@yourgpt/llm-sdk 2.5.0 → 2.5.1-beta.1

package/dist/index.d.ts

@@ -1,13 +1,13 @@
-import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, d as StorageAdapter, e as StorageMessage } from './types-CR8mi9I0.js';
-export { A as AssistantMessage, C as CoreMessage, w as DEFAULT_CAPABILITIES, D as DoGenerateParams, f as DoGenerateResult, E as ErrorChunk, F as FilePart, s as FinishChunk, u as FinishReason, m as GenerateStep, I as ImagePart, L as LanguageModel, M as ModelCapabilities, R as ResponseOptions, v as StorageFile, o as StreamChunk, n as StreamPart, g as SystemMessage, p as TextDeltaChunk, j as TextPart, t as TokenUsage, k as ToolCall, q as ToolCallChunk, h as ToolMessage, l as ToolResult, r as ToolResultChunk, i as UserContentPart, U as UserMessage } from './types-CR8mi9I0.js';
+import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, A as ActionDefinition, d as ToolDefinition, e as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, f as StorageAdapter, R as ResponseFormat, M as McpServerConfig, g as ReasoningEffort, D as DoneEventMessage, h as StreamEvent, i as ToolCallInfo, j as TokenUsageRaw, P as ProviderToolRuntimeOptions, k as Message, l as ToolResponse, m as StorageMessage } from './types-D774b0dg.js';
+export { a6 as AnthropicProviderToolOptions, a3 as AnthropicToolSelectionHints, r as AssistantMessage, a7 as Citation, C as CoreMessage, Y as DEFAULT_CAPABILITIES, o as DoGenerateParams, p as DoGenerateResult, N as ErrorChunk, F as FilePart, J as FinishChunk, Q as FinishReason, x as GenerateStep, I as ImagePart, Z as LLMConfig, L as LanguageModel, n as ModelCapabilities, a5 as OpenAIProviderToolOptions, a2 as OpenAIToolSelectionHints, V as ResponseOptions, X as StorageFile, z as StreamChunk, y as StreamPart, q as SystemMessage, B as TextDeltaChunk, u as TextPart, O as TokenUsage, v as ToolCall, E as ToolCallChunk, a1 as ToolExecution, _ as ToolLocation, s as ToolMessage, a4 as ToolNativeProviderHints, w as ToolResult, H as ToolResultChunk, $ as UnifiedToolCall, a0 as UnifiedToolResult, t as UserContentPart, U as UserMessage } from './types-D774b0dg.js';
 import { z } from 'zod';
-import { A as ActionDefinition, T as ToolDefinition, a as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, L as LLMAdapter, D as DoneEventMessage, S as StreamEvent, b as ToolCallInfo, c as TokenUsageRaw, P as ProviderToolRuntimeOptions, M as Message, d as ToolResponse } from './base-iGi9Va6Z.js';
-export { e as AdapterFactory, m as AnthropicProviderToolOptions, j as AnthropicToolSelectionHints, C as ChatCompletionRequest, n as Citation, f as LLMConfig, l as OpenAIProviderToolOptions, O as OpenAIToolSelectionHints, i as ToolExecution, g as ToolLocation, k as ToolNativeProviderHints, U as UnifiedToolCall, h as UnifiedToolResult } from './base-iGi9Va6Z.js';
-import { A as AIProvider } from './types-BctsnC3g.js';
-export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-BctsnC3g.js';
+import { A as AIProvider } from './types-TMilS-Dz.js';
+export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-TMilS-Dz.js';
+import { L as LLMAdapter } from './base-Cxq3ni0t.js';
+export { A as AdapterFactory, C as ChatCompletionRequest } from './base-Cxq3ni0t.js';
 import * as hono from 'hono';
 import { Hono } from 'hono';
-export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-38yolWJn.js';
+export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-ChORafYS.js';
 
 /**
  * generateText - Generate text using a language model
@@ -388,6 +388,9 @@ interface ChatRequest {
     config?: {
         temperature?: number;
         maxTokens?: number;
+        responseFormat?: ResponseFormat;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
     };
     /** System prompt override */
     systemPrompt?: string;
@@ -797,6 +800,11 @@ declare function createStreamResult(generator: AsyncGenerator<StreamEvent>): Str
  * ```
  */
 
+interface GenerateUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
 interface GenerateResultData {
     text: string;
     messages: DoneEventMessage[];
@@ -810,6 +818,7 @@ interface GenerateResultData {
         result: unknown;
     }>;
     requiresAction: boolean;
+    usage?: GenerateUsage;
     error?: {
         message: string;
         code?: string;
@@ -863,6 +872,12 @@ declare class GenerateResult {
      * Whether client action is required (e.g., tool approval)
      */
     get requiresAction(): boolean;
+    /**
+     * Token usage for the call — present when the underlying provider emitted
+     * a usage payload on the `done` event. May be undefined for streaming
+     * responses that don't yield usage (e.g. mid-stream errors).
