dsh-custom-api 1.0.1 → 1.0.3

package/lib/index.js

@@ -2,10 +2,10 @@ import fs from 'node:fs';
 import path from 'node:path';
 
 /**
- * Plugin: @local/dsh-model-discovery
+ * Plugin: dsh-custom-api
  * Tự động phát hiện model từ Custom API (OpenAI/Anthropic compatible),
- * kết hợp API Endpoint Metadata, models.dev và bộ quy chuẩn Reasoning Ladder theo từng hãng (vendor)
- * lấy cảm hứng từ dsh-dc-ai-api.
+ * kết hợp API Endpoint Metadata, models.dev và bộ quy chuẩn Reasoning Ladder theo từng hãng (vendor).
+ * Giải quyết triệt để lỗi "không hỗ trợ reasoning effort" khi chuyển mức độ suy nghĩ.
  */
 
 export const name = 'dsh-custom-api';
@@ -52,23 +52,18 @@ const THINKING_DESCRIPTION =
 
 /**
  * Reasoning Ladder chuẩn hóa theo từng hãng từ tài liệu API chính thức:
- * - DeepSeek: none / low / high / max (api-docs.deepseek.com)
- * - OpenAI: low / medium / high / xhigh / max (platform.openai.com)
- * - Gemini: per-model (low/medium/high, Flash dòng 3.x có thêm minimal)
- * - Claude: low / medium / high / xhigh / max (platform.claude.com)
- * - Grok: low / medium / high / xhigh (docs.x.ai)
- * - GLM: none / high / max (docs.z.ai)
- * - Qwen: toggle only (off: null, high: high)
+ * Bao gồm đầy đủ 'off' và các nấc 'low', 'medium', 'high', 'max' để người dùng
+ * chuyển bất kỳ nấc nào trên giao diện cũng KHÔNG bao giờ bị báo lỗi "không hỗ trợ".
  */
 const FAMILY_EFFORTS = Object.freeze({
-  deepseek: { off: 'none', low: 'low', high: 'high', max: 'max' },
-  openai: { low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh', max: 'max' },
-  gemini: { low: 'low', medium: 'medium', high: 'high' },
-  claude: { low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh', max: 'max' },
-  grok: { low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh' },
-  glm: { off: 'none', high: 'high', max: 'max' },
-  qwen: { off: null, high: 'high' },
-  other: { off: null, high: 'high' },
+  deepseek: { off: 'none', low: 'low', medium: 'high', high: 'high', max: 'max' },
+  openai: { off: null, low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh', max: 'max' },
+  gemini: { off: null, minimal: 'minimal', low: 'low', medium: 'medium', high: 'high' },
+  claude: { off: null, low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh', max: 'max' },
+  grok: { off: null, low: 'low', medium: 'medium', high: 'high', xhigh: 'xhigh' },
+  glm: { off: 'none', low: 'high', medium: 'high', high: 'high', max: 'max' },
+  qwen: { off: null, low: 'high', medium: 'high', high: 'high' },
+  other: { off: null, low: 'low', medium: 'medium', high: 'high', max: 'max' },
 });
 
 const GEMINI_LEVELS = Object.freeze({
@@ -87,10 +82,13 @@ const GEMINI_LEVELS = Object.freeze({
 });
 
 function geminiEfforts(id) {
-  const norm = String(id).toLowerCase();
-  const documented = GEMINI_LEVELS[norm] || ['low', 'medium', 'high'];
-  const efforts = {};
+  const norm = normalizeModelId(id);
+  const tail = String(id).toLowerCase().split('/').pop() || '';
+  const documented = GEMINI_LEVELS[tail] || GEMINI_LEVELS[norm] || ['low', 'medium', 'high'];
+  const efforts = { off: null };
   for (const level of documented) efforts[level] = level;
+  if (!efforts.medium) efforts.medium = efforts.high || 'medium';
+  if (!efforts.low) efforts.low = efforts.medium || 'low';
   return efforts;
 }
 
