@nebutra/agents 1.1.2 → 3.0.0

package/README.md

@@ -67,7 +67,7 @@ orchestrator.registerAgent(
   new VercelAIAgent({ id: "assistant", name: "Assistant", description: "Helpful", model: "openai/gpt-5.5", instructions: "..." }),
 );
 
-const ctx = createAgentContext("org_123", "user_456");
+const ctx = createAgentContext({ tenantId: "org_123", product: "app" }, "user_456");
 const response = await orchestrator.chat("Hello", ctx);
 // → response.usage tracks tokens for billing
 ```

package/dist/{agent-DDGWuUpe.d.ts → agent-DyyFxZf-.d.ts}

@@ -1,4 +1,4 @@
-import { c as AgentConfig, A as AgentMessage, a as AgentContext, b as AgentResponse } from './types-BytC-HfQ.js';
+import { c as AgentConfig, A as AgentMessage, a as AgentContext, b as AgentResponse } from './types-CRbPsTpE.js';
 
 /**
  * BaseAgent — abstract agent with memory, usage tracking, and tenant scoping.

package/dist/{chunk-QYZDHC5A.js → chunk-7VL333VJ.js}

@@ -1,7 +1,7 @@
 import {
   getAgentsEnv,
   isLangfuseConfigured
-} from "./chunk-BMSL4E4A.js";
+} from "./chunk-FCXWOXII.js";
 
 // src/observability.ts
 import { logger } from "@nebutra/logger";

package/dist/{chunk-XLBS3XUI.js → chunk-C4CC5UEF.js}

@@ -1,17 +1,22 @@
 import {
-  resolveModel
-} from "./chunk-BZKIMAGK.js";
+  resolveModel,
+  toAi302ModelId
+} from "./chunk-HVLFZW6E.js";
 import {
   getAgentsEnv
-} from "./chunk-BMSL4E4A.js";
+} from "./chunk-FCXWOXII.js";
 
 // src/fallback.ts
 import { logger } from "@nebutra/logger";
 var ENV_KEY_BY_PROVIDER = {
   openrouter: "OPENROUTER_API_KEY",
   anthropic: "ANTHROPIC".concat("_API_KEY"),
-  openai: "OPENAI".concat("_API_KEY")
+  openai: "OPENAI".concat("_API_KEY"),
+  ai302: "AI302_API_KEY"
 };
+function ai302BaseUrl() {
+  return globalThis.process?.env?.AI302_BASE_URL ?? "https://api.302.ai/v1";
+}
 function hasProviderKey(provider) {
   const k = ENV_KEY_BY_PROVIDER[provider];
   return Boolean(globalThis.process?.env?.[k]);
@@ -40,6 +45,10 @@ async function buildModel(provider, modelOrPreset) {
       const openaiModelId = modelId.startsWith("openai/") ? modelId.slice("openai/".length) : modelId;
       return createOpenAI({ apiKey })(openaiModelId);
     }
+    case "ai302": {
+      const { createOpenAI } = await import("@ai-sdk/openai");
+      return createOpenAI({ apiKey, baseURL: ai302BaseUrl() })(toAi302ModelId(modelId));
+    }
   }
 }
 var RETRYABLE_STATUS = /* @__PURE__ */ new Set([408, 425, 429, 500, 502, 503, 504]);
@@ -102,7 +111,10 @@ function buildSystemWithCache(systemPrompt) {
 }
 var EMBEDDING_CAPABLE = /* @__PURE__ */ new Set([
   "openrouter",
-  "openai"
+  "openai",
+  // 302.AI serves /v1/embeddings — an unauthenticated POST answers
+  // `Missing 302 Apikey` rather than 404, so the route exists behind auth.
