@thanh01.pmt/curriculum-kit 1.0.8 → 1.0.10

package/dist/{provider-factory-DH3udYlN.d.ts → provider-factory-DxzOVEmh.d.cts}

@@ -9,7 +9,7 @@ interface ModelResolutionOptions {
 }
 declare const NVIDIA_MODELS: {
     readonly codingFast: "nvidia/nemotron-3.5-lightning-30b-a3b";
-    readonly reasoningUltra: "nvidia/nemotron-3-ultra-550b-a55b";
+    readonly reasoningUltra: "nvidia/nemotron-3-ultra-550b-a55b:free";
     readonly superMoE: "nvidia/nemotron-3-super-120b-a12b";
     readonly llamaNemotron: "nvidia/llama-3.1-nemotron-70b-instruct";
     readonly translation: "nvidia/riva-translate-4b-instruct-v2";

package/dist/{provider-factory-DH3udYlN.d.cts → provider-factory-DxzOVEmh.d.ts}

@@ -9,7 +9,7 @@ interface ModelResolutionOptions {
 }
 declare const NVIDIA_MODELS: {
     readonly codingFast: "nvidia/nemotron-3.5-lightning-30b-a3b";
-    readonly reasoningUltra: "nvidia/nemotron-3-ultra-550b-a55b";
+    readonly reasoningUltra: "nvidia/nemotron-3-ultra-550b-a55b:free";
     readonly superMoE: "nvidia/nemotron-3-super-120b-a12b";
     readonly llamaNemotron: "nvidia/llama-3.1-nemotron-70b-instruct";
     readonly translation: "nvidia/riva-translate-4b-instruct-v2";

package/dist/storage/index.d.cts

@@ -1,10 +1,10 @@
 import { I as ICurriculumStorage, a as SotDocument, L as LessonArtifactSummary, P as ProjectStatusReport, c as PipelineState, b as ArtifactLifecycleInfo } from '../types-BUJGYiep.cjs';
 export { A as ArtifactLifecycleState, Q as QualityAuditReport, R as ReviewRecord, S as SmartResumeContext, T as TaskLifecycleState } from '../types-BUJGYiep.cjs';
 import { SupabaseClient } from '@supabase/supabase-js';
-import { M as MilestoneCurriculumBundle } from '../milestoneBundleGenerator-DrBtHzBO.cjs';
+import { M as MilestoneCurriculumBundle } from '../milestoneBundleGenerator-CzSOeAzU.cjs';
 import '../schemas/index.cjs';
 import 'zod';
-import '../provider-factory-DH3udYlN.cjs';
+import '../provider-factory-DxzOVEmh.cjs';
 import 'ai';
 
 declare function computeContentHash(content: string): string;

package/dist/storage/index.d.ts

@@ -1,10 +1,10 @@
 import { I as ICurriculumStorage, a as SotDocument, L as LessonArtifactSummary, P as ProjectStatusReport, c as PipelineState, b as ArtifactLifecycleInfo } from '../types-BUJGYiep.js';
 export { A as ArtifactLifecycleState, Q as QualityAuditReport, R as ReviewRecord, S as SmartResumeContext, T as TaskLifecycleState } from '../types-BUJGYiep.js';
 import { SupabaseClient } from '@supabase/supabase-js';
-import { M as MilestoneCurriculumBundle } from '../milestoneBundleGenerator-CbXhs1CP.js';
+import { M as MilestoneCurriculumBundle } from '../milestoneBundleGenerator-Bkba8OPG.js';
 import '../schemas/index.js';
 import 'zod';
-import '../provider-factory-DH3udYlN.js';
+import '../provider-factory-DxzOVEmh.js';
 import 'ai';
 
 declare function computeContentHash(content: string): string;

