graphlit-client 1.0.20250615003 → 1.0.20250615004

package/README.md

@@ -75,7 +75,7 @@ await client.streamAgent(
     }
   },
   undefined, // conversationId (optional)
-  { id: spec.createSpecification.id } // specification
+  { id: spec.createSpecification.id }, // specification
 );
 ```
 
@@ -134,7 +134,7 @@ GOOGLE_API_KEY=your_key
 
 # Additional streaming providers
 GROQ_API_KEY=your_key          # For Groq models (Llama, Mixtral)
-CEREBRAS_API_KEY=your_key      # For Cerebras models  
+CEREBRAS_API_KEY=your_key      # For Cerebras models
 COHERE_API_KEY=your_key        # For Cohere Command models
 MISTRAL_API_KEY=your_key       # For Mistral models
 DEEPSEEK_API_KEY=your_key      # For Deepseek models
@@ -152,8 +152,9 @@ The SDK now includes automatic retry logic for network errors and transient fail
 ### Default Retry Configuration
 
 By default, the client will automatically retry on these status codes:
+
 - `429` - Too Many Requests
-- `502` - Bad Gateway  
+- `502` - Bad Gateway
 - `503` - Service Unavailable
 - `504` - Gateway Timeout
 
@@ -168,19 +169,19 @@ Configure retry behavior to match your needs:
 ```typescript
 const client = new Graphlit({
   organizationId: "your_org_id",
-  environmentId: "your_env_id", 
+  environmentId: "your_env_id",
   jwtSecret: "your_secret",
   retryConfig: {
-    maxAttempts: 10,              // Maximum retry attempts (default: 5)
-    initialDelay: 500,            // Initial delay in ms (default: 300)
-    maxDelay: 60000,              // Maximum delay in ms (default: 30000)
-    jitter: true,                 // Add randomness to delays (default: true)
+    maxAttempts: 10, // Maximum retry attempts (default: 5)
+    initialDelay: 500, // Initial delay in ms (default: 300)
+    maxDelay: 60000, // Maximum delay in ms (default: 30000)
+    jitter: true, // Add randomness to delays (default: true)
     retryableStatusCodes: [429, 500, 502, 503, 504], // Custom status codes
     onRetry: (attempt, error, operation) => {
       console.log(`Retry attempt ${attempt} for ${operation.operationName}`);
       console.log(`Error: ${error.message}`);
-    }
-  }
+    },
+  },
 });
 ```
 
@@ -196,7 +197,7 @@ const client = new Graphlit();
 client.setRetryConfig({
   maxAttempts: 20,
   initialDelay: 100,
-  retryableStatusCodes: [429, 500, 502, 503, 504, 521, 522, 524]
+  retryableStatusCodes: [429, 500, 502, 503, 504, 521, 522, 524],
 });
 ```
 