+  "ai302"
 ]);
 async function buildEmbeddingModel(provider, modelOrPreset) {
   const modelId = resolveModel(modelOrPreset);
@@ -119,6 +131,12 @@ async function buildEmbeddingModel(provider, modelOrPreset) {
       const openaiModelId = modelId.startsWith("openai/") ? modelId.slice("openai/".length) : modelId;
       return createOpenAI({ apiKey }).textEmbeddingModel(openaiModelId);
     }
+    case "ai302": {
+      const { createOpenAI } = await import("@ai-sdk/openai");
+      return createOpenAI({ apiKey, baseURL: ai302BaseUrl() }).textEmbeddingModel(
+        toAi302ModelId(modelId)
+      );
+    }
     case "anthropic": {
       throw new Error("Anthropic does not expose embedding models");
     }

package/dist/{chunk-BMSL4E4A.js → chunk-FCXWOXII.js}

@@ -1,6 +1,6 @@
 // src/env.ts
 import { z } from "zod";
-var FallbackProviderName = z.enum(["openrouter", "anthropic", "openai"]);
+var FallbackProviderName = z.enum(["openrouter", "anthropic", "openai", "ai302"]);
 var FallbackChain = z.string().transform(
   (raw) => raw.split(",").map((s) => s.trim()).filter(Boolean)
 ).pipe(z.array(FallbackProviderName).min(1));

package/dist/{chunk-RDOFKRI6.js → chunk-HAAM3SJN.js}

@@ -107,6 +107,7 @@ var BaseAgent = class {
     const durationMs = Date.now() - startTime;
     const usage = {
       tenantId: context.tenantId,
+      product: context.product,
       userId: context.userId,
       agentId: this.config.id,
       model: this.config.model,
@@ -145,6 +146,7 @@ var BaseAgent = class {
       const creditCost = Math.max(1, Math.ceil(totalTokens / 1e4));
       await deductCredits({
         organizationId: event.tenantId,
+        product: event.product,
         amount: creditCost,
         description: `Agent execution: ${event.model}`
       });

package/dist/chunk-HVLFZW6E.js

@@ -0,0 +1,71 @@
+// src/sdk/frontier-fallback.generated.ts
+var FRONTIER_FALLBACK = {
+  reasoning: "anthropic/claude-opus-5",
+  flagship: "anthropic/claude-sonnet-5",
+  fast: "anthropic/claude-haiku-4.5",
+  "openai-flagship": "openai/gpt-5.6-sol",
+  "google-flagship": "google/gemini-3.1-pro-preview",
+  "google-fast": "google/gemini-3.7-flash"
+};
+var AI302_ALIASES = {
+  fast: "claude-haiku-4-5-20251001"
+};
+var AI302_OPEN_MODELS = {
+  "302-deepseek": "deepseek-v4-pro",
+  "302-deepseek-fast": "deepseek-v4-flash",
+  "302-qwen": "qwen3.8-max",
+  "302-glm": "glm-5.3",
+  "302-kimi": "kimi-k3",
+  "302-minimax": "MiniMax-M3"
+};
+
+// src/sdk/models.ts
+var models = {
+  // ── Generated frontier tiers — edit via `pnpm gen:frontier-models` ──────────
+  ...FRONTIER_FALLBACK,
+  /** Embedding model */
+  embedding: "openai/text-embedding-3-small",
+  /** Embedding model (high-dimensional) */
+  "embedding-large": "openai/text-embedding-3-large",
+  // --- Open-weight families via 302.AI (use with provider: "ai302") ---
+  // Generated with the tiers above. These replaced three SiliconFlow presets
+  // naming Qwen2.5-72B, DeepSeek-R1 and DeepSeek-V3 — every one superseded,
+  // none with a caller anywhere in the repo, and none checkable without a
+  // SiliconFlow key. 302 serves the same families and lists its catalogue,
+  // so these are resolved rather than remembered.
+  ...AI302_OPEN_MODELS,
+  // --- SenseNova Token Plan presets (use with provider: "sensenova") ---
+  // Base URL: https://token.sensenova.cn/v1
+  // Docs: https://github.com/OpenSenseNova/SenseNova6.7/blob/main/API_CN.md
+  /** SenseNova 6.7 Flash-Lite — small multimodal agent (default for i18n CI) */
+  "sn-flash-lite": "sensenova-6.7-flash-lite",
+  /** DeepSeek V4 Flash via SenseNova Token Plan — long context, cheap */
+  "sn-deepseek-flash": "deepseek-v4-flash",
+  /** Alias: prefer flash-lite for bulk translation */
+  "sn-translate": "sensenova-6.7-flash-lite"
+};
+function resolveModel(modelOrPreset) {
+  if (modelOrPreset in models) {
+    return models[modelOrPreset];
+  }
+  return modelOrPreset;
+}
+var AI302_BY_GATEWAY_ID = Object.fromEntries(
+  Object.entries(AI302_ALIASES).map(([tier, id]) => [
+    FRONTIER_FALLBACK[tier],
+    id
+  ])
+);
+function toAi302ModelId(modelOrPreset) {
+  const modelId = resolveModel(modelOrPreset);
