@yourgpt/llm-sdk 2.5.1-beta.1 → 2.5.1-beta.3

package/dist/providers/anthropic/index.mjs

@@ -87,63 +87,6 @@ function toAnthropicOutputConfig(rf) {
     }
   };
 }
-function toAnthropicMcp(mcpServers) {
-  if (!mcpServers || mcpServers.length === 0) {
-    return { mcpServers: [], tools: [], betas: [] };
-  }
-  const serverEntries = [];
-  const toolEntries = [];
-  for (const mcp of mcpServers) {
-    const authHeader = mcp.headers?.Authorization ?? mcp.headers?.authorization;
-    const token = authHeader?.replace(/^Bearer\s+/i, "");
-    serverEntries.push({
-      type: "url",
-      url: mcp.url,
-      name: mcp.label,
-      ...token ? { authorization_token: token } : {}
-    });
-    if (mcp.allowedTools && mcp.allowedTools.length > 0) {
-      toolEntries.push({
-        type: "mcp_toolset",
-        mcp_server_name: mcp.label,
-        configs: Object.fromEntries(
-          mcp.allowedTools.map((toolName) => [toolName, {}])
-        )
-      });
-    }
-  }
-  return {
-    mcpServers: serverEntries,
-    tools: toolEntries,
-    betas: ["mcp-client-2025-11-20"]
-  };
-}
-function isStringEffort(effort) {
-  return typeof effort === "string" && (effort === "minimal" || effort === "low" || effort === "medium" || effort === "high");
-}
-var ANTHROPIC_ADAPTIVE_MODELS = /(claude-opus-4-7|claude-opus-4-6|claude-sonnet-4-6)/i;
-function toAnthropicThinking(effort, modelId) {
-  if (!effort) return {};
-  if (typeof effort === "object" && "raw" in effort) {
-    return { thinking: effort.raw };
-  }
-  const isAdaptive = !!modelId && ANTHROPIC_ADAPTIVE_MODELS.test(modelId);
-  if (typeof effort === "object" && "budgetTokens" in effort) {
-    return {
-      thinking: { type: "enabled", budget_tokens: effort.budgetTokens }
-    };
-  }
-  if (!isStringEffort(effort)) return {};
-  if (isAdaptive) {
-    const mapped = effort === "minimal" ? "low" : effort;
-    return {
-      thinking: { type: "adaptive" },
-      outputConfigEffort: mapped
-    };
-  }
-  const budget = effort === "high" ? 16e3 : effort === "medium" ? 8e3 : effort === "low" ? 4e3 : 2048;
-  return { thinking: { type: "enabled", budget_tokens: budget } };
-}
 function hasMediaAttachments(message) {
   const attachments = message.metadata?.attachments;
   return attachments?.some(
@@ -979,58 +922,36 @@ var AnthropicAdapter = class {
     if (serverToolConfiguration) {
       options.server_tool_configuration = serverToolConfiguration;
     }
-    const modelForThinking = request.config?.model || this.model;
-    const thinkingTranslation = toAnthropicThinking(
-      request.config?.reasoningEffort,
-      modelForThinking
-    );
     const outputConfig = toAnthropicOutputConfig(responseFormat);
-    if (outputConfig || thinkingTranslation.outputConfigEffort) {
-      options.output_config = {
-        ...outputConfig ?? {},
-        ...thinkingTranslation.outputConfigEffort ? { effort: thinkingTranslation.outputConfigEffort } : {}
-      };
+    if (outputConfig) {
+      options.output_config = outputConfig;
     }
-    if (thinkingTranslation.thinking) {
-      options.thinking = thinkingTranslation.thinking;
-    } else if (this.config.thinking?.type === "enabled") {
+    if (this.config.thinking?.type === "enabled") {
       options.thinking = {
         type: "enabled",
         budget_tokens: this.config.thinking.budgetTokens || 1e4
       };
     }
-    const mcp = toAnthropicMcp(request.config?.mcpServers);
-    const betas = [];
-    if (mcp.mcpServers.length > 0) {
-      options.mcp_servers = mcp.mcpServers;
-      betas.push(...mcp.betas);
-      if (mcp.tools.length > 0) {
-        const existingTools = Array.isArray(options.tools) ? options.tools : [];
-        options.tools = [...existingTools, ...mcp.tools];
-      }
-    }
-    return { options, messages, betas };
+    return { options, messages };
   }
   /**
    * Non-streaming completion (for debugging/comparison with original studio-ai)
    */
   async complete(request) {
     const client = await this.getClient();
-    const { options, betas } = this.buildRequestOptions(request);
+    const { options } = this.buildRequestOptions(request);
     const nonStreamingOptions = {
       ...options,
       stream: false
     };
     try {
-      const finalOptions = betas.length > 0 ? { ...nonStreamingOptions, betas } : nonStreamingOptions;
-      const messagesApi = betas.length > 0 ? client.beta.messages : client.messages;
       logProviderPayload(
         "anthropic",
         "request payload",
-        finalOptions,
+        nonStreamingOptions,
         request.debug
       );
-      const response = await messagesApi.create(finalOptions);
+      const response = await client.messages.create(nonStreamingOptions);
       logProviderPayload(
         "anthropic",
         "response payload",
@@ -1065,19 +986,17 @@ var AnthropicAdapter = class {
   }
   async *stream(request) {
     const client = await this.getClient();
-    const { options, betas } = this.buildRequestOptions(request);
+    const { options } = this.buildRequestOptions(request);
     const messageId = generateMessageId();
     yield { type: "message:start", id: messageId };
     try {
-      const finalOptions = betas.length > 0 ? { ...options, betas } : options;
-      const streamApi = betas.length > 0 ? client.beta.messages : client.messages;
       logProviderPayload(
         "anthropic",
         "request payload",
-        finalOptions,
+        options,
         request.debug
       );
-      const stream = await streamApi.stream(finalOptions);
+      const stream = await client.messages.stream(options);
       let currentToolUse = null;
       let isInThinkingBlock = false;
       const collectedCitations = [];

