graphlit-client 1.0.20250924002 → 1.0.20250924003

package/dist/client.js

@@ -2892,7 +2892,12 @@ class Graphlit {
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
             console.log(`🚀 [Graphlit SDK] Routing to OpenAI streaming provider | Spec: ${specification.name} (${specification.id}) | Messages: ${messages.length} | Tools: ${tools?.length || 0}`);
         }
-        await streamWithOpenAI(specification, messages, tools, openaiClient, (event) => uiAdapter.handleEvent(event), onComplete, abortSignal);
+        // Extract reasoning effort for OpenAI o1 models
+        const reasoningEffort = specification.openAI?.reasoningEffort || undefined;
+        if (reasoningEffort && process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+            console.log(`🧠 [Graphlit SDK] OpenAI reasoning effort: ${reasoningEffort}`);
+        }
+        await streamWithOpenAI(specification, messages, tools, openaiClient, (event) => uiAdapter.handleEvent(event), onComplete, abortSignal, reasoningEffort);
     }
     /**
      * Stream with Anthropic client
@@ -2967,7 +2972,12 @@ class Graphlit {
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
             console.log(`🚀 [Graphlit SDK] Routing to Google streaming provider | Spec: ${specification.name} (${specification.id}) | Messages: ${messages.length} | Tools: ${tools?.length || 0} | SystemPrompt: ${systemPrompt ? "Yes" : "No"}`);
         }
-        await streamWithGoogle(specification, messages, systemPrompt, tools, googleClient, (event) => uiAdapter.handleEvent(event), onComplete, abortSignal);
+        // Get thinking configuration from specification
+        const thinkingConfig = this.getThinkingConfig(specification);
+        if (thinkingConfig && process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+            console.log(`🧠 [Graphlit SDK] Google thinking enabled | Budget: ${thinkingConfig.budget_tokens} tokens`);
+        }
+        await streamWithGoogle(specification, messages, systemPrompt, tools, googleClient, (event) => uiAdapter.handleEvent(event), onComplete, abortSignal, thinkingConfig);
     }
     /**
      * Stream with Groq client (OpenAI-compatible)

package/dist/streaming/providers.d.ts

@@ -5,7 +5,7 @@ import { StreamEvent } from "../types/internal.js";
  * Stream with OpenAI SDK
  */
 export declare function streamWithOpenAI(specification: Specification, messages: OpenAIMessage[], tools: ToolDefinitionInput[] | undefined, openaiClient: any, // OpenAI client instance
-onEvent: (event: StreamEvent) => void, onComplete: (message: string, toolCalls: ConversationToolCall[], usage?: any) => void, abortSignal?: AbortSignal): Promise<void>;
+onEvent: (event: StreamEvent) => void, onComplete: (message: string, toolCalls: ConversationToolCall[], usage?: any) => void, abortSignal?: AbortSignal, reasoningEffort?: string): Promise<void>;
 /**
  * Stream with Anthropic SDK
  */
@@ -19,7 +19,10 @@ onEvent: (event: StreamEvent) => void, onComplete: (message: string, toolCalls:
  * Stream with Google SDK
  */
 export declare function streamWithGoogle(specification: Specification, messages: GoogleMessage[], systemPrompt: string | undefined, tools: ToolDefinitionInput[] | undefined, googleClient: any, // Google GenerativeAI client instance
-onEvent: (event: StreamEvent) => void, onComplete: (message: string, toolCalls: ConversationToolCall[], usage?: any) => void, abortSignal?: AbortSignal): Promise<void>;
+onEvent: (event: StreamEvent) => void, onComplete: (message: string, toolCalls: ConversationToolCall[], usage?: any) => void, abortSignal?: AbortSignal, thinkingConfig?: {
+    type: "enabled";
+    budget_tokens: number;
+}): Promise<void>;
 /**
  * Stream with Groq SDK (OpenAI-compatible)
  */