+  const alias = AI302_BY_GATEWAY_ID[modelId];
+  if (alias) return alias;
+  const slash = modelId.indexOf("/");
+  return slash === -1 ? modelId : modelId.slice(slash + 1);
+}
+
+export {
+  models,
+  resolveModel,
+  toAi302ModelId
+};

package/dist/{chunk-RWQL4HXC.js → chunk-HZQXXUKB.js}

@@ -1,6 +1,6 @@
 import {
   isRetryableError
-} from "./chunk-XLBS3XUI.js";
+} from "./chunk-C4CC5UEF.js";
 
 // src/generation/index.ts
 import { logger } from "@nebutra/logger";

package/dist/{chunk-RNFUEMUB.js → chunk-KC6SOI5Z.js}

@@ -1,9 +1,10 @@
 import {
   resolveApiKey
-} from "./chunk-V6VC2O6Q.js";
+} from "./chunk-YBCIJKC7.js";
 import {
-  resolveModel
-} from "./chunk-BZKIMAGK.js";
+  resolveModel,
+  toAi302ModelId
+} from "./chunk-HVLFZW6E.js";
 
 // src/sdk/provider.ts
 import { createOpenAI } from "@ai-sdk/openai";
@@ -38,6 +39,13 @@ function createModel(modelOrPreset, config) {
       });
       return provider(modelId);
     }
+    case "ai302": {
+      const provider = createOpenAI({
+        apiKey,
+        baseURL: process.env.AI302_BASE_URL ?? "https://api.302.ai/v1"
+      });
+      return provider(toAi302ModelId(modelId));
+    }
     case "gateway": {
       const provider = createOpenAI({
         apiKey,

package/dist/{chunk-GRSMUTUS.js → chunk-S5N743EP.js}

@@ -1,16 +1,16 @@
 import {
   createModel
-} from "./chunk-RNFUEMUB.js";
+} from "./chunk-KC6SOI5Z.js";
 import {
   NebutraAIConfigSchema
-} from "./chunk-V6VC2O6Q.js";
+} from "./chunk-YBCIJKC7.js";
 import {
   assertSafeOpenAIJsonPayload
 } from "./chunk-VZPQOXWW.js";
 import {
   runEmbedWithFallback,
   runWithFallback
-} from "./chunk-XLBS3XUI.js";
+} from "./chunk-C4CC5UEF.js";
 
 // src/sdk/index.ts
 import {

package/dist/{chunk-V6VC2O6Q.js → chunk-YBCIJKC7.js}

@@ -1,13 +1,28 @@
+import {
+  models
+} from "./chunk-HVLFZW6E.js";
+
 // src/sdk/config.ts
 import { z } from "zod";
-var ProviderType = z.enum(["openrouter", "openai", "siliconflow", "sensenova", "gateway"]);
+var ProviderType = z.enum([
+  "openrouter",
+  "openai",
+  "siliconflow",
+  "sensenova",
+  "ai302",
+  "gateway"
+]);
 var NebutraAIConfigSchema = z.object({
   /** Which provider backend to use. Defaults to "openrouter". */
   provider: ProviderType.default("openrouter"),
   /** API key override. Falls back to env vars per provider. */
   apiKey: z.string().optional(),
-  /** Default model id — OpenRouter / models.dev frontier (not Claude 3.x / GPT-4 era). */
-  defaultModel: z.string().default("anthropic/claude-sonnet-4.6"),
+  /**
+   * Default model id. Reads the generated frontier flagship rather than naming
+   * a version, so `pnpm gen:frontier-models` moves it and it cannot go stale
+   * here independently of everywhere else.
+   */
+  defaultModel: z.string().default(models.flagship),
   /** Default temperature for generations. */
   temperature: z.number().min(0).max(2).default(0.7),
   /** Default max tokens for output. */
@@ -24,6 +39,7 @@ function resolveApiKey(config) {
     openai: "OPENAI_API_KEY",
     siliconflow: "SILICONFLOW_API_KEY",
     sensenova: "SENSENOVA_API_KEY",
+    ai302: "AI302_API_KEY",
     gateway: "VERCEL_OIDC_TOKEN"
   };
   const envVar = envMap[config.provider];

package/dist/env.d.ts

@@ -13,6 +13,7 @@ declare const FallbackProviderName: z.ZodEnum<{
     openrouter: "openrouter";
     anthropic: "anthropic";
     openai: "openai";
+    ai302: "ai302";
 }>;
 type FallbackProviderName = z.infer<typeof FallbackProviderName>;
 declare const AgentsEnvSchema: z.ZodObject<{
@@ -24,11 +25,13 @@ declare const AgentsEnvSchema: z.ZodObject<{
         openrouter: "openrouter";
         anthropic: "anthropic";
         openai: "openai";
+        ai302: "ai302";
     }>>>>;
     LLM_EMBEDDING_FALLBACK_CHAIN: z.ZodDefault<z.ZodPipe<z.ZodPipe<z.ZodString, z.ZodTransform<string[], string>>, z.ZodArray<z.ZodEnum<{
         openrouter: "openrouter";
         anthropic: "anthropic";
         openai: "openai";
+        ai302: "ai302";
     }>>>>;
 }, z.core.$strip>;
 type AgentsEnv = z.infer<typeof AgentsEnvSchema>;

package/dist/env.js

@@ -3,7 +3,7 @@ import {
   _resetAgentsEnvCache,
   getAgentsEnv,