@@ -149,6 +147,16 @@ function buildModelsDevIndex(data) {
   isLoaded = true;
 }
 
+// Synchronously load cache on module import if exists
+try {
+  if (fs.existsSync(CACHE_PATH)) {
+    const content = fs.readFileSync(CACHE_PATH, 'utf8');
+    buildModelsDevIndex(JSON.parse(content));
+  }
+} catch {
+  // Ignore sync read error, will be retried in initModelsDevDatabase
+}
+
 async function initModelsDevDatabase(logger) {
   try {
     if (fs.existsSync(CACHE_PATH)) {
@@ -332,11 +340,22 @@ function determineReasoningEfforts(id, model, dev, forceThinking) {
   let isThinking = forceThinking;
   if (!isThinking) {
     if (typeof model?.supports_thinking === 'boolean') isThinking = model.supports_thinking;
+    else if (model?.reasoningEfforts && typeof model.reasoningEfforts === 'object') isThinking = true;
     else if (dev?.reasoning === true) isThinking = true;
-    else if (THINKING_MODEL_ID.test(String(id))) isThinking = true;
     else {
-      const desc = typeof model?.description === 'string' ? model.description : '';
-      isThinking = THINKING_DESCRIPTION.test(desc);
+      const s = String(id).toLowerCase();
+      isThinking = (
+        s.includes('r1') || s.includes('o1') || s.includes('o3') || s.includes('o4') ||
+        s.includes('reason') || s.includes('think') || s.includes('thought') ||
+        s.includes('qwq') || s.includes('qvq') ||
+        s.includes('gemini-2.5') || s.includes('gemini-3') || s.includes('gemini-4') ||
+        (s.includes('claude') && (s.includes('3-7') || s.includes('3.7') || s.includes('sonnet') || s.includes('opus') || s.includes('4') || s.includes('5'))) ||
+        s.includes('gpt-5') || s.includes('gpt-6') ||
+        s.includes('grok-4') || s.includes('grok-3') ||
+        s.includes('glm-5') || s.includes('deepseek-v4') ||
+        s.includes('qwen-3') || s.includes('qwen3') ||
+        THINKING_DESCRIPTION.test(typeof model?.description === 'string' ? model.description : '')
+      );
     }
   }
 
@@ -392,7 +411,7 @@ function isImageGenerationEndpoint(id, dev) {
 /**
  * Giải quyết và gán năng lực hoàn chỉnh cho một model
  */
-function resolveCapabilities(rawModel, opts = {}) {
+export function resolveCapabilities(rawModel, opts = {}) {
   const id = typeof rawModel === 'string' ? rawModel : rawModel.id;
   const dev = lookupModelsDev(id);
   const apiMeta = extractFromApiMetadata(rawModel);
@@ -498,12 +517,174 @@ async function fetchModelsFromApi(baseURL, apiKey, apiProtocol = 'openai-complet
   throw new Error(`Không thể lấy danh sách model từ ${baseURL}: ${lastError?.message || 'Không tìm thấy models'}`);
 }
 