+     */
+    get usage(): GenerateUsage | undefined;
     /**
      * Error if generation failed
      */
@@ -1122,6 +1137,44 @@ declare class Runtime {
      * ```
      */
     generate(request: ChatRequest, options?: GenerateOptions): Promise<GenerateResult>;
+    /**
+     * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+     * structured output. Thin ergonomic wrapper around `generate()` — uses the
+     * same adapter chain and the same translators, just expressed as a single
+     * prompt-in / text-out call.
+     *
+     * @example
+     * ```ts
+     * const result = await runtime.response({
+     *   prompt: "Extract FAQs from this conversation.",
+     *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+     *   reasoningEffort: "high",
+     *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+     * });
+     * const data = JSON.parse(result.text);
+     * ```
+     */
+    response(request: {
+        prompt: string;
+        systemPrompt?: string;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
+        responseFormat?: ResponseFormat;
+        maxTokens?: number;
+        temperature?: number;
+    }): Promise<{
+        text: string;
+        toolCalls: Array<{
+            id: string;
+            name: string;
+            args: Record<string, unknown>;
+        }>;
+        usage?: {
+            promptTokens: number;
+            completionTokens: number;
+            totalTokens: number;
+        };
+    }>;
     /**
      * Create Express-compatible handler middleware
      *
@@ -1536,4 +1589,4 @@ declare function generateThreadId(): string;
  */
 declare function generateToolCallId(): string;
 
-export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, Message, ProviderToolRuntimeOptions, type RequestContext, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };
+export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, McpServerConfig, Message, ProviderToolRuntimeOptions, ReasoningEffort, type RequestContext, ResponseFormat, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };

package/dist/index.js

@@ -135,6 +135,11 @@ function formatToolsForGoogle(tools) {
 // src/core/generate-text.ts
 async function generateText(params) {
   const { model, tools, maxSteps = 1, signal } = params;
+  if (params.responseFormat && model.capabilities.supportsJsonMode === false) {
+    console.warn(
+      `[llm-sdk] ${model.provider}/${model.modelId} does not support structured output (responseFormat); the request will be sent but the provider may ignore it.`
+    );
+  }
   let messages = buildMessages(params);
   const steps = [];
   const allToolCalls = [];
@@ -149,6 +154,9 @@ async function generateText(params) {
       tools: formattedTools,
       temperature: params.temperature,
       maxTokens: params.maxTokens,
+      responseFormat: params.responseFormat,
+      mcpServers: params.mcpServers,
+      reasoningEffort: params.reasoningEffort,
       signal
     });
     const stepToolResults = [];
@@ -266,6 +274,11 @@ function sumUsage(steps) {
 // src/core/stream-text.ts
 async function streamText(params) {
   const { model, tools, maxSteps = 1, signal } = params;
+  if (params.responseFormat && model.capabilities.supportsJsonMode === false) {
+    console.warn(
+      `[llm-sdk] ${model.provider}/${model.modelId} does not support structured output (responseFormat); the request will be sent but the provider may ignore it.`
+    );
+  }
   let fullText = "";
   let finalUsage = {
     promptTokens: 0,
@@ -291,6 +304,9 @@ async function streamText(params) {
           tools: formattedTools,
           temperature: params.temperature,
           maxTokens: params.maxTokens,
+          responseFormat: params.responseFormat,
+          mcpServers: params.mcpServers,
+          reasoningEffort: params.reasoningEffort,
           signal
         })) {
           switch (chunk.type) {
@@ -1159,6 +1175,14 @@ var GenerateResult = class {
   get requiresAction() {
     return this.data.requiresAction;
   }
+  /**
+   * Token usage for the call — present when the underlying provider emitted
+   * a usage payload on the `done` event. May be undefined for streaming
+   * responses that don't yield usage (e.g. mid-stream errors).