   isLangfuseConfigured
-} from "./chunk-BMSL4E4A.js";
+} from "./chunk-FCXWOXII.js";
 export {
   AgentsEnvSchema,
   _resetAgentsEnvCache,

package/dist/fallback.js

@@ -5,9 +5,9 @@ import {
   runEmbedWithFallback,
   runWithFallback,
   withAnthropicCacheControl
-} from "./chunk-XLBS3XUI.js";
-import "./chunk-BZKIMAGK.js";
-import "./chunk-BMSL4E4A.js";
+} from "./chunk-C4CC5UEF.js";
+import "./chunk-HVLFZW6E.js";
+import "./chunk-FCXWOXII.js";
 export {
   buildSystemWithCache,
   filterAvailableProviders,

package/dist/generation/index.js

@@ -5,10 +5,10 @@ import {
   listGenerationProviders,
   mockGenerationProvider,
   registerGenerationProvider
-} from "../chunk-RWQL4HXC.js";
-import "../chunk-XLBS3XUI.js";
-import "../chunk-BZKIMAGK.js";
-import "../chunk-BMSL4E4A.js";
+} from "../chunk-HZQXXUKB.js";
+import "../chunk-C4CC5UEF.js";
+import "../chunk-HVLFZW6E.js";
+import "../chunk-FCXWOXII.js";
 export {
   _resetGenerationRegistry,
   generateImage,

package/dist/index.d.ts

@@ -1,14 +1,14 @@
-import { B as BaseAgent } from './agent-DDGWuUpe.js';
+import { B as BaseAgent } from './agent-DyyFxZf-.js';
 export { AgentsEnv, AgentsEnvSchema, FallbackProviderName, getAgentsEnv, isLangfuseConfigured } from './env.js';
 export { CreateFallbackModelOptions, EmbeddingFallbackOptions, FallbackResult, buildSystemWithCache, filterAvailableProviders, isRetryableError, runEmbedWithFallback, runWithFallback, withAnthropicCacheControl } from './fallback.js';
 export { GenerationCallOptions, GenerationContext, GenerationModality, GenerationProvider, GenerationResult, ImageGenerationRequest, VideoGenerationRequest, _resetGenerationRegistry, generateImage, generateVideo, listGenerationProviders, mockGenerationProvider, registerGenerationProvider } from './generation/index.js';
-import { A as AgentMessage, O as OrchestratorConfig, a as AgentContext, b as AgentResponse, R as RouterConfig, c as AgentConfig } from './types-BytC-HfQ.js';
-export { d as AgentTool, e as AgentUsageEvent, M as MemoryConfig, T as TokenUsage, f as ToolCallResult } from './types-BytC-HfQ.js';
+import { A as AgentMessage, O as OrchestratorConfig, a as AgentContext, b as AgentResponse, R as RouterConfig, c as AgentConfig } from './types-CRbPsTpE.js';
+export { d as AgentTool, e as AgentUsageEvent, M as MemoryConfig, T as TokenUsage, f as ToolCallResult } from './types-CRbPsTpE.js';
 export { TelemetryMetadata, buildTelemetryConfig, flushTelemetry, initLangfuse } from './observability.js';
 export { EmbedOptions, GenerateOptions, GenerateStructuredResult, OPENAI_MAX_PROPERTY_NAME_LENGTH, assertSafeOpenAIJsonPayload, configure, embed, embedMany, findOversizedPropertyName, generateStructured, generateText, getConfig, streamText, validateStructured } from './sdk/index.js';
 export { BUILT_IN_TOOLS, databaseQueryTool, knowledgeBaseTool, webSearchTool } from './tools.js';
 export { GenerateTextResult, ModelMessage, StreamTextResult } from 'ai';
-export { ModelPreset, models, resolveModel } from './sdk/models.js';
+export { ModelPreset, models, resolveModel, toAi302ModelId } from './sdk/models.js';
 export { NebutraAIConfig, NebutraAIConfigSchema, ProviderType, ResolvedNebutraAIConfig } from './sdk/config.js';
 export { createEmbeddingModel, createModel } from './sdk/provider.js';
 import 'zod';
@@ -153,15 +153,21 @@ declare class AgentRouter {
 /**
  * Create a fully-populated AgentContext.
  * Generates a random conversationId when none is provided.
+ *
+ * `scope` names the tenant and the product whose balance the run spends:
+ * balances are per product, so a run cannot be billed without one.
  */
-declare function createAgentContext(tenantId: string, userId: string, conversationId?: string, metadata?: Record<string, unknown>): AgentContext;
+declare function createAgentContext(scope: {
+    tenantId: string;
+    product: string;
+}, userId: string, conversationId?: string, metadata?: Record<string, unknown>): AgentContext;
 /**
  * Validate that a tenant has remaining quota for agent execution.
  *
  * Returns `{ allowed: true, remaining: -1 }` (unlimited) by default.
  * Integrate with @nebutra/billing entitlements for production usage.
