existing.provider ? ` ${GREEN}(${t.current})${RESET}` : "";
+    console.log(`    ${i + 1}. ${labels[i].label}${marker}`);
   }
   const providerIdx = parseInt(await ask(rl, `  ${GRAY}>${RESET} ${t.choose}`, "1"), 10) - 1;
-  const provider = PROVIDERS[Math.max(0, Math.min(providerIdx, PROVIDERS.length - 1))];
+  const provider = providers[Math.max(0, Math.min(providerIdx, providers.length - 1))];
   console.log();
 
-  let baseUrl = provider.base_url;
-  if (provider.name === "Custom") {
+  // Bedrock warning
+  if (provider.id === "bedrock") {
+    console.log(`  ${YELLOW}⚠ ${t.bedrockWarn}${RESET}`);
+    console.log();
+  }
+
+  // --- Base URL ---
+  let baseUrl = provider.baseUrl;
+  if (provider.id === "custom") {
     baseUrl = await ask(rl, `  ${t.baseUrl}`, existing.base_url || "");
     if (!baseUrl) { console.log(`  ${RED}${t.baseUrlRequired}${RESET}`); rl.close(); process.exit(1); }
+    console.log();
   }
 
+  // --- Aliyun coding plan key sub-option ---
+  let useCodingPlan = false;
+  if (provider.supportsCodingPlanKey) {
+    console.log(`  ${CYAN}${t.keyType}:${RESET}`);
+    console.log(`    1. ${t.keyTypeOptions[0]}`);
+    console.log(`    2. ${t.keyTypeOptions[1]}`);
+    const keyTypeChoice = await ask(rl, `  ${GRAY}>${RESET} ${t.choose}`, "1");
+    useCodingPlan = keyTypeChoice === "2";
+    if (useCodingPlan && provider.codingPlanUrl) {
+      baseUrl = provider.codingPlanUrl;
+    }
+    console.log();
+  }
+
+  // --- API Key ---
   const maskedExisting = existing.api_key ? existing.api_key.slice(0, 6) + "..." + existing.api_key.slice(-4) : "";
+  const keyHint = maskedExisting ? t.apiKeyKeep : t.apiKeyRequired;
   const keyPrompt = maskedExisting
-    ? `  ${CYAN}${t.apiKey}${RESET} ${DIM}(${maskedExisting}, ${t.apiKeyKeep})${RESET}`
+    ? `  ${CYAN}${t.apiKey}${RESET} ${DIM}(${maskedExisting})${RESET}`
     : `  ${CYAN}${t.apiKey}${RESET} ${YELLOW}(${t.apiKeyRequired})${RESET}`;
-  const apiKey = await ask(rl, keyPrompt, "");
+  const apiKey = await ask(rl, keyPrompt, "", keyHint);
   const finalKey = apiKey || existing.api_key || "";
   if (!finalKey) { console.log(`  ${RED}${t.apiKeyMissing}${RESET}`); rl.close(); process.exit(1); }
   console.log();
 
