openlit 1.11.0 → 1.12.0

package/dist/instrumentation/openai/wrapper.js

@@ -1,21 +1,71 @@
 "use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
 var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 const api_1 = require("@opentelemetry/api");
 const config_1 = __importDefault(require("../../config"));
-const helpers_1 = __importDefault(require("../../helpers"));
+const helpers_1 = __importStar(require("../../helpers"));
 const semantic_convention_1 = __importDefault(require("../../semantic-convention"));
 const base_wrapper_1 = __importDefault(require("../base-wrapper"));
+function spanCreationAttrs(operationName, requestModel) {
+    return {
+        [semantic_convention_1.default.GEN_AI_OPERATION]: operationName,
+        [semantic_convention_1.default.GEN_AI_PROVIDER_NAME_OTEL]: semantic_convention_1.default.GEN_AI_SYSTEM_OPENAI,
+        [semantic_convention_1.default.GEN_AI_REQUEST_MODEL]: requestModel,
+        [semantic_convention_1.default.SERVER_ADDRESS]: OpenAIWrapper.serverAddress,
+        [semantic_convention_1.default.SERVER_PORT]: OpenAIWrapper.serverPort,
+    };
+}
 class OpenAIWrapper extends base_wrapper_1.default {
     static _patchChatCompletionCreate(tracer) {
         const genAIEndpoint = 'openai.resources.chat.completions';
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'gpt-4o';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT, requestModel),
+                }, effectiveCtx);
                 return api_1.context
-                    .with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                    .with(api_1.trace.setSpan(effectiveCtx, span), async () => {
                     return originalMethod.apply(this, args);
                 })
                     .then((response) => {
@@ -32,7 +82,16 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 })
                     .catch((e) => {
                     helpers_1.default.handleException(span, e);
+                    base_wrapper_1.default.recordMetrics(span, {
+                        genAIEndpoint,
+                        model: requestModel,
+                        aiSystem: OpenAIWrapper.aiSystem,
+                        serverAddress: OpenAIWrapper.serverAddress,
+                        serverPort: OpenAIWrapper.serverPort,
+                        errorType: e?.constructor?.name || '_OTHER',
+                    });
                     span.end();
+                    throw e;
                 });
             };
         };
@@ -50,10 +109,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
         }
         catch (e) {
             helpers_1.default.handleException(span, e);
+            throw e;
         }
         finally {
             span.end();
-            // Record metrics after span has ended if parameters are available
             if (metricParams) {
                 base_wrapper_1.default.recordMetrics(span, metricParams);
             }
@@ -94,7 +153,7 @@ class OpenAIWrapper extends base_wrapper_1.default {
                     },
                 },
             };
-            let toolCalls = [];
+            const toolCalls = [];
             for await (const chunk of response) {
                 timestamps.push(Date.now());
                 result.id = chunk.id;
@@ -115,12 +174,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 if (chunk.choices[0]?.delta.content) {
                     result.choices[0].message.content += chunk.choices[0].delta.content;
                 }
-                // Improved tool calls handling for streaming
                 if (chunk.choices[0]?.delta.tool_calls) {
                     const deltaTools = chunk.choices[0].delta.tool_calls;
                     for (const tool of deltaTools) {
                         const idx = tool.index || 0;
-                        // Extend array if needed
                         while (toolCalls.length <= idx) {
                             toolCalls.push({
                                 id: '',
@@ -129,7 +186,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
                             });
                         }
                         if (tool.id) {
-                            // New tool call
                             toolCalls[idx].id = tool.id;
                             toolCalls[idx].type = tool.type || 'function';
                             if (tool.function?.name) {
@@ -140,7 +196,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
                             }
                         }
                         else if (tool.function?.arguments) {
-                            // Append arguments to existing tool call
                             toolCalls[idx].function.arguments += tool.function.arguments;
                         }
                     }
@@ -169,7 +224,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 };
             }
             args[0].tools = tools;
-            // Calculate TTFT and TBT
             const ttft = timestamps.length > 0 ? (timestamps[0] - startTime) / 1000 : 0;
             let tbt = 0;
             if (timestamps.length > 1) {
@@ -188,107 +242,95 @@ class OpenAIWrapper extends base_wrapper_1.default {
         }
         catch (e) {
             helpers_1.default.handleException(span, e);
+            throw e;
         }
         finally {
             span.end();
-            // Record metrics after span has ended if parameters are available
             if (metricParams) {
                 base_wrapper_1.default.recordMetrics(span, metricParams);
             }