@@ -210,8 +211,8 @@ const client = new Graphlit({
   environmentId: "your_env_id",
   jwtSecret: "your_secret",
   retryConfig: {
-    maxAttempts: 1  // No retries
-  }
+    maxAttempts: 1, // No retries
+  },
 });
 ```
 
@@ -221,17 +222,17 @@ The Graphlit SDK supports real-time streaming responses from 9 different LLM pro
 
 ### Supported Providers
 
-| Provider | Models | SDK Required | API Key |
-|----------|--------|--------------|---------|
-| **OpenAI** | GPT-4, GPT-4o, GPT-4.1, O1, O3, O4 | `openai` | `OPENAI_API_KEY` |
-| **Anthropic** | Claude 3, Claude 3.5, Claude 3.7, Claude 4 | `@anthropic-ai/sdk` | `ANTHROPIC_API_KEY` |
-| **Google** | Gemini 1.5, Gemini 2.0, Gemini 2.5 | `@google/generative-ai` | `GOOGLE_API_KEY` |
-| **Groq** | Llama 4, Llama 3.3, Mixtral, Deepseek R1 | `groq-sdk` | `GROQ_API_KEY` |
-| **Cerebras** | Llama 3.3, Llama 3.1 | `openai` | `CEREBRAS_API_KEY` |
-| **Cohere** | Command R+, Command R, Command R7B, Command A | `cohere-ai` | `COHERE_API_KEY` |
-| **Mistral** | Mistral Large, Medium, Small, Nemo, Pixtral | `@mistralai/mistralai` | `MISTRAL_API_KEY` |
-| **AWS Bedrock** | Nova Premier/Pro, Claude 3.7, Llama 4 | `@aws-sdk/client-bedrock-runtime` | AWS credentials |
-| **Deepseek** | Deepseek Chat, Deepseek Reasoner | `openai` | `DEEPSEEK_API_KEY` |
+| Provider        | Models                                        | SDK Required                      | API Key             |
+| --------------- | --------------------------------------------- | --------------------------------- | ------------------- |
+| **OpenAI**      | GPT-4, GPT-4o, GPT-4.1, O1, O3, O4            | `openai`                          | `OPENAI_API_KEY`    |
+| **Anthropic**   | Claude 3, Claude 3.5, Claude 3.7, Claude 4    | `@anthropic-ai/sdk`               | `ANTHROPIC_API_KEY` |
+| **Google**      | Gemini 1.5, Gemini 2.0, Gemini 2.5            | `@google/generative-ai`           | `GOOGLE_API_KEY`    |
+| **Groq**        | Llama 4, Llama 3.3, Mixtral, Deepseek R1      | `groq-sdk`                        | `GROQ_API_KEY`      |
+| **Cerebras**    | Llama 3.3, Llama 3.1                          | `openai`                          | `CEREBRAS_API_KEY`  |
+| **Cohere**      | Command R+, Command R, Command R7B, Command A | `cohere-ai`                       | `COHERE_API_KEY`    |
+| **Mistral**     | Mistral Large, Medium, Small, Nemo, Pixtral   | `@mistralai/mistralai`            | `MISTRAL_API_KEY`   |
+| **AWS Bedrock** | Nova Premier/Pro, Claude 3.7, Llama 4         | `@aws-sdk/client-bedrock-runtime` | AWS credentials     |
+| **Deepseek**    | Deepseek Chat, Deepseek Reasoner              | `openai`                          | `DEEPSEEK_API_KEY`  |
 
 ### Setting Up Streaming
 
@@ -250,7 +251,9 @@ if (process.env.OPENAI_API_KEY) {
 
 if (process.env.COHERE_API_KEY) {
   const { CohereClient } = await import("cohere-ai");
-  client.setCohereClient(new CohereClient({ token: process.env.COHERE_API_KEY }));
+  client.setCohereClient(
+    new CohereClient({ token: process.env.COHERE_API_KEY }),
+  );
 }
 
 if (process.env.GROQ_API_KEY) {
@@ -263,9 +266,9 @@ const spec = await client.createSpecification({
   name: "Multi-Provider Assistant",
   type: Types.SpecificationTypes.Completion,
   serviceType: Types.ModelServiceTypes.Cohere, // or any supported provider
-  cohere: { 
+  cohere: {
     model: Types.CohereModels.CommandRPlus,
-    temperature: 0.7 
+    temperature: 0.7,
   },
 });
 ```
@@ -310,7 +313,7 @@ await client.streamAgent(
     }
   },
   undefined, // conversationId
-  { id: spec.createSpecification.id } // specification
+  { id: spec.createSpecification.id }, // specification
 );
 ```
 
@@ -338,7 +341,7 @@ const content = await client.ingestUri(
   "https://arxiv.org/pdf/1706.03762.pdf", // Attention Is All You Need paper
   "AI Research Paper", // name
   undefined, // id
-  true // isSynchronous - waits for processing
+  true, // isSynchronous - waits for processing
 );
 
 console.log(`✅ Uploaded: ${content.ingestUri.id}`);
@@ -360,7 +363,7 @@ await client.streamAgent(
     }
   },
   conversation.createConversation.id, // conversationId with content filter
-  { id: spec.createSpecification.id } // specification
+  { id: spec.createSpecification.id }, // specification
 );
 ```
 