-  const defaultModel = provider.model || existing.conductor_model || "";
-  const model = await ask(rl, `  ${CYAN}${t.conductorModel}${RESET}`, defaultModel);
+  // --- Auto-discovery ---
+  let discoveredModels = null;
+  let suggestedTiers = null;
+  let suggestedConductor = null;
+
+  // Try curated models first (for providers without /models endpoint)
+  const curated = getCuratedModels(provider.id);
+
+  if (curated) {
+    // Use curated model list
+    discoveredModels = curated;
+    const classified = classifyModels(curated);
+    suggestedTiers = classified.tiers;
+    suggestedConductor = classified.conductor;
+    console.log(`  ${GREEN}✓${RESET} ${t.discoveryFound.replace("{n}", curated.length)}`);
+    if (suggestedConductor) {
+      console.log(`    ${DIM}Conductor: ${suggestedConductor}${RESET}`);
+    }
+    for (const [tier, models] of Object.entries(suggestedTiers)) {
+      if (models) console.log(`    ${DIM}${tier.toUpperCase()}: ${models}${RESET}`);
+    }
+    console.log();
+  } else if (provider.modelsEndpoint) {
+    // Query /models endpoint
+    console.log(`  ${DIM}${t.discovering}${RESET}`);
+    try {
+      const tempClient = new LLMClient({
+        apiKey: finalKey,
+        baseUrl: baseUrl,
+        authType: provider.authType,
+        apiFormat: provider.apiFormat,
+      });
+      discoveredModels = await tempClient.listModels();
+
+      if (discoveredModels && discoveredModels.length > 0) {
+        const classified = classifyModels(discoveredModels);
+        suggestedTiers = classified.tiers;
+        suggestedConductor = classified.conductor;
+        console.log(`  ${GREEN}✓${RESET} ${t.discoveryFound.replace("{n}", discoveredModels.length)}`);
+        if (suggestedConductor) {
+          console.log(`    ${DIM}Conductor: ${suggestedConductor}${RESET}`);
+        }
+        for (const [tier, models] of Object.entries(suggestedTiers)) {
+          if (models) console.log(`    ${DIM}${tier.toUpperCase()}: ${models}${RESET}`);
+        }
+      } else {
+        console.log(`  ${DIM}${t.discoveryFailed}${RESET}`);
+      }
+    } catch {
+      console.log(`  ${DIM}${t.discoveryFailed}${RESET}`);
+    }
+    console.log();
+  }
+
+  // --- Conductor model ---
+  const defaultModel = suggestedConductor || provider.defaultModel || existing.conductor_model || "";
+  const model = await ask(
+    rl,
+    `  ${CYAN}${t.conductorModel}${RESET}`,
+    defaultModel,
+    isUpdate ? t.enterDefault : "",
+  );
   console.log();
 
+  // --- Worker LLM tiers ---
   console.log(`  ${CYAN}${t.workerTiers}${RESET} ${DIM}(${t.tierHint})${RESET}`);
   const tiers = {};
   for (const tier of ["tier1", "tier2", "tier3", "tier4"]) {
-    const def = provider.tiers[tier] || existing?.tiers?.[tier] || "";
-    tiers[tier] = await ask(rl, `    ${tier.toUpperCase()}`, def);
+    const def = suggestedTiers?.[tier] || provider.defaultTiers[tier] || existing?.tiers?.[tier] || "";
+    tiers[tier] = await ask(
+      rl,
+      `    ${tier.toUpperCase()}`,
+      def,
+      t.discoveryAccept ? "" : "",
+    );
   }
   console.log();
 
-  const defaultAcc = existing.accuracy_threshold?.toString() || "0.9";
-  const accuracy = parseFloat(await ask(rl, `  ${CYAN}${t.accuracy}${RESET}`, defaultAcc));
+  // --- VLM tiers (vision/OCR) ---
+  console.log(`  ${CYAN}${t.vlmTiers}${RESET} ${DIM}(${t.tierHint})${RESET}`);
+  const vlmTiers = {};
+  for (const tier of ["tier1", "tier2", "tier3"]) {
+    const def = provider.defaultVlm?.[tier] || existing?.vlm_tiers?.[tier] || "";
+    vlmTiers[tier] = await ask(
+      rl,
+      `    ${tier.toUpperCase()}`,
+      def,
+    );
+  }
   console.log();
 