         }
     }
     static async _chatCompletionCommonSetter({ args, genAIEndpoint, result, span, ttft = 0, tbt = 0, }) {
-        const traceContent = config_1.default.traceContent;
-        const { messages, frequency_penalty = 0, max_tokens = null, n = 1, presence_penalty = 0, seed = null, stop = null, temperature = 1, top_p, user, stream = false, tools, } = args[0];
-        // Request Params attributes : Start
+        const captureContent = config_1.default.captureMessageContent;
+        const requestModel = args[0]?.model || 'gpt-4o';
+        const { messages, frequency_penalty = 0, max_tokens = null, n = 1, presence_penalty = 0, seed = null, stop = null, temperature = 1, top_p, user, stream = false, tools: _tools, service_tier, } = args[0];
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_TOP_P, top_p || 1);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_MAX_TOKENS, max_tokens || -1);
+        if (max_tokens != null) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_MAX_TOKENS, max_tokens);
+        }
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_TEMPERATURE, temperature);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_PRESENCE_PENALTY, presence_penalty);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FREQUENCY_PENALTY, frequency_penalty);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_SEED, seed ? String(seed) : '');
+        if (presence_penalty) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_PRESENCE_PENALTY, presence_penalty);
+        }
+        if (frequency_penalty) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FREQUENCY_PENALTY, frequency_penalty);
+        }
+        if (seed != null) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_SEED, Number(seed));
+        }
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IS_STREAM, stream);
         if (stop) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_STOP_SEQUENCES, Array.isArray(stop) ? stop : [stop]);
         }
-        if (user) {
-            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_USER, user);
+        if (n && n !== 1) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_CHOICE_COUNT, n);
+        }
+        if (service_tier && service_tier !== 'auto') {
+            span.setAttribute(semantic_convention_1.default.OPENAI_REQUEST_SERVICE_TIER, service_tier);
         }
-        if (traceContent) {
+        if (captureContent) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, helpers_1.default.buildInputMessages(messages || []));
         }
-        // Request Params attributes : End
-        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT);
         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_ID, result.id);
-        const model = result.model || 'gpt-3.5-turbo';
-        const responseModel = result.model || model;
-        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-        // Calculate cost of the operation
-        const cost = helpers_1.default.getChatModelCost(model, pricingInfo, result.usage.prompt_tokens, result.usage.completion_tokens);
+        const responseModel = result.model || requestModel;
+        const pricingInfo = config_1.default.pricingInfo || {};
+        const cost = helpers_1.default.getChatModelCost(requestModel, pricingInfo, result.usage.prompt_tokens, result.usage.completion_tokens);
         OpenAIWrapper.setBaseSpanAttributes(span, {
             genAIEndpoint,
-            model,
+            model: requestModel,
             user,
             cost,
             aiSystem: OpenAIWrapper.aiSystem,
             serverAddress: OpenAIWrapper.serverAddress,
             serverPort: OpenAIWrapper.serverPort,
         });
-        // Response model
         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_MODEL, responseModel);
-        // OpenAI-specific attributes
+        span.setAttribute(semantic_convention_1.default.OPENAI_API_TYPE, 'chat_completions');
         if (result.system_fingerprint) {
-            span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_SYSTEM_FINGERPRINT, result.system_fingerprint);
+            span.setAttribute(semantic_convention_1.default.OPENAI_RESPONSE_SYSTEM_FINGERPRINT, result.system_fingerprint);
         }
         if (result.service_tier) {
-            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_SERVICE_TIER, result.service_tier);
-        }
-        // Token usage
-        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS, result.usage.prompt_tokens);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_OUTPUT_TOKENS, result.usage.completion_tokens);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_TOTAL_TOKENS, result.usage.total_tokens);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_CLIENT_TOKEN_USAGE, result.usage.total_tokens);
-        // Enhanced token details
-        if (result.usage.completion_tokens_details) {
-            if (result.usage.completion_tokens_details.reasoning_tokens) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_REASONING_TOKENS, result.usage.completion_tokens_details.reasoning_tokens);
-                span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_COMPLETION_TOKENS_DETAILS_REASONING, result.usage.completion_tokens_details.reasoning_tokens);
-            }
-            if (result.usage.completion_tokens_details.audio_tokens) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_COMPLETION_TOKENS_DETAILS_AUDIO, result.usage.completion_tokens_details.audio_tokens);
-            }