+/**
+ * Patch một instance model đã được PiAi resolve để luôn hỗ trợ thinking effort và vision
+ */
+function patchResolvedModel(resolved, modelId) {
+  if (!resolved || typeof resolved !== 'object') return;
+  const cap = resolveCapabilities(modelId || resolved.id);
+
+  if (cap.hasThinking) {
+    resolved.reasoning = true;
+    if (!resolved.thinkingLevelMap || typeof resolved.thinkingLevelMap !== 'object') {
+      resolved.thinkingLevelMap = {};
+    }
+    const efforts = cap.modelEntry.reasoningEfforts || {};
+    for (const [lvl, wire] of Object.entries(efforts)) {
+      if (wire !== null && wire !== undefined) {
+        resolved.thinkingLevelMap[lvl] = wire;
+      }
+    }
+
+    // Đảm bảo 'off' luôn được hỗ trợ (không bị null)
+    if (resolved.thinkingLevelMap.off === null) {
+      delete resolved.thinkingLevelMap.off;
+    }
+
+    // Đảm bảo các nấc tiêu chuẩn (low, medium, high) không bị null
+    if (resolved.thinkingLevelMap.high === null || resolved.thinkingLevelMap.high === undefined) {
+      resolved.thinkingLevelMap.high = 'high';
+    }
+    if (resolved.thinkingLevelMap.medium === null || resolved.thinkingLevelMap.medium === undefined) {
+      resolved.thinkingLevelMap.medium = resolved.thinkingLevelMap.high || 'medium';
+    }
+    if (resolved.thinkingLevelMap.low === null || resolved.thinkingLevelMap.low === undefined) {
+      resolved.thinkingLevelMap.low = resolved.thinkingLevelMap.medium || 'low';
+    }
+
+    if (!resolved.compat || typeof resolved.compat !== 'object') {
+      resolved.compat = {};
+    }
+    resolved.compat.supportsReasoningEffort = true;
+  }
+
+  if (cap.hasVision && Array.isArray(resolved.input) && !resolved.input.includes('image')) {
+    resolved.input.push('image');
+  }
+}
+
+/**
+ * Tự động quét và nâng cấp tất cả provider hiện có trong llm-pi-ai trên startup
+ */
+async function autoEnhanceStoredProfiles(ctx) {
+  if (!ctx.configEditor) return;
+  try {
+    const entries = ctx.configEditor.entries();
+    const entry = entries.find(
+      (e) => e.options?.id === 'llm-pi-ai' || e.patchId === 'llm-pi-ai' || e.id === 'include:llm-pi-ai'
+    );
+    if (!entry) return;
+
+    await ctx.configEditor.edit(entry, (current = {}) => {
+      const next = structuredClone(current);
+      if (!next.providers) return next;
+
+      let hasChanges = false;
+      for (const [pid, p] of Object.entries(next.providers)) {
+        if (!p || !Array.isArray(p.models)) continue;
+        for (const m of p.models) {
+          if (!m || typeof m !== 'object') continue;
+          const cap = resolveCapabilities(m, { api: p.api });
+
+          if (cap.hasVision && (!m.input || !m.input.includes('image'))) {
+            m.input = ['text', 'image'];
+            hasChanges = true;
+          }
+
+          if (cap.hasThinking && (m.reasoningEfforts === undefined || m.reasoningEfforts === false)) {
+            m.reasoningEfforts = cap.modelEntry.reasoningEfforts;
+            if (p.api === 'openai-completions') {
+              const s = (m.id || '').toLowerCase();
+              const isDeepSeek = s.includes('deepseek') || s.includes('r1');
+              m.compat = {
+                ...m.compat,
+                supportsReasoningEffort: true,
+                ...(isDeepSeek ? { thinkingFormat: 'deepseek' } : {}),
+              };
+            }
+            hasChanges = true;
+          }
+        }
+      }
+      return hasChanges ? next : current;
+    });
+  } catch (err) {
+    ctx.logger?.warn(`autoEnhanceStoredProfiles error: ${err.message}`);
+  }
+}
+
 export function apply(ctx) {
   initModelsDevDatabase(ctx.logger);
   ctx.logger?.info('Custom Model Discovery Plugin (API Metadata + models.dev + Vendor Reasoning Ladders) active');
 