package/dist/{streamRunner-C7j5LKon.d.cts → streamRunner-CcpmxOf1.d.cts}

@@ -15,6 +15,25 @@ interface StreamRunnerOptions {
     maxTokens?: number;
     temperature?: number;
     reasoningEffort?: 'off' | 'low' | 'medium' | 'high' | 'max';
+    /**
+     * Host-provided inference override (Vercel AI SDK gateway, test stubs, …).
+     * When set, ALL built-in provider routes (raw-fetch NVIDIA/OpenRouter/
+     * DeepSeek/DashScope/Gemini + designated fallback chain) are bypassed and
+     * every inference call goes through this function instead. It receives the
+     * fully-formed system instruction (persona + project context + guidelines)
+     * and the user prompt, plus the same onChunk contract as the built-in
+     * routes: stream text deltas via ('content'|'thought'), and report token
+     * spend via onChunk(JSON.stringify({ promptTokens, completionTokens,
+     * totalTokens, reasoningTokens }), 'usage'). Return the full completion text.
+     */
+    customInference?: (request: {
+        systemInstruction: string;
+        userPrompt: string;
+        messages: ChatMessage[];
+        temperature: number;
+        maxTokens: number;
+        onChunk: (token: string, type: ChunkType) => void;
+    }) => Promise<string>;
 }
 /**
  * Checks if a specific AI provider is enabled via environment variables.
@@ -36,7 +55,7 @@ interface FallbackTarget {
  * Resolves the designated last-resort fallback chain from environment variables.
  * Comma-separated lists are parsed in left-to-right priority order:
  * - FALLBACK_AI_PROVIDER=nvidia,openrouter (Must be enabled / in ALLOWED_AI_PROVIDERS)
- * - FALLBACK_AI_MODEL=nvidia/nemotron-3-ultra-550b-a55b,@preset/coding-free,deepseek/deepseek-chat:free
+ * - FALLBACK_AI_MODEL=nvidia/nemotron-3-ultra-550b-a55b:free,@preset/coding-free,deepseek/deepseek-chat:free
  */
 declare function getDesignatedFallbackChain(): FallbackTarget[];
 declare function getDesignatedFallbackConfig(): FallbackTarget;

package/dist/{streamRunner-C7j5LKon.d.ts → streamRunner-CcpmxOf1.d.ts}

@@ -15,6 +15,25 @@ interface StreamRunnerOptions {
     maxTokens?: number;
     temperature?: number;
     reasoningEffort?: 'off' | 'low' | 'medium' | 'high' | 'max';
+    /**
+     * Host-provided inference override (Vercel AI SDK gateway, test stubs, …).
+     * When set, ALL built-in provider routes (raw-fetch NVIDIA/OpenRouter/
+     * DeepSeek/DashScope/Gemini + designated fallback chain) are bypassed and
+     * every inference call goes through this function instead. It receives the
+     * fully-formed system instruction (persona + project context + guidelines)
+     * and the user prompt, plus the same onChunk contract as the built-in
+     * routes: stream text deltas via ('content'|'thought'), and report token
+     * spend via onChunk(JSON.stringify({ promptTokens, completionTokens,
+     * totalTokens, reasoningTokens }), 'usage'). Return the full completion text.
+     */
+    customInference?: (request: {
+        systemInstruction: string;
+        userPrompt: string;
+        messages: ChatMessage[];
+        temperature: number;
+        maxTokens: number;
+        onChunk: (token: string, type: ChunkType) => void;
+    }) => Promise<string>;
 }
 /**
  * Checks if a specific AI provider is enabled via environment variables.
@@ -36,7 +55,7 @@ interface FallbackTarget {
  * Resolves the designated last-resort fallback chain from environment variables.
  * Comma-separated lists are parsed in left-to-right priority order:
  * - FALLBACK_AI_PROVIDER=nvidia,openrouter (Must be enabled / in ALLOWED_AI_PROVIDERS)
- * - FALLBACK_AI_MODEL=nvidia/nemotron-3-ultra-550b-a55b,@preset/coding-free,deepseek/deepseek-chat:free
+ * - FALLBACK_AI_MODEL=nvidia/nemotron-3-ultra-550b-a55b:free,@preset/coding-free,deepseek/deepseek-chat:free
  */
 declare function getDesignatedFallbackChain(): FallbackTarget[];
 declare function getDesignatedFallbackConfig(): FallbackTarget;