-  // Advanced thresholds (Enter to keep defaults)
-  const advLabel = lang === "zh" ? "高级阈值" : "Advanced Thresholds";
-  const skipHint = lang === "zh" ? "回车使用默认值" : "Enter to keep defaults";
-  console.log(`  ${CYAN}${advLabel}${RESET} ${DIM}(${skipHint})${RESET}`);
-  const systemicThreshold = parseFloat(await ask(rl, `    ${lang === "zh" ? "系统性问题阈值" : "Systemic threshold"}`, existing.systemic_threshold?.toString() || "0.10"));
-  const spotCheckRate = parseFloat(await ask(rl, `    ${lang === "zh" ? "抽查比率" : "Spot-check rate"}`, existing.spot_check_rate?.toString() || "0.10"));
-  const tierTolerance = parseFloat(await ask(rl, `    ${lang === "zh" ? "降级容差" : "Tier downgrade tolerance"}`, existing.tier_tolerance?.toString() || "0.05"));
+  // --- Worker LLM provider (optional) ---
+  console.log(`  ${CYAN}${t.workerConfig}${RESET}`);
+  const sameProvider = await ask(rl, `  ${t.workerSameProvider}`, "Y", t.yesNo);
+  let workerProvider = "";
+  let workerApiKey = "";
+  let workerBaseUrl = "";
+  let workerAuthType = "";
+  let workerApiFormat = "";
+
+  if (sameProvider.toLowerCase() === "n" || sameProvider.toLowerCase() === "no") {
+    // Pick a different provider for workers
+    console.log();
+    console.log(`  ${CYAN}${t.providerPrompt} (Worker):${RESET}`);
+    for (let i = 0; i < labels.length; i++) {
+      console.log(`    ${i + 1}. ${labels[i].label}`);
+    }
+    const wIdx = parseInt(await ask(rl, `  ${GRAY}>${RESET} ${t.choose}`, "1"), 10) - 1;
+    const wp = providers[Math.max(0, Math.min(wIdx, providers.length - 1))];
+    workerProvider = wp.id;
+    workerAuthType = wp.authType;
+    workerApiFormat = wp.apiFormat;
+    workerBaseUrl = wp.baseUrl;
+
+    if (wp.id === "custom") {
+      workerBaseUrl = await ask(rl, `  ${t.baseUrl}`, existing.worker_base_url || "");
+    }
+
+    // Worker API key
+    const wMasked = existing.worker_api_key ? existing.worker_api_key.slice(0, 6) + "..." + existing.worker_api_key.slice(-4) : "";
+    const wKeyHint = wMasked ? t.apiKeyKeep : t.apiKeyRequired;
+    workerApiKey = await ask(
+      rl,
+      `  ${CYAN}${t.apiKey} (Worker)${RESET}`,
+      "",
+      wKeyHint,
+    );
+    workerApiKey = workerApiKey || existing.worker_api_key || "";
+  }
   console.log();
 
   rl.close();
 
+  // Preserve existing thresholds or set defaults (editable via 'kc-beta config')
+  const accuracy = existing.accuracy_threshold ?? 0.9;
+  const systemicThreshold = existing.systemic_threshold ?? 0.10;
+  const spotCheckRate = existing.spot_check_rate ?? 0.10;
+  const tierTolerance = existing.tier_tolerance ?? 0.05;
+
   const config = {
     language: lang,
-    provider: provider.name.toLowerCase(),
+    provider: provider.id,
     api_key: finalKey,
     base_url: baseUrl,
+    auth_type: provider.authType,
+    api_format: provider.apiFormat,
     conductor_model: model,
     tiers,
+    vlm_tiers: vlmTiers,
+    // Worker LLM (optional — empty means use conductor config)
+    worker_provider: workerProvider,
+    worker_api_key: workerApiKey,
+    worker_base_url: workerBaseUrl,
+    worker_auth_type: workerAuthType,
+    worker_api_format: workerApiFormat,
+    // Thresholds
     accuracy_threshold: accuracy,
     systemic_threshold: systemicThreshold,
     spot_check_rate: spotCheckRate,
@@ -190,5 +349,9 @@ export async function onboard() {
   console.log(`  ${GREEN}✓${RESET} ${t.saved} ${GRAY}${CONFIG_PATH}${RESET}`);
   console.log();
   console.log(`  ${t.runHint.replace("{cmd}", `${BOLD}kc-beta${RESET}`)}`);
+  const configHint = lang === "zh"
+    ? `  ${DIM}运行 ${BOLD}kc-beta config${RESET}${DIM} 调整阈值和高级设置。${RESET}`
+    : `  ${DIM}Run ${BOLD}kc-beta config${RESET}${DIM} to adjust thresholds and advanced settings.${RESET}`;
+  console.log(configHint);
   console.log();
 }