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

@@ -1,7 +1,7 @@
-import { b as AzureProviderConfig, A as AIProvider } from '../../types-BvkiJ1dd.mjs';
-import '../../types-D774b0dg.mjs';
+import { b as AzureProviderConfig, A as AIProvider } from '../../types-BSSiJW2o.mjs';
+import '../../types-BkQCSiIt.mjs';
 import 'zod';
-import '../../base-BYQKp9TW.mjs';
+import '../../base-tNgbBaSo.mjs';
 
 /**
  * Azure OpenAI Provider

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

@@ -1,7 +1,7 @@
-import { b as AzureProviderConfig, A as AIProvider } from '../../types-TMilS-Dz.js';
-import '../../types-D774b0dg.js';
+import { b as AzureProviderConfig, A as AIProvider } from '../../types-BQ31QIsA.js';
+import '../../types-BkQCSiIt.js';
 import 'zod';
-import '../../base-Cxq3ni0t.js';
+import '../../base-C58Dsr9p.js';
 
 /**
  * Azure OpenAI Provider

package/dist/providers/azure/index.js

@@ -112,6 +112,7 @@ function normalizeObjectJsonSchema(schema) {
 function toOpenAIResponseFormat(rf) {
   if (!rf) return void 0;
   if (rf.type === "json_object") return { type: "json_object" };
+  if (rf.type !== "json_schema" || !rf.json_schema) return void 0;
   return {
     type: "json_schema",
     json_schema: {

package/dist/providers/azure/index.mjs

@@ -110,6 +110,7 @@ function normalizeObjectJsonSchema(schema) {
 function toOpenAIResponseFormat(rf) {
   if (!rf) return void 0;
   if (rf.type === "json_object") return { type: "json_object" };
+  if (rf.type !== "json_schema" || !rf.json_schema) return void 0;
   return {
     type: "json_schema",
     json_schema: {

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

@@ -1,5 +1,7 @@
-import { L as LanguageModel } from '../../types-D774b0dg.mjs';
+import { L as LanguageModel } from '../../types-BkQCSiIt.mjs';
+import { A as AIProvider } from '../../types-BSSiJW2o.mjs';
 import 'zod';
+import '../../base-tNgbBaSo.mjs';
 
 /**
  * Fireworks Provider
@@ -46,4 +48,67 @@ interface FireworksProviderOptions {
  */
 declare function fireworks(modelId: string, options?: FireworksProviderOptions): LanguageModel;
 