@@ -374,7 +377,7 @@ const webpage = await client.ingestUri(
   "https://en.wikipedia.org/wiki/Artificial_intelligence", // uri
   "AI Wikipedia Page", // name
   undefined, // id
-  true // isSynchronous
+  true, // isSynchronous
 );
 
 // Wait for content to be indexed
@@ -389,7 +392,7 @@ const conversation = await client.createConversation({
 const response = await client.promptAgent(
   "Summarize the key points about AI from this Wikipedia page",
   conversation.createConversation.id, // conversationId with filter
-  { id: spec.createSpecification.id } // specification (create one as shown above)
+  { id: spec.createSpecification.id }, // specification (create one as shown above)
 );
 
 console.log(response.message);
@@ -412,7 +415,9 @@ if (process.env.OPENAI_API_KEY) {
 
 if (process.env.COHERE_API_KEY) {
   const { CohereClient } = await import("cohere-ai");
-  client.setCohereClient(new CohereClient({ token: process.env.COHERE_API_KEY }));
+  client.setCohereClient(
+    new CohereClient({ token: process.env.COHERE_API_KEY }),
+  );
 }
 
 if (process.env.GROQ_API_KEY) {
@@ -425,24 +430,24 @@ const providers = [
   {
     name: "OpenAI GPT-4o",
     serviceType: Types.ModelServiceTypes.OpenAi,
-    openAI: { model: Types.OpenAiModels.Gpt4O_128K }
+    openAI: { model: Types.OpenAiModels.Gpt4O_128K },
   },
   {
     name: "Cohere Command R+",
     serviceType: Types.ModelServiceTypes.Cohere,
-    cohere: { model: Types.CohereModels.CommandRPlus }
+    cohere: { model: Types.CohereModels.CommandRPlus },
   },
   {
     name: "Groq Llama",
     serviceType: Types.ModelServiceTypes.Groq,
-    groq: { model: Types.GroqModels.Llama_3_3_70B }
-  }
+    groq: { model: Types.GroqModels.Llama_3_3_70B },
+  },
 ];
 
 // Compare responses
 for (const provider of providers) {
   console.log(`\n🤖 ${provider.name}:`);
-  
+
   const spec = await client.createSpecification({
     ...provider,
     type: Types.SpecificationTypes.Completion,
@@ -456,7 +461,7 @@ for (const provider of providers) {
       }
     },
     undefined,
-    { id: spec.createSpecification.id }
+    { id: spec.createSpecification.id },
   );
 }
 ```
@@ -518,7 +523,7 @@ await client.streamAgent(
   undefined, // conversationId
   { id: spec.createSpecification.id }, // specification
   [weatherTool], // tools
-  toolHandlers // handlers
+  toolHandlers, // handlers
 );
 ```
 
@@ -563,7 +568,7 @@ class KnowledgeAssistant {
         url, // uri
         url.split("/").pop() || "Document", // name
         undefined, // id
-        true // isSynchronous - wait for processing
+        true, // isSynchronous - wait for processing
       );
       this.contentIds.push(content.ingestUri.id);
     }
@@ -593,7 +598,7 @@ class KnowledgeAssistant {
         }
       },
       this.conversationId, // Maintains conversation context
-      { id: this.specificationId! } // specification
+      { id: this.specificationId! }, // specification
     );
   }
 }
@@ -623,7 +628,7 @@ const document = await client.ingestUri(
   "https://example.com/document.pdf", // uri
   "Document #12345", // name
   undefined, // id
-  true // isSynchronous
+  true, // isSynchronous
 );
 
