@yourgpt/llm-sdk 2.5.1-beta.2 → 2.5.1-beta.4

package/dist/adapters/index.js

@@ -487,6 +487,7 @@ var OpenAIAdapter = class _OpenAIAdapter {
     if (baseUrl.includes("generativelanguage.googleapis.com")) return "google";
     if (baseUrl.includes("x.ai")) return "xai";
     if (baseUrl.includes("azure")) return "azure";
+    if (baseUrl.includes("fireworks.ai")) return "fireworks";
     return "openai";
   }
   async getClient() {
@@ -1304,6 +1305,11 @@ var AnthropicAdapter = class {
         content,
         toolCalls,
         thinking: thinking || void 0,
+        usage: response.usage ? {
+          promptTokens: response.usage.input_tokens ?? 0,
+          completionTokens: response.usage.output_tokens ?? 0,
+          totalTokens: (response.usage.input_tokens ?? 0) + (response.usage.output_tokens ?? 0)
+        } : void 0,
         rawResponse: response
       };
     } catch (error) {
@@ -2162,6 +2168,11 @@ var GoogleAdapter = class {
     return {
       content: textContent,
       toolCalls,
+      usage: response.usageMetadata ? {
+        promptTokens: response.usageMetadata.promptTokenCount ?? 0,
+        completionTokens: (response.usageMetadata.candidatesTokenCount ?? 0) + (response.usageMetadata.thoughtsTokenCount ?? 0),
+        totalTokens: response.usageMetadata.totalTokenCount ?? 0
+      } : void 0,
       rawResponse: response
     };
   }
@@ -2390,6 +2401,11 @@ var AzureAdapter = class {
     return {
       content: message?.content || "",
       toolCalls,
+      usage: response.usage ? {
+        promptTokens: response.usage.prompt_tokens ?? 0,
+        completionTokens: response.usage.completion_tokens ?? 0,
+        totalTokens: response.usage.total_tokens ?? (response.usage.prompt_tokens ?? 0) + (response.usage.completion_tokens ?? 0)
+      } : void 0,
       rawResponse: response
     };
   }

package/dist/adapters/index.mjs

@@ -485,6 +485,7 @@ var OpenAIAdapter = class _OpenAIAdapter {
     if (baseUrl.includes("generativelanguage.googleapis.com")) return "google";
     if (baseUrl.includes("x.ai")) return "xai";
     if (baseUrl.includes("azure")) return "azure";
+    if (baseUrl.includes("fireworks.ai")) return "fireworks";
     return "openai";
   }
   async getClient() {
@@ -1302,6 +1303,11 @@ var AnthropicAdapter = class {
         content,
         toolCalls,
         thinking: thinking || void 0,
+        usage: response.usage ? {
+          promptTokens: response.usage.input_tokens ?? 0,
+          completionTokens: response.usage.output_tokens ?? 0,
+          totalTokens: (response.usage.input_tokens ?? 0) + (response.usage.output_tokens ?? 0)
+        } : void 0,
         rawResponse: response
       };
     } catch (error) {
@@ -2160,6 +2166,11 @@ var GoogleAdapter = class {
     return {
       content: textContent,
       toolCalls,
+      usage: response.usageMetadata ? {
+        promptTokens: response.usageMetadata.promptTokenCount ?? 0,
+        completionTokens: (response.usageMetadata.candidatesTokenCount ?? 0) + (response.usageMetadata.thoughtsTokenCount ?? 0),
+        totalTokens: response.usageMetadata.totalTokenCount ?? 0
+      } : void 0,
       rawResponse: response
     };
   }
@@ -2388,6 +2399,11 @@ var AzureAdapter = class {
     return {
       content: message?.content || "",
       toolCalls,
+      usage: response.usage ? {
+        promptTokens: response.usage.prompt_tokens ?? 0,
+        completionTokens: response.usage.completion_tokens ?? 0,
+        totalTokens: response.usage.total_tokens ?? (response.usage.prompt_tokens ?? 0) + (response.usage.completion_tokens ?? 0)
+      } : void 0,
       rawResponse: response
     };
   }