+            span.setAttribute(semantic_convention_1.default.OPENAI_RESPONSE_SERVICE_TIER, result.service_tier);
         }
-        if (result.usage.prompt_tokens_details) {
-            if (result.usage.prompt_tokens_details.cached_tokens) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_PROMPT_TOKENS_DETAILS_CACHE_READ, result.usage.prompt_tokens_details.cached_tokens);
-            }
-            if (result.usage.prompt_tokens_details.audio_tokens) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_PROMPT_TOKENS_DETAILS_CACHE_WRITE, result.usage.prompt_tokens_details.audio_tokens);
-            }
+        const inputTokens = result.usage.prompt_tokens;
+        const outputTokens = result.usage.completion_tokens;
+        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS, inputTokens);
+        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_OUTPUT_TOKENS, outputTokens);
+        if (result.usage.prompt_tokens_details?.cached_tokens) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS, result.usage.prompt_tokens_details.cached_tokens);
+        }
+        if (result.usage.prompt_tokens_details?.cache_creation_tokens) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS, result.usage.prompt_tokens_details.cache_creation_tokens);
         }
-        // TTFT and TBT metrics
         if (ttft > 0) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_SERVER_TTFT, ttft);
         }
         if (tbt > 0) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_SERVER_TBT, tbt);
         }
-        // Finish reason
         if (result.choices[0].finish_reason) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_FINISH_REASON, [result.choices[0].finish_reason]);
         }
-        // Output type
         const outputType = typeof result.choices[0].message.content === 'string'
             ? semantic_convention_1.default.GEN_AI_OUTPUT_TYPE_TEXT
             : semantic_convention_1.default.GEN_AI_OUTPUT_TYPE_JSON;
         span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_TYPE, outputType);
-        // Tool calls handling
         if (result.choices[0].message.tool_calls) {
             const toolCalls = result.choices[0].message.tool_calls;
             const toolNames = toolCalls.map((t) => t.function?.name || '').filter(Boolean);
             const toolIds = toolCalls.map((t) => t.id || '').filter(Boolean);
             const toolArgs = toolCalls.map((t) => t.function?.arguments || '').filter(Boolean);
-            const toolTypes = toolCalls.map((t) => t.type || '').filter(Boolean);
             if (toolNames.length > 0) {
                 span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_NAME, toolNames.join(', '));
             }
@@ -296,23 +338,41 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_CALL_ID, toolIds.join(', '));
             }
             if (toolArgs.length > 0) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_CALL_ARGUMENTS, toolArgs);
-            }
-            if (toolTypes.length > 0) {
-                const toolTypesStr = toolTypes.join(', ');
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_TYPE, toolTypesStr);
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_TYPE_OTEL, toolTypesStr);
+                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_ARGS, toolArgs.join(', '));
             }
         }
-        // Content
-        if (traceContent) {
+        let inputMessagesJson;
+        let outputMessagesJson;
+        if (captureContent) {
             const toolCalls = result.choices[0].message.tool_calls;
-            const outputJson = helpers_1.default.buildOutputMessages(result.choices[0].message.content || '', result.choices[0].finish_reason || 'stop', toolCalls);
-            span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES, outputJson);
+            outputMessagesJson = helpers_1.default.buildOutputMessages(result.choices[0].message.content || '', result.choices[0].finish_reason || 'stop', toolCalls);
+            span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES, outputMessagesJson);
+            inputMessagesJson = helpers_1.default.buildInputMessages(messages || []);
+        }
+        if (!config_1.default.disableEvents) {
+            const eventAttrs = {
+                [semantic_convention_1.default.GEN_AI_OPERATION]: semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT,
+                [semantic_convention_1.default.GEN_AI_REQUEST_MODEL]: requestModel,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_MODEL]: responseModel,
+                [semantic_convention_1.default.SERVER_ADDRESS]: OpenAIWrapper.serverAddress,
+                [semantic_convention_1.default.SERVER_PORT]: OpenAIWrapper.serverPort,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_ID]: result.id,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_FINISH_REASON]: [result.choices[0].finish_reason],
+                [semantic_convention_1.default.GEN_AI_OUTPUT_TYPE]: outputType,
+                [semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS]: inputTokens,
+                [semantic_convention_1.default.GEN_AI_USAGE_OUTPUT_TOKENS]: outputTokens,
+            };
+            if (captureContent) {
+                if (inputMessagesJson)
+                    eventAttrs[semantic_convention_1.default.GEN_AI_INPUT_MESSAGES] = inputMessagesJson;
+                if (outputMessagesJson)
+                    eventAttrs[semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES] = outputMessagesJson;
+            }
+            helpers_1.default.emitInferenceEvent(span, eventAttrs);