 // Wait for content to be indexed
@@ -634,7 +639,7 @@ const extraction = await client.extractContents(
   "Extract the key information from this document",
   undefined, // tools
   undefined, // specification
-  { contents: [{ id: document.ingestUri.id }] } // filter
+  { contents: [{ id: document.ingestUri.id }] }, // filter
 );
 
 console.log("Extracted data:", extraction.extractContents);
@@ -653,7 +658,7 @@ for (const url of documentUrls) {
     url, // uri
     url.split("/").pop() || "Document", // name
     undefined, // id
-    true // isSynchronous
+    true, // isSynchronous
   );
   ids.push(content.ingestUri.id);
 }
@@ -666,7 +671,7 @@ const summary = await client.summarizeContents(
       prompt: "Create an executive summary of these documents",
     },
   ], // summarizations
-  { contents: ids.map((id) => ({ id })) } // filter
+  { contents: ids.map((id) => ({ id })) }, // filter
 );
 
 console.log("Summary:", summary.summarizeContents);
@@ -680,13 +685,13 @@ const content = await client.ingestUri(
   "https://example.com/large-document.pdf", // uri
   undefined, // name
   undefined, // id
-  true // isSynchronous
+  true, // isSynchronous
 );
 console.log("✅ Content ready!");
 
 // Option 2: Asynchronous processing (for large files)
 const content = await client.ingestUri(
-  "https://example.com/very-large-video.mp4" // uri
+  "https://example.com/very-large-video.mp4", // uri
   // isSynchronous defaults to false
 );
 
@@ -724,7 +729,7 @@ const result = await client.promptAgent(
   {
     // Only allow retrieval from specific content
     contents: [{ id: "content-id-1" }, { id: "content-id-2" }],
-  }
+  },
 );
 
 // Example 2: Streaming with content filter
@@ -745,7 +750,7 @@ await client.streamAgent(
   {
     // Filter by collection
     collections: [{ id: "technical-docs-collection" }],
-  }
+  },
 );
 ```
 
@@ -775,7 +780,7 @@ await client.streamAgent(
   {
     // Force this content into context
     contents: [{ id: fileContent.content.id }],
-  }
+  },
 );
 ```
 
@@ -801,7 +806,7 @@ await client.promptAgent(
   {
     // Always include the specific code file
     contents: [{ id: "implementation-file-id" }],
-  }
+  },
 );
 ```
 
@@ -846,7 +851,7 @@ await client.updateProject({
 
 // Now all content will be automatically summarized
 const content = await client.ingestUri(
-  "https://example.com/report.pdf" // uri
+  "https://example.com/report.pdf", // uri
 );
 ```
 