+   */
+  get usage() {
+    return this.data.usage;
+  }
   /**
    * Error if generation failed
    */
@@ -2945,6 +2969,7 @@ var Runtime = class {
     const toolResults = [];
     let messages = [];
     let requiresAction = false;
+    let usage;
     let error;
     try {
       for await (const event of generator) {
@@ -2975,6 +3000,13 @@ var Runtime = class {
           case "done":
             messages = event.messages || [];
             requiresAction = event.requiresAction || false;
+            if (event.usage) {
+              usage = {
+                promptTokens: event.usage.prompt_tokens,
+                completionTokens: event.usage.completion_tokens,
+                totalTokens: event.usage.total_tokens ?? event.usage.prompt_tokens + event.usage.completion_tokens
+              };
+            }
             break;
           case "error":
             error = { message: event.message, code: event.code };
@@ -3006,9 +3038,48 @@ var Runtime = class {
       toolCalls,
       toolResults,
       requiresAction,
+      usage,
       error
     });
   }
+  /**
+   * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+   * structured output. Thin ergonomic wrapper around `generate()` — uses the
+   * same adapter chain and the same translators, just expressed as a single
+   * prompt-in / text-out call.
+   *
+   * @example
+   * ```ts
+   * const result = await runtime.response({
+   *   prompt: "Extract FAQs from this conversation.",
+   *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+   *   reasoningEffort: "high",
+   *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+   * });
+   * const data = JSON.parse(result.text);
+   * ```
+   */
+  async response(request) {
+    const result = await this.generate({
+      messages: [{ role: "user", content: request.prompt }],
+      systemPrompt: request.systemPrompt,
+      config: {
+        temperature: request.temperature,
+        maxTokens: request.maxTokens,
+        responseFormat: request.responseFormat,
+        mcpServers: request.mcpServers,
+        reasoningEffort: request.reasoningEffort
+      }
+    });
+    if (result.error) {
+      throw new Error(`[llm-sdk] response() failed: ${result.error.message}`);
+    }
+    return {
+      text: result.text,
+      toolCalls: result.toolCalls,
+      usage: result.usage
+    };
+  }
   /**
    * Create Express-compatible handler middleware
    *

package/dist/index.mjs

@@ -133,6 +133,11 @@ function formatToolsForGoogle(tools) {
 // src/core/generate-text.ts
 async function generateText(params) {
   const { model, tools, maxSteps = 1, signal } = params;
+  if (params.responseFormat && model.capabilities.supportsJsonMode === false) {
+    console.warn(
+      `[llm-sdk] ${model.provider}/${model.modelId} does not support structured output (responseFormat); the request will be sent but the provider may ignore it.`
+    );
+  }
   let messages = buildMessages(params);
   const steps = [];
   const allToolCalls = [];
@@ -147,6 +152,9 @@ async function generateText(params) {
       tools: formattedTools,
       temperature: params.temperature,
       maxTokens: params.maxTokens,
+      responseFormat: params.responseFormat,
+      mcpServers: params.mcpServers,
+      reasoningEffort: params.reasoningEffort,
       signal
     });
     const stepToolResults = [];
@@ -264,6 +272,11 @@ function sumUsage(steps) {
 // src/core/stream-text.ts
 async function streamText(params) {
   const { model, tools, maxSteps = 1, signal } = params;
+  if (params.responseFormat && model.capabilities.supportsJsonMode === false) {
+    console.warn(
+      `[llm-sdk] ${model.provider}/${model.modelId} does not support structured output (responseFormat); the request will be sent but the provider may ignore it.`
+    );
+  }
   let fullText = "";
   let finalUsage = {
     promptTokens: 0,
@@ -289,6 +302,9 @@ async function streamText(params) {
           tools: formattedTools,
           temperature: params.temperature,
           maxTokens: params.maxTokens,
+          responseFormat: params.responseFormat,
+          mcpServers: params.mcpServers,
+          reasoningEffort: params.reasoningEffort,
           signal
         })) {
           switch (chunk.type) {
@@ -1157,6 +1173,14 @@ var GenerateResult = class {
   get requiresAction() {
     return this.data.requiresAction;
   }
+  /**
+   * Token usage for the call — present when the underlying provider emitted
+   * a usage payload on the `done` event. May be undefined for streaming
+   * responses that don't yield usage (e.g. mid-stream errors).