         }
         return {
             genAIEndpoint,
-            model,
+            model: requestModel,
             user,
             cost,
             aiSystem: OpenAIWrapper.aiSystem,
@@ -320,23 +380,29 @@ class OpenAIWrapper extends base_wrapper_1.default {
     }
     static _patchEmbedding(tracer) {
         const genAIEndpoint = 'openai.resources.embeddings';
-        const traceContent = config_1.default.traceContent;
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
-                return api_1.context.with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'text-embedding-ada-002';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_EMBEDDING} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_EMBEDDING, requestModel),
+                }, effectiveCtx);
+                return api_1.context.with(api_1.trace.setSpan(effectiveCtx, span), async () => {
+                    const captureContent = config_1.default.captureMessageContent;
                     let metricParams;
                     try {
                         const response = await originalMethod.apply(this, args);
-                        const model = response.model || 'text-embedding-ada-002';
-                        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-                        const cost = helpers_1.default.getEmbedModelCost(model, pricingInfo, response.usage.prompt_tokens);
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_EMBEDDING);
+                        const _responseModel = response.model || requestModel;
+                        const pricingInfo = config_1.default.pricingInfo || {};
+                        const cost = helpers_1.default.getEmbedModelCost(requestModel, pricingInfo, response.usage.prompt_tokens);
                         const { dimensions, encoding_format = 'float', input, user } = args[0];
-                        // Set base span attributes
                         OpenAIWrapper.setBaseSpanAttributes(span, {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
@@ -344,25 +410,18 @@ class OpenAIWrapper extends base_wrapper_1.default {
                             serverPort: OpenAIWrapper.serverPort,
                         });
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IS_STREAM, false);
-                        // Request Params attributes : Start
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_ENCODING_FORMATS, [encoding_format]);
                         if (dimensions) {
-                            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_EMBEDDING_DIMENSION, dimensions);
+                            span.setAttribute(semantic_convention_1.default.GEN_AI_EMBEDDINGS_DIMENSION_COUNT, dimensions);
                         }
-                        if (user) {
-                            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_USER, user);
-                        }
-                        if (traceContent) {
+                        if (captureContent) {
                             const formattedInput = typeof input === 'string' ? input : JSON.stringify(input);
                             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, formattedInput);
                         }
-                        // Request Params attributes : End
                         span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS, response.usage.prompt_tokens);
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_TOTAL_TOKENS, response.usage.total_tokens);
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_CLIENT_TOKEN_USAGE, response.usage.prompt_tokens);
                         metricParams = {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
@@ -387,38 +446,40 @@ class OpenAIWrapper extends base_wrapper_1.default {
         const genAIEndpoint = 'openai.resources.fine_tuning.jobs';
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
-                return api_1.context.with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'gpt-3.5-turbo';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_FINETUNING} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_FINETUNING, requestModel),
+                }, effectiveCtx);
+                return api_1.context.with(api_1.trace.setSpan(effectiveCtx, span), async () => {
                     let metricParams;
                     try {
                         const response = await originalMethod.apply(this, args);
-                        const model = response.model || 'gpt-3.5-turbo';
                         const { hyperparameters = {}, suffix = '', training_file, user, validation_file, } = args[0];
-                        // Set base span attributes
                         OpenAIWrapper.setBaseSpanAttributes(span, {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             aiSystem: OpenAIWrapper.aiSystem,
                             serverAddress: OpenAIWrapper.serverAddress,
                             serverPort: OpenAIWrapper.serverPort,
                         });
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_FINETUNING);
-                        // Request Params attributes : Start
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_TRAINING_FILE, training_file);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_VALIDATION_FILE, validation_file);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FINETUNE_BATCH_SIZE, hyperparameters?.batch_size || 'auto');
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FINETUNE_MODEL_LRM, hyperparameters?.learning_rate_multiplier || 'auto');