package/dist/index.d.mts

@@ -1001,7 +1001,7 @@ declare class Runtime {
      * 5. Loop continues until no more tool calls or max iterations reached
      * 6. Returns all new messages in the done event for client to append
      */
-    processChatWithLoop(request: ChatRequest, signal?: AbortSignal, _accumulatedMessages?: DoneEventMessage[], _isRecursive?: boolean, _httpRequest?: Request, _toolSearchState?: ToolSearchState): AsyncGenerator<StreamEvent>;
+    processChatWithLoop(request: ChatRequest, signal?: AbortSignal, _accumulatedMessages?: DoneEventMessage[], _isRecursive?: boolean, _httpRequest?: Request, _toolSearchState?: ToolSearchState, _iteration?: number): AsyncGenerator<StreamEvent>;
     /**
      * Non-streaming agent loop implementation
      *

package/dist/index.d.ts

@@ -1001,7 +1001,7 @@ declare class Runtime {
      * 5. Loop continues until no more tool calls or max iterations reached
      * 6. Returns all new messages in the done event for client to append
      */
-    processChatWithLoop(request: ChatRequest, signal?: AbortSignal, _accumulatedMessages?: DoneEventMessage[], _isRecursive?: boolean, _httpRequest?: Request, _toolSearchState?: ToolSearchState): AsyncGenerator<StreamEvent>;
+    processChatWithLoop(request: ChatRequest, signal?: AbortSignal, _accumulatedMessages?: DoneEventMessage[], _isRecursive?: boolean, _httpRequest?: Request, _toolSearchState?: ToolSearchState, _iteration?: number): AsyncGenerator<StreamEvent>;
     /**
      * Non-streaming agent loop implementation
      *

package/dist/index.js

@@ -2112,7 +2112,7 @@ var Runtime = class {
    * 5. Loop continues until no more tool calls or max iterations reached
    * 6. Returns all new messages in the done event for client to append
    */
-  async *processChatWithLoop(request, signal, _accumulatedMessages, _isRecursive, _httpRequest, _toolSearchState) {
+  async *processChatWithLoop(request, signal, _accumulatedMessages, _isRecursive, _httpRequest, _toolSearchState, _iteration = 0) {
     const debug = this.config.debug;
     if (request.streaming === false) {
       if (debug) {
@@ -2131,7 +2131,25 @@ var Runtime = class {
       return;
     }
     const newMessages = _accumulatedMessages || [];
-    this.config.maxIterations ?? 20;
+    const maxIterations = this.config.maxIterations ?? 20;
+    if (signal?.aborted) {
+      yield {
+        type: "error",
+        message: "Aborted",
+        code: "ABORTED"
+      };
+      return;
+    }
+    if (_iteration >= maxIterations) {
+      if (debug) {
+        console.log(`[Copilot SDK] Max iterations (${maxIterations}) reached`);
+      }
+      yield {
+        type: "done",
+        messages: newMessages.length > 0 ? newMessages : void 0
+      };
+      return;
+    }
     const allTools = this.collectToolsForRequest(request);
     const nativeToolSearch = this.resolveNativeToolSearchForRequest(request);
     const nativeToolCatalog = nativeToolSearch ? this.buildNativeToolCatalogForRequest(request, allTools) : null;
@@ -2393,7 +2411,8 @@ var Runtime = class {
           true,
           // Mark as recursive
           _httpRequest,
-          nextToolSearchState
+          nextToolSearchState,
+          _iteration + 1
         )) {
           yield event;
         }
@@ -2490,7 +2509,9 @@ var Runtime = class {
           _accumulatedMessages,
           _isRecursive,
           _httpRequest,
-          toolSearchState
+          toolSearchState,
+          iteration
+          // carry the non-streaming loop's depth into the streaming sub-call
         )) {
           yield event;
         }