@@ -879,7 +884,7 @@ await client.streamAgent(
     }
   },
   undefined,
-  { id: conversationSpec.createSpecification.id }
+  { id: conversationSpec.createSpecification.id },
 );
 ```
 

package/dist/client.js

@@ -133,7 +133,8 @@ class Graphlit {
     constructor(organizationIdOrOptions, environmentId, jwtSecret, ownerId, userId, apiUri) {
         // Handle both old constructor signature and new options object
         let options;
-        if (typeof organizationIdOrOptions === 'object' && organizationIdOrOptions !== null) {
+        if (typeof organizationIdOrOptions === "object" &&
+            organizationIdOrOptions !== null) {
             // New constructor with options object
             options = organizationIdOrOptions;
         }
@@ -145,7 +146,7 @@ class Graphlit {
                 jwtSecret,
                 ownerId,
                 userId,
-                apiUri
+                apiUri,
             };
         }
         this.apiUri =
@@ -158,7 +159,8 @@ class Graphlit {
             dotenv.config();
             this.organizationId =
                 options.organizationId || process.env.GRAPHLIT_ORGANIZATION_ID;
-            this.environmentId = options.environmentId || process.env.GRAPHLIT_ENVIRONMENT_ID;
+            this.environmentId =
+                options.environmentId || process.env.GRAPHLIT_ENVIRONMENT_ID;
             this.jwtSecret = options.jwtSecret || process.env.GRAPHLIT_JWT_SECRET;
             // optional: for multi-tenant support
             this.ownerId = options.ownerId || process.env.GRAPHLIT_OWNER_ID;
@@ -179,7 +181,7 @@ class Graphlit {
             maxDelay: 30000,
             retryableStatusCodes: [429, 502, 503, 504],
             jitter: true,
-            ...options.retryConfig
+            ...options.retryConfig,
         };
         if (!this.organizationId) {
             throw new Error("Graphlit organization identifier is required.");
@@ -223,7 +225,9 @@ class Graphlit {
                     if (statusCode && this.retryConfig.retryableStatusCodes) {
                         const shouldRetry = this.retryConfig.retryableStatusCodes.includes(statusCode);
                         // Call onRetry callback if provided
-                        if (shouldRetry && this.retryConfig.onRetry && _operation.getContext().retryCount !== undefined) {
+                        if (shouldRetry &&
+                            this.retryConfig.onRetry &&
+                            _operation.getContext().retryCount !== undefined) {
                             const attempt = _operation.getContext().retryCount + 1;
                             this.retryConfig.onRetry(attempt, error, _operation);
                         }
@@ -332,7 +336,7 @@ class Graphlit {
     setRetryConfig(retryConfig) {
         this.retryConfig = {
             ...this.retryConfig,
-            ...retryConfig
+            ...retryConfig,
         };
         // Refresh client to apply new retry configuration
         this.refreshClient();

package/dist/streaming/llm-formatters.d.ts

@@ -79,10 +79,13 @@ export declare function formatMessagesForAnthropic(messages: ConversationMessage
 export declare function formatMessagesForGoogle(messages: ConversationMessage[]): GoogleMessage[];
 /**
  * Cohere message format
+ * Note: For Cohere v7 SDK, messages are handled differently:
+ * - Current message is passed as 'message' parameter
+ * - Previous messages are passed as 'chatHistory' array
  */
 export interface CohereMessage {
-    role: "user" | "assistant" | "system" | "tool";
-    content: string;
+    role: "USER" | "CHATBOT" | "SYSTEM" | "TOOL";
+    message: string;
     tool_calls?: Array<{
         id: string;
         name: string;
@@ -90,12 +93,11 @@ export interface CohereMessage {
     }>;
     tool_results?: Array<{
         call: {
-            id: string;
             name: string;
             parameters: Record<string, any>;
         };
         outputs: Array<{
-            text: string;
+            output: string;
         }>;
     }>;
 }

package/dist/streaming/llm-formatters.js

@@ -284,14 +284,14 @@ export function formatMessagesForCohere(messages) {
         switch (message.role) {
             case ConversationRoleTypes.System:
                 formattedMessages.push({
-                    role: "system",
-                    content: trimmedMessage,
+                    role: "SYSTEM",
+                    message: trimmedMessage,
                 });
                 break;
             case ConversationRoleTypes.Assistant:
                 const assistantMessage = {
-                    role: "assistant",
-                    content: trimmedMessage,
+                    role: "CHATBOT",
+                    message: trimmedMessage,
                 };
                 // Add tool calls if present