  */
-declare function checkAgentQuota(tenantId: string): Promise<{
+declare function checkAgentQuota(tenantId: string, product: string): Promise<{
     allowed: boolean;
     remaining: number;
 }>;

package/dist/index.js

@@ -7,14 +7,14 @@ import {
   getConfig,
   streamText,
   validateStructured
-} from "./chunk-GRSMUTUS.js";
+} from "./chunk-S5N743EP.js";
 import {
   createEmbeddingModel,
   createModel
-} from "./chunk-RNFUEMUB.js";
+} from "./chunk-KC6SOI5Z.js";
 import {
   NebutraAIConfigSchema
-} from "./chunk-V6VC2O6Q.js";
+} from "./chunk-YBCIJKC7.js";
 import {
   BUILT_IN_TOOLS,
   databaseQueryTool,
@@ -28,7 +28,7 @@ import {
   listGenerationProviders,
   mockGenerationProvider,
   registerGenerationProvider
-} from "./chunk-RWQL4HXC.js";
+} from "./chunk-HZQXXUKB.js";
 import {
   OPENAI_MAX_PROPERTY_NAME_LENGTH,
   assertSafeOpenAIJsonPayload,
@@ -38,7 +38,7 @@ import {
   buildTelemetryConfig,
   flushTelemetry,
   initLangfuse
-} from "./chunk-QYZDHC5A.js";
+} from "./chunk-7VL333VJ.js";
 import {
   buildSystemWithCache,
   filterAvailableProviders,
@@ -46,22 +46,23 @@ import {
   runEmbedWithFallback,
   runWithFallback,
   withAnthropicCacheControl
-} from "./chunk-XLBS3XUI.js";
+} from "./chunk-C4CC5UEF.js";
 import {
   models,
-  resolveModel
-} from "./chunk-BZKIMAGK.js";
+  resolveModel,
+  toAi302ModelId
+} from "./chunk-HVLFZW6E.js";
 import {
   AgentsEnvSchema,
   getAgentsEnv,
   isLangfuseConfigured
-} from "./chunk-BMSL4E4A.js";
+} from "./chunk-FCXWOXII.js";
 import {
   BaseAgent,
   clearMemory,
   getMemory,
   saveMemory
-} from "./chunk-RDOFKRI6.js";
+} from "./chunk-HAAM3SJN.js";
 
 // src/context.ts
 var MAX_BIO_CHARS = 2e3;
@@ -228,18 +229,19 @@ ${agentList}`,
 };
 
 // src/tenant.ts
-function createAgentContext(tenantId, userId, conversationId, metadata) {
+function createAgentContext(scope, userId, conversationId, metadata) {
   return {
-    tenantId,
+    tenantId: scope.tenantId,
+    product: scope.product,
     userId,
     conversationId: conversationId ?? crypto.randomUUID(),
     ...metadata !== void 0 ? { metadata } : {}
   };
 }
-async function checkAgentQuota(tenantId) {
+async function checkAgentQuota(tenantId, product) {
   try {
     const { getCreditBalance } = await import("@nebutra/billing/credits");
-    const balance = await getCreditBalance(tenantId);
+    const balance = await getCreditBalance(tenantId, product);
     if (balance.balance > 0) {
       return { allowed: true, remaining: balance.balance };
     }
@@ -275,7 +277,7 @@ var AgentOrchestrator = class {
   }
   /** Route a message to the best agent and execute. */
   async chat(message, context) {
-    await this.assertQuota(context.tenantId);
+    await this.assertQuota(context.tenantId, context.product);
     const agentConfigs = [...this.agents.values()].map((a) => a.config);
     const agentId = await this.router.route(message, agentConfigs, context, this.defaultAgentId);
     const agent = this.agents.get(agentId);
@@ -286,8 +288,8 @@ var AgentOrchestrator = class {
     return agent.run(messages, context);
   }
   /** Check tenant quota before execution. */
-  async assertQuota(tenantId) {
-    const { allowed } = await checkAgentQuota(tenantId);
+  async assertQuota(tenantId, product) {
+    const { allowed } = await checkAgentQuota(tenantId, product);
     if (!allowed) {
       throw new Error(`Tenant "${tenantId}" has exceeded agent execution quota`);
     }
@@ -339,6 +341,7 @@ export {
   runWithFallback,
   saveMemory,
   streamText,
+  toAi302ModelId,
   validateStructured,
   webSearchTool,
   withAnthropicCacheControl

package/dist/observability.js

@@ -3,8 +3,8 @@ import {
   buildTelemetryConfig,
   flushTelemetry,
   initLangfuse
-} from "./chunk-QYZDHC5A.js";
-import "./chunk-BMSL4E4A.js";
+} from "./chunk-7VL333VJ.js";
+import "./chunk-FCXWOXII.js";
 export {
   _resetLangfuseCache,
   buildTelemetryConfig,

package/dist/providers/langchain.d.ts

@@ -1,5 +1,5 @@
-import { B as BaseAgent } from '../agent-DDGWuUpe.js';
-import { c as AgentConfig } from '../types-BytC-HfQ.js';
+import { B as BaseAgent } from '../agent-DyyFxZf-.js';
+import { c as AgentConfig } from '../types-CRbPsTpE.js';
 
 /**
  * LangChain.js agent adapter — stub.