package/src/config.js

@@ -1,6 +1,7 @@
 import fs from "node:fs";
 import path from "node:path";
 import os from "node:os";
+import { getProviderById } from "./providers.js";
 
 const GLOBAL_CONFIG_DIR = path.join(os.homedir(), ".kc_agent");
 const GLOBAL_CONFIG_PATH = path.join(GLOBAL_CONFIG_DIR, "config.json");
@@ -43,31 +44,46 @@ function loadEnvFile(envPath) {
 
 /**
  * Load settings by merging: global config (lowest) -> workspace .env (highest).
+ * Supports both new generic keys (LLM_API_KEY) and legacy keys (SILICONFLOW_API_KEY).
  * @param {string} [workspacePath] - Optional workspace directory for .env override
  */
 export function loadSettings(workspacePath) {
   const gc = loadGlobalConfig();
   const env = workspacePath ? loadEnvFile(path.join(workspacePath, ".env")) : {};
 
-  return {
-    // Conductor LLM
-    llmApiKey: env.SILICONFLOW_API_KEY || gc.api_key || "",
-    llmBaseUrl: env.SILICONFLOW_BASE_URL || gc.base_url || "https://api.siliconflow.cn/v1",
+  // Resolve provider metadata for authType/apiFormat defaults
+  const provider = gc.provider || "siliconflow";
+  const providerDef = getProviderById(provider);
+
+  const settings = {
+    // Provider identity
+    provider,
+    authType: gc.auth_type || providerDef?.authType || "bearer",
+    apiFormat: gc.api_format || providerDef?.apiFormat || "openai",
+
+    // Conductor LLM (generic keys with legacy fallback)
+    llmApiKey: env.LLM_API_KEY || env.SILICONFLOW_API_KEY || gc.api_key || "",
+    llmBaseUrl: env.LLM_BASE_URL || env.SILICONFLOW_BASE_URL || gc.base_url || "https://api.siliconflow.cn/v1",
     kcModel: gc.conductor_model || "glm-5",
     kcMaxTokens: 65536,
 
-    // Worker LLMs (SiliconFlow)
-    siliconflowApiKey: env.SILICONFLOW_API_KEY || gc.api_key || "",
-    siliconflowBaseUrl: env.SILICONFLOW_BASE_URL || gc.base_url || "https://api.siliconflow.cn/v1",
-
     // Tier models (from .env or global config tiers)
     tier1: env.TIER1 || gc.tiers?.tier1 || "",
     tier2: env.TIER2 || gc.tiers?.tier2 || "",
     tier3: env.TIER3 || gc.tiers?.tier3 || "",
     tier4: env.TIER4 || gc.tiers?.tier4 || "",
 
-    // OCR models
-    ocrModelTier1: env.OCR_MODEL_TIER1 || "zai-org/GLM-4.6V",
+    // VLM tiers (vision/OCR models)
+    vlmTier1: env.VLM_TIER1 || gc.vlm_tiers?.tier1 || "",
+    vlmTier2: env.VLM_TIER2 || gc.vlm_tiers?.tier2 || "",
+    vlmTier3: env.VLM_TIER3 || gc.vlm_tiers?.tier3 || "",
+
+    // Worker LLM — optional, defaults to conductor config
+    workerProvider: gc.worker_provider || "",
+    workerApiKey: env.WORKER_API_KEY || gc.worker_api_key || "",
+    workerBaseUrl: env.WORKER_BASE_URL || gc.worker_base_url || "",
+    workerAuthType: gc.worker_auth_type || "",
+    workerApiFormat: gc.worker_api_format || "",
 