                 if (message.toolCalls && message.toolCalls.length > 0) {
@@ -300,32 +300,37 @@ export function formatMessagesForCohere(messages) {
                         .map((toolCall) => ({
                         id: toolCall.id,
                         name: toolCall.name,
-                        parameters: toolCall.arguments ? JSON.parse(toolCall.arguments) : {},
+                        parameters: toolCall.arguments
+                            ? JSON.parse(toolCall.arguments)
+                            : {},
                     }));
                 }
                 formattedMessages.push(assistantMessage);
                 break;
             case ConversationRoleTypes.Tool:
-                // Cohere expects tool results as tool messages
+                // Cohere expects tool results as TOOL messages
                 formattedMessages.push({
-                    role: "tool",
-                    content: trimmedMessage,
-                    tool_results: [{
+                    role: "TOOL",
+                    message: trimmedMessage,
+                    tool_results: [
+                        {
                             call: {
-                                id: message.toolCallId || "",
                                 name: "", // Would need to be tracked from the tool call
                                 parameters: {},
                             },
-                            outputs: [{
-                                    text: trimmedMessage,
-                                }],
-                        }],
+                            outputs: [
+                                {
+                                    output: trimmedMessage, // Changed from 'text' to 'output'
+                                },
+                            ],
+                        },
+                    ],
                 });
                 break;
             default: // User messages
                 formattedMessages.push({
-                    role: "user",
-                    content: trimmedMessage,
+                    role: "USER",
+                    message: trimmedMessage,
                 });
                 break;
         }

package/dist/streaming/providers.js

@@ -321,6 +321,30 @@ onEvent, onComplete) {
         onComplete(fullMessage, toolCalls);
     }
     catch (error) {
+        // Handle OpenAI-specific errors
+        const errorMessage = error.message || error.toString();
+        // Check for rate limit errors
+        if (error.status === 429 ||
+            error.statusCode === 429 ||
+            error.code === "rate_limit_exceeded") {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [OpenAI] Rate limit hit`);
+            }
+            const rateLimitError = new Error("OpenAI rate limit exceeded");
+            rateLimitError.statusCode = 429;
+            throw rateLimitError;
+        }
+        // Check for network errors
+        if (errorMessage.includes("fetch failed") ||
+            error.code === "ECONNRESET" ||
+            error.code === "ETIMEDOUT") {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [OpenAI] Network error: ${errorMessage}`);
+            }
+            const networkError = new Error(`OpenAI network error: ${errorMessage}`);
+            networkError.statusCode = 503; // Service unavailable
+            throw networkError;
+        }
         // Don't emit error event here - let the client handle it to avoid duplicates
         throw error;
     }
@@ -631,6 +655,30 @@ onEvent, onComplete) {
         onComplete(fullMessage, validToolCalls);
     }
     catch (error) {
+        // Handle Anthropic-specific errors
+        const errorMessage = error.message || error.toString();
+        // Check for overloaded errors
+        if (error.type === "overloaded_error" ||
+            errorMessage.includes("Overloaded")) {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [Anthropic] Service overloaded`);
+            }
+            // Treat overloaded as a rate limit error for retry logic
+            const overloadError = new Error("Anthropic service overloaded");
+            overloadError.statusCode = 503; // Service unavailable
+            throw overloadError;
+        }
+        // Check for rate limit errors
+        if (error.status === 429 ||
+            error.statusCode === 429 ||
+            error.type === "rate_limit_error") {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [Anthropic] Rate limit hit`);
+            }
+            const rateLimitError = new Error("Anthropic rate limit exceeded");
+            rateLimitError.statusCode = 429;
+            throw rateLimitError;
+        }
         // Don't emit error event here - let the client handle it to avoid duplicates
         throw error;