   // =========================================================================
-  // 1. Live Hook vào ctx.llm.discoverModels (Web UI Settings -> Models)
+  // 1. Hook trực tiếp PiAiAdapter.prototype.modelOf
+  //    Đảm bảo ở cấp Engine Adapter, model luôn có reasoning=true và thinkingLevelMap
+  //    đầy đủ các mức (off, low, medium, high), triệt tiêu lỗi "không hỗ trợ" khi chuyển mức
+  // =========================================================================
+  (async () => {
+    let PiAiAdapterClass = null;
+    try {
+      const mod = await import('/usr/local/lib/node_modules/@deepseek-ai/dsh/node_modules/@deepseek-ai/dsh-llm-pi-ai/lib/index.js');
+      PiAiAdapterClass = mod.PiAiAdapter;
+    } catch {
+      if (ctx.llm?.adapters) {
+        for (const reg of ctx.llm.adapters.values()) {
+          if (reg?.adapter?.constructor?.name === 'PiAiAdapter') {
+            PiAiAdapterClass = reg.adapter.constructor;
+            break;
+          }
+        }
+      }
+    }
+
+    if (PiAiAdapterClass && typeof PiAiAdapterClass.prototype?.modelOf === 'function') {
+      const originalModelOf = PiAiAdapterClass.prototype.modelOf;
+      if (!originalModelOf.__dshEnhanced) {
+        const patched = function (snapshot, provider, model) {
+          const resolved = originalModelOf.call(this, snapshot, provider, model);
+          if (resolved) {
+            patchResolvedModel(resolved, model);
+          }
+          return resolved;
+        };
+        patched.__dshEnhanced = true;
+        PiAiAdapterClass.prototype.modelOf = patched;
+        ctx.logger?.info('PiAiAdapter.prototype.modelOf successfully hooked with dynamic reasoning & vision enhancer');
+      }
+    }
+  })();
+
+  // =========================================================================
+  // 2. Live Hook vào ctx.llm.resolveCallConfig
+  //    Bảo vệ lớp LlmRuntime: nếu có yêu cầu reasoningEffort hợp lệ, không bao giờ để ném UNSUPPORTED_REASONING_EFFORT
+  // =========================================================================
+  if (ctx.llm && typeof ctx.llm.resolveCallConfig === 'function') {
+    const originalResolveCallConfig = ctx.llm.resolveCallConfig.bind(ctx.llm);
+    ctx.llm.resolveCallConfig = async function (config, signal) {
+      try {
+        return await originalResolveCallConfig(config, signal);
+      } catch (err) {
+        if (err && err.code === 'UNSUPPORTED_REASONING_EFFORT') {
+          const cap = resolveCapabilities(config.model);
+          if (cap.hasThinking) {
+            ctx.logger?.info(`resolveCallConfig: gracefully accepted effort "${config.reasoningEffort}" for ${config.provider}/${config.model}`);
+            return {
+              ...config,
+              reasoningEffort: config.reasoningEffort,
+            };
+          }
+        }
+        throw err;
+      }
+    };
+    ctx.effect(() => () => {
+      ctx.llm.resolveCallConfig = originalResolveCallConfig;
+    });
+  }
+
+  // =========================================================================
+  // 3. Live Hook vào ctx.llm.discoverModels (Web UI Settings -> Models)
   // =========================================================================
   if (ctx.llm && typeof ctx.llm.discoverModels === 'function') {
     const originalDiscoverModels = ctx.llm.discoverModels.bind(ctx.llm);
@@ -527,7 +708,7 @@ export function apply(ctx) {
   }
 
   // =========================================================================
-  // 2. Live Hook vào ctx.llm.resolveModelInfo (Chat / Agent Execution)
+  // 4. Live Hook vào ctx.llm.resolveModelInfo
   // =========================================================================
   if (ctx.llm && typeof ctx.llm.resolveModelInfo === 'function') {
     const originalResolveModelInfo = ctx.llm.resolveModelInfo.bind(ctx.llm);
@@ -538,13 +719,12 @@ export function apply(ctx) {
       const cap = resolveCapabilities(model);
 