     // Document parsing
     mineruApiUrl: env.MINERU_API_URL || "",
@@ -90,9 +106,46 @@ export function loadSettings(workspacePath) {
     maxIterations: parseInt(env.MAX_ITERATIONS || "20", 10),
     monitorFrequency: env.MONITOR_FREQUENCY || "mid",
 
+    // Web search
+    tavilyApiKey: env.TAVILY_API_KEY || gc.tavily_api_key || "",
+
+    // Context management
+    kcContextLimit: parseInt(env.KC_CONTEXT_LIMIT || "200000", 10),
+
     // Language
     language: env.LANGUAGE || gc.language || "en",
   };
+
+  // Effective worker config (falls back to conductor config)
+  settings.effectiveWorkerProvider = () => settings.workerProvider || settings.provider;
+  settings.effectiveWorkerApiKey = () => settings.workerApiKey || settings.llmApiKey;
+  settings.effectiveWorkerBaseUrl = () => {
+    if (settings.workerBaseUrl) return settings.workerBaseUrl;
+    // If worker uses a different provider, use that provider's default base URL
+    if (settings.workerProvider && settings.workerProvider !== settings.provider) {
+      const wp = getProviderById(settings.workerProvider);
+      return wp?.baseUrl || settings.llmBaseUrl;
+    }
+    return settings.llmBaseUrl;
+  };
+  settings.effectiveWorkerAuthType = () => {
+    if (settings.workerAuthType) return settings.workerAuthType;
+    if (settings.workerProvider && settings.workerProvider !== settings.provider) {
+      const wp = getProviderById(settings.workerProvider);
+      return wp?.authType || settings.authType;
+    }
+    return settings.authType;
+  };
+  settings.effectiveWorkerApiFormat = () => {
+    if (settings.workerApiFormat) return settings.workerApiFormat;
+    if (settings.workerProvider && settings.workerProvider !== settings.provider) {
+      const wp = getProviderById(settings.workerProvider);
+      return wp?.apiFormat || settings.apiFormat;
+    }
+    return settings.apiFormat;
+  };
+
+  return settings;
 }
 
 export { GLOBAL_CONFIG_DIR, GLOBAL_CONFIG_PATH };