package/dist/index.mjs

@@ -2110,7 +2110,7 @@ var Runtime = class {
    * 5. Loop continues until no more tool calls or max iterations reached
    * 6. Returns all new messages in the done event for client to append
    */
-  async *processChatWithLoop(request, signal, _accumulatedMessages, _isRecursive, _httpRequest, _toolSearchState) {
+  async *processChatWithLoop(request, signal, _accumulatedMessages, _isRecursive, _httpRequest, _toolSearchState, _iteration = 0) {
     const debug = this.config.debug;
     if (request.streaming === false) {
       if (debug) {
@@ -2129,7 +2129,25 @@ var Runtime = class {
       return;
     }
     const newMessages = _accumulatedMessages || [];
-    this.config.maxIterations ?? 20;
+    const maxIterations = this.config.maxIterations ?? 20;
+    if (signal?.aborted) {
+      yield {
+        type: "error",
+        message: "Aborted",
+        code: "ABORTED"
+      };
+      return;
+    }
+    if (_iteration >= maxIterations) {
+      if (debug) {
+        console.log(`[Copilot SDK] Max iterations (${maxIterations}) reached`);
+      }
+      yield {
+        type: "done",
+        messages: newMessages.length > 0 ? newMessages : void 0
+      };
+      return;
+    }
     const allTools = this.collectToolsForRequest(request);
     const nativeToolSearch = this.resolveNativeToolSearchForRequest(request);
     const nativeToolCatalog = nativeToolSearch ? this.buildNativeToolCatalogForRequest(request, allTools) : null;
@@ -2391,7 +2409,8 @@ var Runtime = class {
           true,
           // Mark as recursive
           _httpRequest,
-          nextToolSearchState
+          nextToolSearchState,
+          _iteration + 1
         )) {
           yield event;
         }
@@ -2488,7 +2507,9 @@ var Runtime = class {
           _accumulatedMessages,
           _isRecursive,
           _httpRequest,
-          toolSearchState
+          toolSearchState,
+          iteration
+          // carry the non-streaming loop's depth into the streaming sub-call
         )) {
           yield event;
         }

package/dist/providers/anthropic/index.js

@@ -980,6 +980,11 @@ var AnthropicAdapter = class {
         content,
         toolCalls,
         thinking: thinking || void 0,
+        usage: response.usage ? {
+          promptTokens: response.usage.input_tokens ?? 0,
+          completionTokens: response.usage.output_tokens ?? 0,
+          totalTokens: (response.usage.input_tokens ?? 0) + (response.usage.output_tokens ?? 0)
+        } : void 0,
         rawResponse: response
       };
     } catch (error) {

package/dist/providers/anthropic/index.mjs

@@ -978,6 +978,11 @@ var AnthropicAdapter = class {
         content,
         toolCalls,
         thinking: thinking || void 0,
+        usage: response.usage ? {
+          promptTokens: response.usage.input_tokens ?? 0,
+          completionTokens: response.usage.output_tokens ?? 0,
+          totalTokens: (response.usage.input_tokens ?? 0) + (response.usage.output_tokens ?? 0)
+        } : void 0,
         rawResponse: response
       };
     } catch (error) {

package/dist/providers/azure/index.js

@@ -419,6 +419,11 @@ var AzureAdapter = class {
     return {
       content: message?.content || "",
       toolCalls,
+      usage: response.usage ? {
+        promptTokens: response.usage.prompt_tokens ?? 0,
+        completionTokens: response.usage.completion_tokens ?? 0,
+        totalTokens: response.usage.total_tokens ?? (response.usage.prompt_tokens ?? 0) + (response.usage.completion_tokens ?? 0)
+      } : void 0,
       rawResponse: response
     };
   }

package/dist/providers/azure/index.mjs

@@ -417,6 +417,11 @@ var AzureAdapter = class {
     return {
       content: message?.content || "",
       toolCalls,
+      usage: response.usage ? {
+        promptTokens: response.usage.prompt_tokens ?? 0,
+        completionTokens: response.usage.completion_tokens ?? 0,
+        totalTokens: response.usage.total_tokens ?? (response.usage.prompt_tokens ?? 0) + (response.usage.completion_tokens ?? 0)
+      } : void 0,
       rawResponse: response
     };
   }

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

@@ -1,5 +1,7 @@
 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-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 };