package/dist/providers/langchain.js

@@ -1,6 +1,6 @@
 import {
   BaseAgent
-} from "../chunk-RDOFKRI6.js";
+} from "../chunk-HAAM3SJN.js";
 
 // src/providers/langchain.ts
 var LangChainAgent = class extends BaseAgent {

package/dist/providers/vercel-ai.d.ts

@@ -1,5 +1,5 @@
-import { B as BaseAgent } from '../agent-DDGWuUpe.js';
-import { A as AgentMessage, a as AgentContext, b as AgentResponse } from '../types-BytC-HfQ.js';
+import { B as BaseAgent } from '../agent-DyyFxZf-.js';
+import { A as AgentMessage, a as AgentContext, b as AgentResponse } from '../types-CRbPsTpE.js';
 
 /**
  * Vercel AI SDK agent adapter.

package/dist/providers/vercel-ai.js

@@ -3,16 +3,16 @@ import {
 } from "../chunk-VZPQOXWW.js";
 import {
   buildTelemetryConfig
-} from "../chunk-QYZDHC5A.js";
+} from "../chunk-7VL333VJ.js";
 import {
   runWithFallback,
   withAnthropicCacheControl
-} from "../chunk-XLBS3XUI.js";
-import "../chunk-BZKIMAGK.js";
-import "../chunk-BMSL4E4A.js";
+} from "../chunk-C4CC5UEF.js";
+import "../chunk-HVLFZW6E.js";
+import "../chunk-FCXWOXII.js";
 import {
   BaseAgent
-} from "../chunk-RDOFKRI6.js";
+} from "../chunk-HAAM3SJN.js";
 
 // src/providers/vercel-ai.ts
 var VercelAIAgent = class extends BaseAgent {

package/dist/sdk/config.d.ts

@@ -16,11 +16,13 @@ import { z } from 'zod';
  * - openai:       Direct OpenAI API access
  * - siliconflow:  SiliconFlow cloud — Qwen, DeepSeek, etc. (OpenAI-compatible, China-optimized)
  * - sensenova:    商汤 SenseNova — OpenAI-compatible (`compatible-mode/v1`)
+ * - ai302:        302.AI aggregator — OpenAI-compatible, broad model catalogue
  * - gateway:      Vercel AI Gateway with OIDC auth (for Vercel-deployed apps)
  */
 declare const ProviderType: z.ZodEnum<{
     openrouter: "openrouter";
     openai: "openai";
+    ai302: "ai302";
     siliconflow: "siliconflow";
     sensenova: "sensenova";
     gateway: "gateway";
@@ -30,6 +32,7 @@ declare const NebutraAIConfigSchema: z.ZodObject<{
     provider: z.ZodDefault<z.ZodEnum<{
         openrouter: "openrouter";
         openai: "openai";
+        ai302: "ai302";
         siliconflow: "siliconflow";
         sensenova: "sensenova";
         gateway: "gateway";

package/dist/sdk/config.js

@@ -2,7 +2,8 @@ import {
   NebutraAIConfigSchema,
   ProviderType,
   resolveApiKey
-} from "../chunk-V6VC2O6Q.js";
+} from "../chunk-YBCIJKC7.js";
+import "../chunk-HVLFZW6E.js";
 export {
   NebutraAIConfigSchema,
   ProviderType,

package/dist/sdk/index.d.ts

@@ -3,7 +3,7 @@ import { ModelMessage, JSONValue, GenerateTextResult, StreamTextResult } from 'a
 export { GenerateTextResult, ModelMessage, StreamTextResult } from 'ai';
 import { NebutraAIConfig, ResolvedNebutraAIConfig } from './config.js';
 export { NebutraAIConfigSchema, ProviderType } from './config.js';
-export { ModelPreset, models, resolveModel } from './models.js';
+export { ModelPreset, models, resolveModel, toAi302ModelId } from './models.js';
 export { createEmbeddingModel, createModel } from './provider.js';
 import 'zod';
 

package/dist/sdk/index.js

@@ -7,25 +7,26 @@ import {
   getConfig,
   streamText,
   validateStructured
-} from "../chunk-GRSMUTUS.js";
+} from "../chunk-S5N743EP.js";
 import {
   createEmbeddingModel,
   createModel
-} from "../chunk-RNFUEMUB.js";
+} from "../chunk-KC6SOI5Z.js";
 import {
   NebutraAIConfigSchema
-} from "../chunk-V6VC2O6Q.js";
+} from "../chunk-YBCIJKC7.js";
 import {
   OPENAI_MAX_PROPERTY_NAME_LENGTH,
   assertSafeOpenAIJsonPayload,
   findOversizedPropertyName
 } from "../chunk-VZPQOXWW.js";
-import "../chunk-XLBS3XUI.js";
+import "../chunk-C4CC5UEF.js";
 import {
   models,
-  resolveModel
-} from "../chunk-BZKIMAGK.js";
-import "../chunk-BMSL4E4A.js";
+  resolveModel,
+  toAi302ModelId
+} from "../chunk-HVLFZW6E.js";
+import "../chunk-FCXWOXII.js";
 export {
   NebutraAIConfigSchema,
   OPENAI_MAX_PROPERTY_NAME_LENGTH,
@@ -42,5 +43,6 @@ export {
   models,
   resolveModel,
   streamText,
+  toAi302ModelId,
   validateStructured
 };

package/dist/sdk/models.d.ts

@@ -1,46 +1,26 @@
-/**
- * Model presets for common Nebutra use cases.