-export { type FireworksProviderOptions, fireworks as createFireworks, fireworks };
+/**
+ * Fireworks Provider
+ *
+ * Fireworks.ai is a high-performance inference platform for open-source models
+ * (Llama, DeepSeek, Qwen, Mixtral, Gemma, and more).
+ *
+ * Uses an OpenAI-compatible API — set FIREWORKS_API_KEY in your environment.
+ *
+ * @see https://fireworks.ai/docs
+ *
+ * @example
+ * ```ts
+ * // Modern pattern — returns LanguageModel directly
+ * import { fireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { generateText } from '@yourgpt/llm-sdk';
+ *
+ * const result = await generateText({
+ *   model: fireworks('accounts/fireworks/models/llama-v3p1-70b-instruct'),
+ *   prompt: 'Hello!',
+ * });
+ *
+ * // Runtime pattern — returns AIProvider for createRuntime / fallback chain
+ * import { createFireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { createRuntime } from '@yourgpt/llm-sdk';
+ *
+ * const provider = createFireworks({ apiKey: '...' });
+ * const runtime = createRuntime({ provider, model: 'accounts/fireworks/models/deepseek-v3p1' });
+ * ```
+ */
+
+interface FireworksProviderConfig {
+    /** API key (defaults to FIREWORKS_API_KEY env var) */
+    apiKey?: string;
+    /** Base URL for API */
+    baseUrl?: string;
+}
+/**
+ * Create a Fireworks provider (callable, for use with createRuntime).
+ *
+ * Fireworks exposes an OpenAI-compatible API, so this reuses the OpenAI adapter
+ * pointed at the Fireworks base URL — the same approach used by Together AI and
+ * OpenRouter. The adapter speaks the runtime's API (generate/stream, rawMessages,
+ * message:delta/end events) and sets `stream_options.include_usage`, so streaming
+ * usage is reported correctly for credit accounting.
+ *
+ * @example
+ * ```typescript
+ * import { createFireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { createRuntime } from '@yourgpt/llm-sdk';
+ *
+ * const fireworks = createFireworks({ apiKey: '...' });
+ * const runtime = createRuntime({
+ *   provider: fireworks,
+ *   model: 'accounts/fireworks/models/deepseek-v3p1',
+ * });
+ *
+ * // Handle incoming chat requests
+ * return runtime.handleRequest(request);
+ * ```
+ */
+declare function createFireworks(config?: FireworksProviderConfig): AIProvider;
+declare const createFireworksProvider: typeof createFireworks;
+
+export { type FireworksProviderConfig, type FireworksProviderOptions, createFireworks, fireworks as createFireworksModel, createFireworksProvider, fireworks };

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

@@ -1,5 +1,7 @@
-import { L as LanguageModel } from '../../types-D774b0dg.js';
+import { L as LanguageModel } from '../../types-BkQCSiIt.js';
+import { A as AIProvider } from '../../types-BQ31QIsA.js';
 import 'zod';
+import '../../base-C58Dsr9p.js';
 
 /**
  * Fireworks Provider
@@ -46,4 +48,67 @@ interface FireworksProviderOptions {
  */
 declare function fireworks(modelId: string, options?: FireworksProviderOptions): LanguageModel;
 
-export { type FireworksProviderOptions, fireworks as createFireworks, fireworks };
+/**
+ * Fireworks Provider
+ *
+ * Fireworks.ai is a high-performance inference platform for open-source models
+ * (Llama, DeepSeek, Qwen, Mixtral, Gemma, and more).
+ *
+ * Uses an OpenAI-compatible API — set FIREWORKS_API_KEY in your environment.
+ *
+ * @see https://fireworks.ai/docs
+ *
+ * @example
+ * ```ts
+ * // Modern pattern — returns LanguageModel directly
+ * import { fireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { generateText } from '@yourgpt/llm-sdk';
+ *
+ * const result = await generateText({
+ *   model: fireworks('accounts/fireworks/models/llama-v3p1-70b-instruct'),
+ *   prompt: 'Hello!',
+ * });
+ *
+ * // Runtime pattern — returns AIProvider for createRuntime / fallback chain
+ * import { createFireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { createRuntime } from '@yourgpt/llm-sdk';
+ *
+ * const provider = createFireworks({ apiKey: '...' });
+ * const runtime = createRuntime({ provider, model: 'accounts/fireworks/models/deepseek-v3p1' });
+ * ```
+ */
+
+interface FireworksProviderConfig {
+    /** API key (defaults to FIREWORKS_API_KEY env var) */
+    apiKey?: string;
+    /** Base URL for API */
+    baseUrl?: string;
+}
+/**
+ * Create a Fireworks provider (callable, for use with createRuntime).
+ *
+ * Fireworks exposes an OpenAI-compatible API, so this reuses the OpenAI adapter
+ * pointed at the Fireworks base URL — the same approach used by Together AI and
+ * OpenRouter. The adapter speaks the runtime's API (generate/stream, rawMessages,
+ * message:delta/end events) and sets `stream_options.include_usage`, so streaming
+ * usage is reported correctly for credit accounting.
+ *
+ * @example
+ * ```typescript
+ * import { createFireworks } from '@yourgpt/llm-sdk/fireworks';
+ * import { createRuntime } from '@yourgpt/llm-sdk';
+ *
+ * const fireworks = createFireworks({ apiKey: '...' });
+ * const runtime = createRuntime({
+ *   provider: fireworks,
+ *   model: 'accounts/fireworks/models/deepseek-v3p1',
+ * });
+ *
+ * // Handle incoming chat requests
+ * return runtime.handleRequest(request);
+ * ```
+ */
+declare function createFireworks(config?: FireworksProviderConfig): AIProvider;
+declare const createFireworksProvider: typeof createFireworks;
+
+export { type FireworksProviderConfig, type FireworksProviderOptions, createFireworks, fireworks as createFireworksModel, createFireworksProvider, fireworks };