package/dist/streaming/providers.js

@@ -76,7 +76,7 @@ function cleanSchemaForGoogle(schema) {
  * Stream with OpenAI SDK
  */
 export async function streamWithOpenAI(specification, messages, tools, openaiClient, // OpenAI client instance
-onEvent, onComplete, abortSignal) {
+onEvent, onComplete, abortSignal, reasoningEffort) {
     let fullMessage = "";
     let toolCalls = [];
     let usageData = null;
@@ -105,7 +105,7 @@ onEvent, onComplete, abortSignal) {
             throw new Error(`No model name found for specification: ${specification.name} (service: ${specification.serviceType})`);
         }
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
-            console.log(`🤖 [OpenAI] Model Config: Service=OpenAI | Model=${modelName} | Temperature=${specification.openAI?.temperature} | MaxTokens=${specification.openAI?.completionTokenLimit || "null"} | Tools=${tools?.length || 0} | Spec="${specification.name}"`);
+            console.log(`🤖 [OpenAI] Model Config: Service=OpenAI | Model=${modelName} | Temperature=${specification.openAI?.temperature} | MaxTokens=${specification.openAI?.completionTokenLimit || "null"} | Tools=${tools?.length || 0} | ReasoningEffort=${reasoningEffort || "none"} | Spec="${specification.name}"`);
         }
         const streamConfig = {
             model: modelName,
@@ -131,6 +131,14 @@ onEvent, onComplete, abortSignal) {
                 },
             }));
         }
+        // Add reasoning effort for o1 models
+        if (reasoningEffort) {
+            // OpenAI o1 models support reasoning_effort parameter
+            streamConfig.reasoning_effort = reasoningEffort.toLowerCase();
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`🧠 [OpenAI] Reasoning effort set to: ${reasoningEffort}`);
+            }
+        }
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
             console.log(`⏱️ [OpenAI] Starting LLM call at: ${new Date().toISOString()}`);
         }
@@ -894,7 +902,7 @@ onEvent, onComplete, abortSignal, thinkingConfig) {
  * Stream with Google SDK
  */
 export async function streamWithGoogle(specification, messages, systemPrompt, tools, googleClient, // Google GenerativeAI client instance
-onEvent, onComplete, abortSignal) {
+onEvent, onComplete, abortSignal, thinkingConfig) {
     let fullMessage = "";
     let toolCalls = [];
     let usageData = null;
@@ -923,7 +931,7 @@ onEvent, onComplete, abortSignal) {
             throw new Error(`No model name found for Google specification: ${specification.name}`);
         }
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
-            console.log(`🤖 [Google] Model Config: Service=Google | Model=${modelName} | Temperature=${specification.google?.temperature} | MaxTokens=${specification.google?.completionTokenLimit || "null"} | SystemPrompt=${systemPrompt ? "Yes" : "No"} | Tools=${tools?.length || 0} | Spec="${specification.name}"`);
+            console.log(`🤖 [Google] Model Config: Service=Google | Model=${modelName} | Temperature=${specification.google?.temperature} | MaxTokens=${specification.google?.completionTokenLimit || "null"} | SystemPrompt=${systemPrompt ? "Yes" : "No"} | Tools=${tools?.length || 0} | Thinking=${!!thinkingConfig} | Spec="${specification.name}"`);
         }
         const streamConfig = {
             model: modelName,
@@ -969,12 +977,25 @@ onEvent, onComplete, abortSignal) {
                 },
             ]
             : undefined;
+        // Add thinking configuration if provided
+        // Note: Google's thinking API is still in preview and may require specific model support
+        const generationConfig = {
+            temperature: streamConfig.temperature,
+            maxOutputTokens: streamConfig.max_tokens,
+        };
+        if (thinkingConfig) {
+            // Google Gemini Flash 2.5+ supports thinking mode
+            // The API may use a different parameter name than Anthropic
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`🧠 [Google] Extended thinking enabled | Budget: ${thinkingConfig.budget_tokens} tokens`);
+            }
+            // Note: Google's thinking API implementation may differ from Anthropic
+            // This is a placeholder for when Google releases their thinking API
+            // generationConfig.thinking = { enabled: true, maxTokens: thinkingConfig.budget_tokens };
+        }
         const model = googleClient.getGenerativeModel({
             model: modelName,
-            generationConfig: {
-                temperature: streamConfig.temperature,
-                maxOutputTokens: streamConfig.max_tokens,
-            },
+            generationConfig,
             tools: googleTools,
         });
         // Convert messages to Google chat format

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "graphlit-client",
-  "version": "1.0.20250924002",
+  "version": "1.0.20250924003",
   "description": "Graphlit API Client for TypeScript",
   "type": "module",
   "main": "./dist/client.js",