+   */
+  get usage() {
+    return this.data.usage;
+  }
   /**
    * Error if generation failed
    */
@@ -2943,6 +2967,7 @@ var Runtime = class {
     const toolResults = [];
     let messages = [];
     let requiresAction = false;
+    let usage;
     let error;
     try {
       for await (const event of generator) {
@@ -2973,6 +2998,13 @@ var Runtime = class {
           case "done":
             messages = event.messages || [];
             requiresAction = event.requiresAction || false;
+            if (event.usage) {
+              usage = {
+                promptTokens: event.usage.prompt_tokens,
+                completionTokens: event.usage.completion_tokens,
+                totalTokens: event.usage.total_tokens ?? event.usage.prompt_tokens + event.usage.completion_tokens
+              };
+            }
             break;
           case "error":
             error = { message: event.message, code: event.code };
@@ -3004,9 +3036,48 @@ var Runtime = class {
       toolCalls,
       toolResults,
       requiresAction,
+      usage,
       error
     });
   }
+  /**
+   * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+   * structured output. Thin ergonomic wrapper around `generate()` — uses the
+   * same adapter chain and the same translators, just expressed as a single
+   * prompt-in / text-out call.
+   *
+   * @example
+   * ```ts
+   * const result = await runtime.response({
+   *   prompt: "Extract FAQs from this conversation.",
+   *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+   *   reasoningEffort: "high",
+   *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+   * });
+   * const data = JSON.parse(result.text);
+   * ```
+   */
+  async response(request) {
+    const result = await this.generate({
+      messages: [{ role: "user", content: request.prompt }],
+      systemPrompt: request.systemPrompt,
+      config: {
+        temperature: request.temperature,
+        maxTokens: request.maxTokens,
+        responseFormat: request.responseFormat,
+        mcpServers: request.mcpServers,
+        reasoningEffort: request.reasoningEffort
+      }
+    });
+    if (result.error) {
+      throw new Error(`[llm-sdk] response() failed: ${result.error.message}`);
+    }
+    return {
+      text: result.text,
+      toolCalls: result.toolCalls,
+      usage: result.usage
+    };
+  }
   /**
    * Create Express-compatible handler middleware
    *

package/dist/providers/anthropic/index.d.mts

@@ -1,7 +1,7 @@
-import { L as LanguageModel } from '../../types-CR8mi9I0.mjs';
-import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-DRqxMIjF.mjs';
+import { L as LanguageModel } from '../../types-D774b0dg.mjs';
+import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-BvkiJ1dd.mjs';
 import 'zod';
-import '../../base-D-U61JaB.mjs';
+import '../../base-BYQKp9TW.mjs';
 
 /**
  * Anthropic Provider - Modern Pattern

package/dist/providers/anthropic/index.d.ts

@@ -1,7 +1,7 @@
-import { L as LanguageModel } from '../../types-CR8mi9I0.js';
-import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-BctsnC3g.js';
+import { L as LanguageModel } from '../../types-D774b0dg.js';
+import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-TMilS-Dz.js';
 import 'zod';
-import '../../base-iGi9Va6Z.js';
+import '../../base-Cxq3ni0t.js';
 
 /**
  * Anthropic Provider - Modern Pattern