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FINETUNE_MODEL_EPOCHS, hyperparameters?.n_epochs || 'auto');
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FINETUNE_MODEL_SUFFIX, suffix);
-                        // Request Params attributes : End
                         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_ID, response.id);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS, response.usage.prompt_tokens);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_FINETUNE_STATUS, response.status);
-                        // Store metric parameters for use after span ends
                         metricParams = {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             aiSystem: OpenAIWrapper.aiSystem,
                         };
@@ -426,10 +487,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
                     }
                     catch (e) {
                         helpers_1.default.handleException(span, e);
+                        throw e;
                     }
                     finally {
                         span.end();
-                        // Record metrics after span has ended if parameters are available
                         if (metricParams) {
                             base_wrapper_1.default.recordMetrics(span, metricParams);
                         }
@@ -440,51 +501,56 @@ class OpenAIWrapper extends base_wrapper_1.default {
     }
     static _patchImageGenerate(tracer) {
         const genAIEndpoint = 'openai.resources.images';
-        const traceContent = config_1.default.traceContent;
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
-                return api_1.context.with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'dall-e-2';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE, requestModel),
+                }, effectiveCtx);
+                return api_1.context.with(api_1.trace.setSpan(effectiveCtx, span), async () => {
+                    const captureContent = config_1.default.captureMessageContent;
                     let metricParams;
                     try {
                         const response = await originalMethod.apply(this, args);
                         const { prompt, quality = 'standard', response_format = 'url', size = '1024x1024', style = 'vivid', user, } = args[0];
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE);
-                        const model = response.model || 'dall-e-2';
-                        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-                        // Calculate cost of the operation
+                        const responseModel = response.model || requestModel;
+                        const pricingInfo = config_1.default.pricingInfo || {};
                         const cost = (response.data?.length || 1) *
-                            helpers_1.default.getImageModelCost(model, pricingInfo, size, quality);
+                            helpers_1.default.getImageModelCost(responseModel, pricingInfo, size, quality);
                         OpenAIWrapper.setBaseSpanAttributes(span, {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
                             serverAddress: OpenAIWrapper.serverAddress,
                             serverPort: OpenAIWrapper.serverPort,
                         });
-                        // Request Params attributes : Start
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_SIZE, size);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_QUALITY, quality);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_STYLE, style);
-                        if (traceContent) {
+                        if (captureContent) {
                             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, prompt);
                         }
-                        // Request Params attributes : End
-                        let imagesCount = 0;
                         if (response.data) {
+                            const imageUrls = [];
+                            const revisedPrompts = [];
                             for (const items of response.data) {
-                                span.setAttribute(`${semantic_convention_1.default.GEN_AI_CONTENT_REVISED_PROMPT}.${imagesCount}`, items.revised_prompt || '');
-                                const attributeName = `${semantic_convention_1.default.GEN_AI_RESPONSE_IMAGE}.${imagesCount}`;
-                                span.setAttribute(attributeName, items[response_format]);
-                                imagesCount++;
+                                revisedPrompts.push(items.revised_prompt || '');
+                                const value = items[response_format];
+                                imageUrls.push(value && !String(value).startsWith('data:') ? value : '[base64_image_data]');
                             }
+                            span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_IMAGE, imageUrls);
+                            span.setAttribute(semantic_convention_1.default.GEN_AI_CONTENT_REVISED_PROMPT, revisedPrompts);
                         }
-                        // Store metric parameters for use after span ends
                         metricParams = {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
@@ -493,10 +559,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
                     }
                     catch (e) {
                         helpers_1.default.handleException(span, e);
+                        throw e;
                     }