     }
@@ -993,16 +1041,57 @@ onEvent, onComplete) {
  */
 export async function streamWithGroq(specification, messages, tools, groqClient, // Groq client instance (OpenAI-compatible)
 onEvent, onComplete) {
-    // Groq uses the same API as OpenAI, so we can reuse the OpenAI streaming logic
-    return streamWithOpenAI(specification, messages, tools, groqClient, onEvent, onComplete);
+    try {
+        // Groq uses the same API as OpenAI, so we can reuse the OpenAI streaming logic
+        return await streamWithOpenAI(specification, messages, tools, groqClient, onEvent, onComplete);
+    }
+    catch (error) {
+        // Handle Groq-specific errors
+        const errorMessage = error.message || error.toString();
+        // Check for tool calling errors
+        if (error.status === 400 &&
+            errorMessage.includes("Failed to call a function")) {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [Groq] Tool calling error: ${errorMessage}`);
+            }
+            // Groq may have limitations with certain tool schemas
+            // Re-throw with a more descriptive error
+            throw new Error(`Groq tool calling error: ${errorMessage}. The model may not support the provided tool schema format.`);
+        }
+        // Handle rate limits
+        if (error.status === 429 || error.statusCode === 429) {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [Groq] Rate limit hit (429)`);
+            }
+            const rateLimitError = new Error("Groq rate limit exceeded");
+            rateLimitError.statusCode = 429;
+            throw rateLimitError;
+        }
+        throw error;
+    }
 }
 /**
  * Stream with Cerebras SDK (OpenAI-compatible)
  */
 export async function streamWithCerebras(specification, messages, tools, cerebrasClient, // OpenAI client instance configured for Cerebras
 onEvent, onComplete) {
-    // Cerebras uses the same API as OpenAI, so we can reuse the OpenAI streaming logic
-    return streamWithOpenAI(specification, messages, tools, cerebrasClient, onEvent, onComplete);
+    try {
+        // Cerebras uses the same API as OpenAI, so we can reuse the OpenAI streaming logic
+        return await streamWithOpenAI(specification, messages, tools, cerebrasClient, onEvent, onComplete);
+    }
+    catch (error) {
+        // Handle Cerebras-specific 429 errors
+        if (error.status === 429 || error.statusCode === 429) {
+            if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
+                console.log(`⚠️ [Cerebras] Rate limit hit (429)`);
+            }
+            // Re-throw with proper status code for retry logic
+            const rateLimitError = new Error("Cerebras rate limit exceeded");
+            rateLimitError.statusCode = 429;
+            throw rateLimitError;
+        }
+        throw error;
+    }
 }
 /**
  * Stream with Deepseek SDK (OpenAI-compatible)
@@ -1037,7 +1126,7 @@ onEvent, onComplete) {
                 serviceType: specification.serviceType,
                 deepseek: specification.deepseek,
                 hasDeepseekModel: !!specification.deepseek?.model,
-                deepseekModelValue: specification.deepseek?.model
+                deepseekModelValue: specification.deepseek?.model,
             });
         }
         const modelName = getModelName(specification);
@@ -1046,7 +1135,7 @@ onEvent, onComplete) {
                 name: specification.name,
                 serviceType: specification.serviceType,
                 deepseek: specification.deepseek,
-                hasCustomModelName: !!specification.deepseek?.modelName
+                hasCustomModelName: !!specification.deepseek?.modelName,
             });
             throw new Error(`No model name found for specification: ${specification.name} (service: ${specification.serviceType})`);
         }
@@ -1109,9 +1198,12 @@ onEvent, onComplete) {
                 // Performance metrics tracking (internal only)
                 if (tokenCount % 10 === 0) {
                     const totalTokens = tokenCount + toolArgumentTokens;
-                    const tokensPerSecond = totalTokens > 0 ? totalTokens / ((currentTime - startTime) / 1000) : 0;
+                    const tokensPerSecond = totalTokens > 0
+                        ? totalTokens / ((currentTime - startTime) / 1000)
+                        : 0;
                     const avgInterTokenDelay = interTokenDelays.length > 0
-                        ? interTokenDelays.reduce((a, b) => a + b, 0) / interTokenDelays.length
+                        ? interTokenDelays.reduce((a, b) => a + b, 0) /
+                            interTokenDelays.length
                         : 0;
                 }
             }
@@ -1231,10 +1323,22 @@ onEvent, onComplete) {
         if (messages.length === 0) {