- *
- * All IDs use "vendor/model" format (OpenRouter / SiliconFlow).
- * When using direct OpenAI provider, only OpenAI models are valid.
- * When using Vercel AI Gateway, use "provider/model" format.
- * When using SiliconFlow, use "Vendor/Model" format (e.g. "Qwen/Qwen2.5-72B-Instruct").
- *
- * These hardcoded ids are the FALLBACK tier of the hybrid model strategy — they
- * are the current Pareto frontier (audited 2026-06-05 against OpenRouter ∩
- * models.dev). For always-fresh resolution use `resolveFrontierModel(tier)` from
- * `@nebutra/ai-providers/catalog`, which picks the newest routable model per tier
- * at runtime and falls back to exactly these values when offline.
- */
 declare const models: {
-    /** High-quality reasoning — default for complex tasks */
-    readonly flagship: "anthropic/claude-sonnet-4.6";
-    /** Deep reasoning for architecture and research */
-    readonly reasoning: "anthropic/claude-opus-4.8";
-    /** Fast + cheap — chat, summaries, classification */
-    readonly fast: "anthropic/claude-haiku-4.5";
-    /** OpenAI flagship */
-    readonly "openai-flagship": "openai/gpt-5.5";
-    /** Google flagship */
-    readonly "google-flagship": "google/gemini-3.1-pro-preview";
-    /** Google fast */
-    readonly "google-fast": "google/gemini-3.5-flash";
-    /** Embedding model */
-    readonly embedding: "openai/text-embedding-3-small";
-    /** Embedding model (high-dimensional) */
-    readonly "embedding-large": "openai/text-embedding-3-large";
-    /** SiliconFlow — Qwen 2.5 72B (flagship open-source) */
-    readonly "sf-qwen": "Qwen/Qwen2.5-72B-Instruct";
-    /** SiliconFlow — DeepSeek R1 reasoning (via SiliconFlow Pro) */
-    readonly "sf-deepseek-r1": "Pro/deepseek-ai/DeepSeek-R1";
-    /** SiliconFlow — DeepSeek V3 (fast, capable) */
-    readonly "sf-deepseek-v3": "deepseek-ai/DeepSeek-V3";
     /** SenseNova 6.7 Flash-Lite — small multimodal agent (default for i18n CI) */
     readonly "sn-flash-lite": "sensenova-6.7-flash-lite";
     /** DeepSeek V4 Flash via SenseNova Token Plan — long context, cheap */
     readonly "sn-deepseek-flash": "deepseek-v4-flash";
     /** Alias: prefer flash-lite for bulk translation */
     readonly "sn-translate": "sensenova-6.7-flash-lite";
+    readonly "302-deepseek": "deepseek-v4-pro";
+    readonly "302-deepseek-fast": "deepseek-v4-flash";
+    readonly "302-qwen": "qwen3.8-max";
+    readonly "302-glm": "glm-5.3";
+    readonly "302-kimi": "kimi-k3";
+    readonly "302-minimax": "MiniMax-M3";
+    /** Embedding model */
+    readonly embedding: "openai/text-embedding-3-small";
+    /** Embedding model (high-dimensional) */
+    readonly "embedding-large": "openai/text-embedding-3-large";
+    readonly reasoning: "anthropic/claude-opus-5";
+    readonly flagship: "anthropic/claude-sonnet-5";
+    readonly fast: "anthropic/claude-haiku-4.5";
+    readonly "openai-flagship": "openai/gpt-5.6-sol";
+    readonly "google-flagship": "google/gemini-3.1-pro-preview";
+    readonly "google-fast": "google/gemini-3.7-flash";
 };
 type ModelPreset = keyof typeof models;
 /**
@@ -48,5 +28,6 @@ type ModelPreset = keyof typeof models;
  * If the input is not a preset key, returns it as-is (passthrough).
  */
 declare function resolveModel(modelOrPreset: string): string;
+declare function toAi302ModelId(modelOrPreset: string): string;
 
-export { type ModelPreset, models, resolveModel };
+export { type ModelPreset, models, resolveModel, toAi302ModelId };

package/dist/sdk/models.js

@@ -1,8 +1,10 @@
 import {
   models,
-  resolveModel
-} from "../chunk-BZKIMAGK.js";
+  resolveModel,
+  toAi302ModelId
+} from "../chunk-HVLFZW6E.js";
 export {
   models,
-  resolveModel
+  resolveModel,
+  toAi302ModelId
 };

package/dist/sdk/provider.js

@@ -1,9 +1,9 @@
 import {
   createEmbeddingModel,
   createModel
-} from "../chunk-RNFUEMUB.js";
-import "../chunk-V6VC2O6Q.js";
-import "../chunk-BZKIMAGK.js";
+} from "../chunk-KC6SOI5Z.js";
+import "../chunk-YBCIJKC7.js";
+import "../chunk-HVLFZW6E.js";
 export {
   createEmbeddingModel,
   createModel

package/dist/tools.d.ts

@@ -1,4 +1,4 @@
-import { d as AgentTool } from './types-BytC-HfQ.js';
+import { d as AgentTool } from './types-CRbPsTpE.js';
 
 /**
  * Built-in tool registry for agents.