                     finally {
                         span.end();
-                        // Record metrics after span has ended if parameters are available
                         if (metricParams) {
                             base_wrapper_1.default.recordMetrics(span, metricParams);
                         }
@@ -507,52 +573,57 @@ class OpenAIWrapper extends base_wrapper_1.default {
     }
     static _patchImageVariation(tracer) {
         const genAIEndpoint = 'openai.resources.images';
-        const traceContent = config_1.default.traceContent;
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
-                return api_1.context.with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'dall-e-2';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE, requestModel),
+                }, effectiveCtx);
+                return api_1.context.with(api_1.trace.setSpan(effectiveCtx, span), async () => {
+                    const captureContent = config_1.default.captureMessageContent;
                     let metricParams;
                     try {
                         const response = await originalMethod.apply(this, args);
                         const { prompt, quality = 'standard', response_format = 'url', size = '1024x1024', style = 'vivid', user, } = args[0];
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_IMAGE);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_ID, response.created);
-                        const model = response.model || 'dall-e-2';
-                        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-                        // Calculate cost of the operation
+                        const responseModel = response.model || requestModel;
+                        const pricingInfo = config_1.default.pricingInfo || {};
                         const cost = (response.data?.length || 1) *
-                            helpers_1.default.getImageModelCost(model, pricingInfo, size, quality);
+                            helpers_1.default.getImageModelCost(responseModel, pricingInfo, size, quality);
                         OpenAIWrapper.setBaseSpanAttributes(span, {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
                             serverAddress: OpenAIWrapper.serverAddress,
                             serverPort: OpenAIWrapper.serverPort,
                         });
-                        // Request Params attributes : Start
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_SIZE, size);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_QUALITY, quality);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IMAGE_STYLE, style);
-                        if (traceContent) {
+                        if (captureContent) {
                             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, prompt);
                         }
-                        // Request Params attributes : End
-                        let imagesCount = 0;
                         if (response.data) {
+                            const imageUrls = [];
+                            const revisedPrompts = [];
                             for (const items of response.data) {
-                                span.setAttribute(`${semantic_convention_1.default.GEN_AI_CONTENT_REVISED_PROMPT}.${imagesCount}`, items.revised_prompt || '');
-                                const attributeName = `${semantic_convention_1.default.GEN_AI_RESPONSE_IMAGE}.${imagesCount}`;
-                                span.setAttribute(attributeName, items[response_format]);
-                                imagesCount++;
+                                revisedPrompts.push(items.revised_prompt || '');
+                                const value = items[response_format];
+                                imageUrls.push(value && !String(value).startsWith('data:') ? value : '[base64_image_data]');
                             }
+                            span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_IMAGE, imageUrls);
+                            span.setAttribute(semantic_convention_1.default.GEN_AI_CONTENT_REVISED_PROMPT, revisedPrompts);
                         }
-                        // Store metric parameters for use after span ends
                         metricParams = {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
@@ -561,10 +632,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
                     }
                     catch (e) {
                         helpers_1.default.handleException(span, e);
+                        throw e;
                     }
                     finally {
                         span.end();
-                        // Record metrics after span has ended if parameters are available
                         if (metricParams) {
                             base_wrapper_1.default.recordMetrics(span, metricParams);
                         }
@@ -575,41 +646,44 @@ class OpenAIWrapper extends base_wrapper_1.default {
     }
     static _patchAudioCreate(tracer) {
         const genAIEndpoint = 'openai.resources.audio.speech';
-        const traceContent = config_1.default.traceContent;
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
-                return api_1.context.with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'tts-1';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_AUDIO} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_AUDIO, requestModel),
+                }, effectiveCtx);