package/dist/workflow/index.cjs

@@ -191,7 +191,7 @@ function isModelAllowed(model) {
 }
 function getDesignatedFallbackChain() {
   const rawProviders = resolveApiKey("FALLBACK_AI_PROVIDER") || resolveApiKey("FALLBACK_AI_PROVIDERS") || resolveApiKey("DEFAULT_AI_PROVIDER") || "nvidia,openrouter";
-  const rawModels = resolveApiKey("FALLBACK_AI_MODEL") || resolveApiKey("FALLBACK_AI_MODELS") || resolveApiKey("DEFAULT_AI_MODEL") || "nvidia/nemotron-3-ultra-550b-a55b,@preset/coding-free,deepseek/deepseek-chat:free";
+  const rawModels = resolveApiKey("FALLBACK_AI_MODEL") || resolveApiKey("FALLBACK_AI_MODELS") || resolveApiKey("DEFAULT_AI_MODEL") || "nvidia/nemotron-3-ultra-550b-a55b:free,@preset/coding-free,deepseek/deepseek-chat:free";
   const providerList = rawProviders.split(",").map((p) => p.trim().toLowerCase()).filter((p) => p.length > 0 && isProviderEnabled(p));
   const modelList = rawModels.split(",").map((m) => m.trim()).filter((m) => m.length > 0 && isModelAllowed(m));
   const chain = [];
@@ -200,7 +200,7 @@ function getDesignatedFallbackChain() {
     if (provider === "nvidia") {
       providerModels = modelList.filter((m) => m.startsWith("nvidia/") || m.startsWith("meta/") || m.startsWith("mistralai/"));
       if (providerModels.length === 0) {
-        providerModels = ["nvidia/nemotron-3-ultra-550b-a55b", "nvidia/nemotron-3.5-lightning-30b-a3b"];
+        providerModels = ["nvidia/nemotron-3-ultra-550b-a55b:free", "nvidia/nemotron-3.5-lightning-30b-a3b"];
       }
     } else if (provider === "openrouter") {
       providerModels = modelList.filter((m) => m.startsWith("@") || m.includes("/") || m.includes(":"));
@@ -236,7 +236,7 @@ function getDesignatedFallbackChain() {
 function getDesignatedFallbackConfig() {
   const chain = getDesignatedFallbackChain();
   if (chain.length > 0) return chain[0];
-  return { provider: "nvidia", model: "nvidia/nemotron-3-ultra-550b-a55b" };
+  return { provider: "nvidia", model: "nvidia/nemotron-3-ultra-550b-a55b:free" };
 }
 function extractThoughtAndContent(rawText) {
   let text = rawText || "";
@@ -429,6 +429,32 @@ Guidelines:
 3. Keep technical terminology and official API symbols intact.
 4. Format output using clean Markdown, clear sections, bullet points, and code blocks where required.
 5. Do NOT use emojis in headings, tab titles, or UI navigation labels.`;
+  if (options.customInference) {
+    try {
+      const text = await options.customInference({
+        systemInstruction,
+        userPrompt: messages.map((m) => m.content).join("\n\n"),
+        messages,
+        temperature,
+        maxTokens,
+        onChunk: onChunk ?? (() => {
+        })
+      });
+      if (text && text.trim()) return text;
+      throw createAiInferenceError({
+        errorCode: "ERR_AI_ALL_PROVIDERS_FAILED",
+        message: "customInference returned empty output",
+        rawError: { customInference: "empty response" }
+      });
+    } catch (e) {
+      if (e?.errorCode || e?.name === "AiInferenceError") throw e;
+      throw createAiInferenceError({
+        errorCode: "ERR_AI_ALL_PROVIDERS_FAILED",
+        message: `customInference failed: ${e?.message || e}`,
+        rawError: { customInference: String(e?.message || e) }
+      });
+    }
+  }
   const formattedMessages = [
     { role: "system", content: systemInstruction },
     ...messages.map((m) => ({
@@ -441,8 +467,8 @@ Guidelines:
   const isNvidiaPreferred = requestedModel.startsWith("nvidia/") || Boolean(nvidiaKey && !openrouterKey) && isProviderEnabled("nvidia");
   if (nvidiaKey && isProviderEnabled("nvidia")) {
     const candidateNemotronModels = [
-      requestedModel.startsWith("nvidia/") ? requestedModel : "nvidia/nemotron-3-ultra-550b-a55b",
-      "nvidia/nemotron-3-ultra-550b-a55b",
+      requestedModel.startsWith("nvidia/") ? requestedModel : "nvidia/nemotron-3-ultra-550b-a55b:free",
+      "nvidia/nemotron-3-ultra-550b-a55b:free",
       "nvidia/nemotron-3.5-lightning-30b-a3b",
       "nvidia/nemotron-3-super-120b-a12b",
       "nvidia/llama-3.1-nemotron-70b-instruct"
@@ -804,7 +830,7 @@ var init_streamRunner = __esm({
 // src/ai/provider-factory.ts
 init_streamRunner();
 var DEFAULT_MODELS = {
-  nvidia: "nvidia/nemotron-3-ultra-550b-a55b",
+  nvidia: "nvidia/nemotron-3-ultra-550b-a55b:free",
   openrouter: "@preset/coding-free",
   alibaba: "qwen3.7-plus",
   dashscope: "qwen3.7-plus",
@@ -832,7 +858,7 @@ function getAIModel(options = {}) {
       apiKey,
       baseURL: options.baseURL || "https://integrate.api.nvidia.com/v1"
     });
-    return nvidia.chat(resolvedModelName || "nvidia/nemotron-3-ultra-550b-a55b");
+    return nvidia.chat(resolvedModelName || "nvidia/nemotron-3-ultra-550b-a55b:free");
   }
   if (provider === "openrouter") {
     const apiKey = options.apiKey || process.env.OPENROUTER_API_KEY || process.env.OPENAI_API_KEY || "";