package/dist/{types-BytC-HfQ.d.ts → types-CRbPsTpE.d.ts}

@@ -30,6 +30,8 @@ interface AgentTool {
 }
 interface AgentContext {
     readonly tenantId: string;
+    /** The product whose balance pays for this run (ADR 2026-09-27 product wallets). */
+    readonly product: string;
     readonly userId: string;
     readonly conversationId: string;
     readonly metadata?: Record<string, unknown>;
@@ -75,6 +77,7 @@ interface RouterConfig {
 }
 interface AgentUsageEvent {
     readonly tenantId: string;
+    readonly product: string;
     readonly userId: string;
     readonly agentId: string;
     readonly model: string;

package/package.json

@@ -1,11 +1,13 @@
 {
   "name": "@nebutra/agents",
-  "version": "1.1.2",
+  "version": "3.0.0",
   "description": "Nebutra AI runtime: multi-agent orchestration + Vercel AI SDK helpers (absorbed @nebutra/ai-sdk in 1.0.0)",
   "private": false,
   "license": "MIT",
   "type": "module",
   "nebutra": {
+    "status": "wip",
+    "graph": "runtime",
     "featureId": "agents",
     "category": "ai",
     "surface": "model-runtime",
@@ -84,9 +86,9 @@
     "langfuse": "^3.38.20",
     "langfuse-vercel": "^3.38.20",
     "zod": "^4.3.6",
-    "@nebutra/billing": "0.1.3",
-    "@nebutra/cache": "0.0.3",
-    "@nebutra/logger": "0.1.2"
+    "@nebutra/billing": "3.0.0",
+    "@nebutra/cache": "3.0.0",
+    "@nebutra/logger": "3.0.0"
   },
   "devDependencies": {
     "@types/node": "^25.9.1",

package/dist/chunk-BZKIMAGK.js

@@ -1,46 +0,0 @@
-// src/sdk/models.ts
-var models = {
-  /** High-quality reasoning — default for complex tasks */
-  flagship: "anthropic/claude-sonnet-4.6",
-  /** Deep reasoning for architecture and research */
-  reasoning: "anthropic/claude-opus-4.8",
-  /** Fast + cheap — chat, summaries, classification */
-  fast: "anthropic/claude-haiku-4.5",
-  /** OpenAI flagship */
-  "openai-flagship": "openai/gpt-5.5",
-  /** Google flagship */
-  "google-flagship": "google/gemini-3.1-pro-preview",
-  /** Google fast */
-  "google-fast": "google/gemini-3.5-flash",
-  /** Embedding model */
-  embedding: "openai/text-embedding-3-small",
-  /** Embedding model (high-dimensional) */
-  "embedding-large": "openai/text-embedding-3-large",
-  // --- SiliconFlow presets (use with provider: "siliconflow") ---
-  /** SiliconFlow — Qwen 2.5 72B (flagship open-source) */
-  "sf-qwen": "Qwen/Qwen2.5-72B-Instruct",
-  /** SiliconFlow — DeepSeek R1 reasoning (via SiliconFlow Pro) */
-  "sf-deepseek-r1": "Pro/deepseek-ai/DeepSeek-R1",
-  /** SiliconFlow — DeepSeek V3 (fast, capable) */
-  "sf-deepseek-v3": "deepseek-ai/DeepSeek-V3",
-  // --- SenseNova Token Plan presets (use with provider: "sensenova") ---
-  // Base URL: https://token.sensenova.cn/v1
-  // Docs: https://github.com/OpenSenseNova/SenseNova6.7/blob/main/API_CN.md
-  /** SenseNova 6.7 Flash-Lite — small multimodal agent (default for i18n CI) */
-  "sn-flash-lite": "sensenova-6.7-flash-lite",
-  /** DeepSeek V4 Flash via SenseNova Token Plan — long context, cheap */
-  "sn-deepseek-flash": "deepseek-v4-flash",
-  /** Alias: prefer flash-lite for bulk translation */
-  "sn-translate": "sensenova-6.7-flash-lite"
-};
-function resolveModel(modelOrPreset) {
-  if (modelOrPreset in models) {
-    return models[modelOrPreset];
-  }
-  return modelOrPreset;
-}
-
-export {
-  models,
-  resolveModel
-};