+                return api_1.context.with(api_1.trace.setSpan(effectiveCtx, span), async () => {
+                    const captureContent = config_1.default.captureMessageContent;
                     let metricParams;
                     try {
                         const response = await originalMethod.apply(this, args);
                         const { input, user, voice, response_format = 'mp3', speed = 1 } = args[0];
-                        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_AUDIO);
-                        const model = response.model || 'tts-1';
-                        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-                        // Calculate cost of the operation
-                        const cost = helpers_1.default.getAudioModelCost(model, pricingInfo, input);
+                        const responseModel = response.model || requestModel;
+                        const pricingInfo = config_1.default.pricingInfo || {};
+                        const cost = helpers_1.default.getAudioModelCost(responseModel, pricingInfo, input);
                         OpenAIWrapper.setBaseSpanAttributes(span, {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
                             serverAddress: OpenAIWrapper.serverAddress,
                             serverPort: OpenAIWrapper.serverPort,
                         });
-                        // Request Params attributes : Start
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_AUDIO_VOICE, voice);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_AUDIO_RESPONSE_FORMAT, response_format);
                         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_AUDIO_SPEED, speed);
-                        if (traceContent) {
+                        if (captureContent) {
                             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, input);
                         }
-                        // Request Params attributes : End
-                        // Store metric parameters for use after span ends
                         metricParams = {
                             genAIEndpoint,
-                            model,
+                            model: requestModel,
                             user,
                             cost,
                             aiSystem: OpenAIWrapper.aiSystem,
@@ -618,10 +692,10 @@ class OpenAIWrapper extends base_wrapper_1.default {
                     }
                     catch (e) {
                         helpers_1.default.handleException(span, e);
+                        throw e;
                     }
                     finally {
                         span.end();
-                        // Record metrics after span has ended if parameters are available
                         if (metricParams) {
                             base_wrapper_1.default.recordMetrics(span, metricParams);
                         }
@@ -634,9 +708,17 @@ class OpenAIWrapper extends base_wrapper_1.default {
         const genAIEndpoint = 'openai.resources.responses';
         return (originalMethod) => {
             return async function (...args) {
-                const span = tracer.startSpan(genAIEndpoint, { kind: api_1.SpanKind.CLIENT });
+                if ((0, helpers_1.isFrameworkLlmActive)())
+                    return originalMethod.apply(this, args);
+                const requestModel = args[0]?.model || 'gpt-4o';
+                const spanName = `${semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT} ${requestModel}`;
+                const effectiveCtx = (0, helpers_1.getFrameworkParentContext)() ?? api_1.context.active();
+                const span = tracer.startSpan(spanName, {
+                    kind: api_1.SpanKind.CLIENT,
+                    attributes: spanCreationAttrs(semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT, requestModel),
+                }, effectiveCtx);
                 return api_1.context
-                    .with(api_1.trace.setSpan(api_1.context.active(), span), async () => {
+                    .with(api_1.trace.setSpan(effectiveCtx, span), async () => {
                     return originalMethod.apply(this, args);
                 })
                     .then((response) => {
@@ -653,7 +735,16 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 })
                     .catch((e) => {
                     helpers_1.default.handleException(span, e);
+                    base_wrapper_1.default.recordMetrics(span, {
+                        genAIEndpoint,
+                        model: requestModel,
+                        aiSystem: OpenAIWrapper.aiSystem,
+                        serverAddress: OpenAIWrapper.serverAddress,
+                        serverPort: OpenAIWrapper.serverPort,
+                        errorType: e?.constructor?.name || '_OTHER',
+                    });
                     span.end();
+                    throw e;
                 });
             };
         };
@@ -671,6 +762,7 @@ class OpenAIWrapper extends base_wrapper_1.default {
         }
         catch (e) {
             helpers_1.default.handleException(span, e);
+            throw e;
         }
         finally {
             span.end();
@@ -684,7 +776,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
         const timestamps = [];
         const startTime = Date.now();
         try {
-            const { input } = args[0];
             const result = {
                 id: '',
                 model: '',
@@ -700,7 +791,7 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 },
             };