package/src/model-tiers.json

@@ -0,0 +1,153 @@
+{
+  "_comment": "Model selections per provider. LLM tiers 1-4, VLM tiers 1-3. Edit this file directly to update model assignments.",
+
+  "siliconflow": {
+    "conductor": "Pro/zai-org/GLM-5",
+    "llm": {
+      "tier1": "Pro/zai-org/GLM-5, Pro/moonshotai/Kimi-K2.5",
+      "tier2": "Pro/deepseek-ai/DeepSeek-V3.2, Pro/MiniMaxAI/MiniMax-M2.5",
+      "tier3": "Qwen/Qwen3.5-122B-A10B",
+      "tier4": "Qwen/Qwen3.5-35B-A3B"
+    },
+    "vlm": {
+      "tier1": "Pro/Qwen/Qwen2.5-VL-72B-Instruct",
+      "tier2": "Qwen/Qwen2.5-VL-32B-Instruct",
+      "tier3": "Qwen/Qwen2.5-VL-7B-Instruct"
+    }
+  },
+
+  "aliyun": {
+    "conductor": "qwen3.6-plus",
+    "llm": {
+      "tier1": "qwen3.6-plus",
+      "tier2": "",
+      "tier3": "",
+      "tier4": ""
+    },
+    "vlm": {
+      "tier1": "qwen-vl-max",
+      "tier2": "qwen-vl-plus",
+      "tier3": ""
+    }
+  },
+
+  "volcanocloud": {
+    "conductor": "doubao-seed-2-0-pro-260215",
+    "llm": {
+      "tier1": "doubao-seed-2-0-pro-260215, deepseek-v3-2-251201",
+      "tier2": "glm-4-7-251222, doubao-1-5-pro-32k-250115",
+      "tier3": "doubao-seed-2-0-mini-260215",
+      "tier4": "doubao-seed-2-0-lite-260215, doubao-1-5-lite-32k-250115"
+    },
+    "vlm": {
+      "tier1": "doubao-vision-pro-32k-241028",
+      "tier2": "doubao-vision-lite-32k-241028",
+      "tier3": ""
+    }
+  },
+
+  "anthropic": {
+    "conductor": "claude-sonnet-4-20250514",
+    "llm": {
+      "tier1": "claude-sonnet-4-20250514",
+      "tier2": "claude-sonnet-4-20250514",
+      "tier3": "claude-haiku-4-5-20251001",
+      "tier4": "claude-haiku-4-5-20251001"
+    },
+    "vlm": {
+      "tier1": "claude-sonnet-4-20250514",
+      "tier2": "claude-haiku-4-5-20251001",
+      "tier3": "claude-haiku-4-5-20251001"
+    }
+  },
+
+  "openai": {
+    "conductor": "gpt-4o",
+    "llm": {
+      "tier1": "gpt-4o",
+      "tier2": "gpt-4o-mini",
+      "tier3": "gpt-4o-mini",
+      "tier4": "gpt-4o-mini"
+    },
+    "vlm": {
+      "tier1": "gpt-4o",
+      "tier2": "gpt-4o-mini",
+      "tier3": "gpt-4o-mini"
+    }
+  },
+
+  "zhipu": {
+    "conductor": "glm-4-plus",
+    "llm": {
+      "tier1": "glm-4-plus",
+      "tier2": "glm-4-air",
+      "tier3": "glm-4-flash",
+      "tier4": "glm-4-flash"
+    },
+    "vlm": {
+      "tier1": "glm-4v-plus",
+      "tier2": "glm-4v",
+      "tier3": "glm-4v-flash"
+    }
+  },
+
+  "minimax": {
+    "conductor": "MiniMax-M2.5",
+    "llm": {
+      "tier1": "MiniMax-M2.5",
+      "tier2": "MiniMax-M2.5",
+      "tier3": "MiniMax-M1",
+      "tier4": "MiniMax-M1"
+    },
+    "vlm": {
+      "tier1": "",
+      "tier2": "",
+      "tier3": ""
+    }
+  },
+
+  "openrouter": {
+    "conductor": "anthropic/claude-sonnet-4-20250514",
+    "llm": {
+      "tier1": "anthropic/claude-sonnet-4-20250514",
+      "tier2": "google/gemini-2.5-flash",
+      "tier3": "google/gemini-2.5-flash",
+      "tier4": "google/gemini-2.5-flash"
+    },
+    "vlm": {
+      "tier1": "anthropic/claude-sonnet-4-20250514",
+      "tier2": "google/gemini-2.5-flash",
+      "tier3": "google/gemini-2.5-flash"
+    }
+  },
+
+  "bedrock": {
+    "conductor": "anthropic.claude-sonnet-4-20250514-v1:0",
+    "llm": {
+      "tier1": "anthropic.claude-sonnet-4-20250514-v1:0",
+      "tier2": "anthropic.claude-sonnet-4-20250514-v1:0",
+      "tier3": "anthropic.claude-haiku-4-5-20251001-v1:0",
+      "tier4": "anthropic.claude-haiku-4-5-20251001-v1:0"
+    },
+    "vlm": {
+      "tier1": "anthropic.claude-sonnet-4-20250514-v1:0",
+      "tier2": "anthropic.claude-haiku-4-5-20251001-v1:0",
+      "tier3": "anthropic.claude-haiku-4-5-20251001-v1:0"
+    }
+  },
+
+  "custom": {
+    "conductor": "",
+    "llm": {
+      "tier1": "",
+      "tier2": "",
+      "tier3": "",
+      "tier4": ""
+    },
+    "vlm": {
+      "tier1": "",
+      "tier2": "",
+      "tier3": ""
+    }
+  }
+}