             throw new Error("No messages found for Cohere streaming");
         }
+        // Cohere v7 expects a single message and optional chatHistory
+        // Extract the last message as the current message
+        const lastMessage = messages[messages.length - 1];
+        const chatHistory = messages.slice(0, -1);
+        if (!lastMessage || !lastMessage.message) {
+            throw new Error("Last message must have message property for Cohere streaming");
+        }
         const streamConfig = {
             model: modelName,
-            messages: messages, // All messages in chronological order
+            message: lastMessage.message, // Current message (singular)
         };
+        // Add chat history if there are previous messages
+        if (chatHistory.length > 0) {
+            // Messages already have 'message' property from formatter
+            streamConfig.chatHistory = chatHistory;
+        }
         // Only add temperature if it's defined
         if (specification.cohere?.temperature !== undefined) {
             streamConfig.temperature = specification.cohere.temperature;
@@ -1283,12 +1387,9 @@ onEvent, onComplete) {
         }
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
             console.log(`🔍 [Cohere] Final stream config:`, JSON.stringify(streamConfig, null, 2));
-            console.log(`🔍 [Cohere] Cohere client methods available:`, Object.getOwnPropertyNames(cohereClient));
-            console.log(`🔍 [Cohere] Has chatStream method:`, typeof cohereClient.chatStream === 'function');
-            console.log(`🔍 [Cohere] Has chat property:`, !!cohereClient.chat);
-            if (cohereClient.chat) {
-                console.log(`🔍 [Cohere] Chat methods:`, Object.getOwnPropertyNames(cohereClient.chat));
-            }
+            console.log(`🔍 [Cohere] Current message: "${streamConfig.message}"`);
+            console.log(`🔍 [Cohere] Chat history length: ${streamConfig.chatHistory?.length || 0}`);
+            console.log(`🔍 [Cohere] Has tools: ${!!streamConfig.tools}`);
             console.log(`⏱️ [Cohere] Starting stream request at: ${new Date().toISOString()}`);
         }
         let stream;
@@ -1498,7 +1599,7 @@ onEvent, onComplete) {
                 name: specification.name,
                 serviceType: specification.serviceType,
                 bedrock: specification.bedrock,
-                hasCustomModelName: !!specification.bedrock?.modelName
+                hasCustomModelName: !!specification.bedrock?.modelName,
             });
             throw new Error(`No model name found for Bedrock specification: ${specification.name} (service: ${specification.serviceType}, bedrock.model: ${specification.bedrock?.model})`);
         }
@@ -1511,9 +1612,13 @@ onEvent, onComplete) {
         // The AWS SDK expects content as an array of content blocks
         const converseMessages = messages.map((msg) => ({
             role: msg.role,
-            content: [{
-                    text: typeof msg.content === 'string' ? msg.content : msg.content.toString()
-                }]
+            content: [
+                {
+                    text: typeof msg.content === "string"
+                        ? msg.content
+                        : msg.content.toString(),
+                },
+            ],
         }));
         // Prepare the request using Converse API format
         // Using 'any' type because:
@@ -1649,9 +1754,25 @@ onEvent, onComplete) {
         if (process.env.DEBUG_GRAPHLIT_SDK_STREAMING) {
             console.error(`❌ [Bedrock] Stream error:`, error);
         }
+        // Handle specific Bedrock errors
+        const errorMessage = error.message || error.toString();
+        const errorName = error.name || "";
+        // Check for throttling errors
+        if (errorName === "ThrottlingException" ||
+            errorMessage.includes("Too many tokens") ||
+            errorMessage.includes("Too many requests")) {
+            onEvent({
+                type: "error",
+                error: `Bedrock rate limit: ${errorMessage}`,
+            });
+            // Re-throw with a specific error type that the retry logic can handle
+            const rateLimitError = new Error(errorMessage);
+            rateLimitError.statusCode = 429; // Treat as rate limit
+            throw rateLimitError;
+        }
         onEvent({
             type: "error",
-            error: `Bedrock streaming error: ${error}`,
+            error: `Bedrock streaming error: ${errorMessage}`,
         });
         throw error;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "graphlit-client",
-  "version": "1.0.20250615003",
+  "version": "1.0.20250615004",
   "description": "Graphlit API Client for TypeScript",
   "type": "module",
   "main": "./dist/client.js",