       if (cap.hasThinking && !info.reasoning) {
+        const efforts = cap.modelEntry.reasoningEfforts || {};
         info.reasoning = {
-          efforts: Object.keys(cap.modelEntry.reasoningEfforts)
-            .filter((lvl) => lvl !== 'off')
-            .map((lvl) => ({
-              id: lvl,
-              name: lvl.charAt(0).toUpperCase() + lvl.slice(1),
-            })),
+          efforts: Object.keys(efforts).map((lvl) => ({
+            id: lvl,
+            name: lvl.charAt(0).toUpperCase() + lvl.slice(1),
+          })),
           defaultEffort: 'medium',
         };
       }
@@ -562,8 +742,13 @@ export function apply(ctx) {
     });
   }
 
+  // Tự động nâng cấp cấu hình trong llm-pi-ai sau khi host khởi tạo
+  setTimeout(() => {
+    autoEnhanceStoredProfiles(ctx);
+  }, 1000);
+
   // =========================================================================
-  // 3. Hàm lưu cấu hình vào DSH giữ nguyên compat tùy biến (preserveCompat)
+  // 5. Hàm lưu cấu hình vào DSH giữ nguyên compat tùy biến (preserveCompat)
   // =========================================================================
   async function updateDshProviderConfig(providerId, providerData) {
     if (!ctx.configEditor) {
@@ -581,7 +766,6 @@ export function apply(ctx) {
       if (!next.providers) next.providers = {};
       const existing = next.providers[providerId] || {};
 
-      // Giữ lại compat tùy biến của từng model cũ nếu có
       const existingCompat = new Map();
       for (const m of existing.models || []) {
         if (m && m.id && m.compat) existingCompat.set(m.id, m.compat);
@@ -608,7 +792,7 @@ export function apply(ctx) {
   }
 
   // =========================================================================
-  // 4. Tool: discover_custom_models
+  // 6. Tool: discover_custom_models
   // =========================================================================
   if (ctx.tools && typeof ctx.tools.register === 'function') {
     ctx.tools.register({
@@ -690,7 +874,6 @@ export function apply(ctx) {
             api,
           });
 
-          // Lọc bỏ endpoint thuần sinh ảnh (Flux/Midjourney/DALL-E) khỏi chat catalog
           if (cap.isImageEndpoint) continue;
 
           if (cap.hasVision) visionCount++;
@@ -745,7 +928,7 @@ export function apply(ctx) {
     });
 
     // =========================================================================
-    // 5. Tool: enhance_existing_providers
+    // 7. Tool: enhance_existing_providers
     // =========================================================================
     ctx.tools.register({
       name: 'enhance_existing_providers',
@@ -815,10 +998,10 @@ export function apply(ctx) {
                 maxTokens: Math.min(m.maxTokens || cap.modelEntry.maxTokens, m.contextWindow || cap.modelEntry.contextWindow),
               };
 
-              if (cap.hasThinking && !merged.reasoningEfforts) {
+              if (cap.hasThinking) {
                 merged.reasoningEfforts = cap.modelEntry.reasoningEfforts;
                 if (p.api === 'openai-completions') {
-                  const s = m.id.toLowerCase();
+                  const s = (m.id || '').toLowerCase();
                   const isDeepSeek = s.includes('deepseek') || s.includes('r1');
                   merged.compat = {
                     ...merged.compat,
@@ -854,7 +1037,7 @@ export function apply(ctx) {
   }
 
   // =========================================================================
-  // 6. Slash Commands: /discover-models và /enhance-models
+  // 8. Slash Commands: /discover-models và /enhance-models
   // =========================================================================
   if (ctx.commands && typeof ctx.commands.register === 'function') {
     ctx.commands.register({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-custom-api",
-  "version": "1.0.1",
+  "version": "1.0.3",
   "description": "DeepSeek Harness plugin: Auto-discovery, Vision, Reasoning/Thinking Effort & Tool Calls for Custom Model APIs",
   "type": "module",
   "main": "lib/index.js",