             let llmResponse = '';
-            let responseTools = [];
+            const responseTools = [];
             for await (const chunk of response) {
                 timestamps.push(Date.now());
                 if (chunk.type === 'response.output_text.delta') {
@@ -740,7 +831,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 }
                 yield chunk;
             }
-            // Construct output array
             if (llmResponse) {
                 result.output.push({
                     type: 'message',
@@ -750,7 +840,6 @@ class OpenAIWrapper extends base_wrapper_1.default {
             if (responseTools.length > 0) {
                 result.output.push(...responseTools);
             }
-            // Calculate TTFT and TBT
             const ttft = timestamps.length > 0 ? (timestamps[0] - startTime) / 1000 : 0;
             let tbt = 0;
             if (timestamps.length > 1) {
@@ -769,6 +858,7 @@ class OpenAIWrapper extends base_wrapper_1.default {
         }
         catch (e) {
             helpers_1.default.handleException(span, e);
+            throw e;
         }
         finally {
             span.end();
@@ -778,65 +868,57 @@ class OpenAIWrapper extends base_wrapper_1.default {
         }
     }
     static async _responsesCommonSetter({ args, genAIEndpoint, result, span, ttft = 0, tbt = 0, }) {
-        const traceContent = config_1.default.traceContent;
+        const captureContent = config_1.default.captureMessageContent;
+        const requestModel = args[0]?.model || 'gpt-4o';
         const { input, temperature = 1.0, top_p = 1.0, max_output_tokens, reasoning, stream = false, } = args[0];
-        // Normalize Responses API input to messages array for buildInputMessages
         const responsesMessages = typeof input === 'string'
             ? [{ role: 'user', content: input }]
             : (Array.isArray(input) ? input : []);
-        // Request Params attributes
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_TEMPERATURE, temperature);
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_TOP_P, top_p);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_MAX_TOKENS, max_output_tokens || -1);
+        if (max_output_tokens != null) {
+            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_MAX_TOKENS, max_output_tokens);
+        }
         span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_IS_STREAM, stream);
         if (reasoning?.effort) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_REASONING_EFFORT, reasoning.effort);
         }
-        if (traceContent) {
+        if (captureContent) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_INPUT_MESSAGES, helpers_1.default.buildInputMessages(responsesMessages));
         }
-        span.setAttribute(semantic_convention_1.default.GEN_AI_OPERATION, semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT);
-        const model = result.model || 'gpt-4o';
-        const responseModel = result.model || model;
-        const pricingInfo = await config_1.default.updatePricingJson(config_1.default.pricing_json);
-        // Calculate cost
+        const responseModel = result.model || requestModel;
+        const pricingInfo = config_1.default.pricingInfo || {};
         const inputTokens = result.usage?.input_tokens || 0;
         const outputTokens = result.usage?.output_tokens || 0;
-        const cost = helpers_1.default.getChatModelCost(model, pricingInfo, inputTokens, outputTokens);
+        const cost = helpers_1.default.getChatModelCost(requestModel, pricingInfo, inputTokens, outputTokens);
         OpenAIWrapper.setBaseSpanAttributes(span, {
             genAIEndpoint,
-            model,
+            model: requestModel,
             user: '',
             cost,
             aiSystem: OpenAIWrapper.aiSystem,
             serverAddress: OpenAIWrapper.serverAddress,
             serverPort: OpenAIWrapper.serverPort,
         });
-        // Response attributes
         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_ID, result.id);
         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_MODEL, responseModel);
         span.setAttribute(semantic_convention_1.default.GEN_AI_RESPONSE_FINISH_REASON, [result.status || 'completed']);
         span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_TYPE, semantic_convention_1.default.GEN_AI_OUTPUT_TYPE_TEXT);
+        span.setAttribute(semantic_convention_1.default.OPENAI_API_TYPE, 'responses');
         if (result.service_tier) {
-            span.setAttribute(semantic_convention_1.default.GEN_AI_REQUEST_SERVICE_TIER, result.service_tier);
+            span.setAttribute(semantic_convention_1.default.OPENAI_RESPONSE_SERVICE_TIER, result.service_tier);
         }
-        // Token usage
         span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS, inputTokens);
         span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_OUTPUT_TOKENS, outputTokens);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_TOTAL_TOKENS, inputTokens + outputTokens);
-        span.setAttribute(semantic_convention_1.default.GEN_AI_CLIENT_TOKEN_USAGE, inputTokens + outputTokens);
-        // Reasoning tokens
         if (result.usage?.output_tokens_details?.reasoning_tokens) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_USAGE_REASONING_TOKENS, result.usage.output_tokens_details.reasoning_tokens);
         }
-        // TTFT and TBT metrics
         if (ttft > 0) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_SERVER_TTFT, ttft);
         }
         if (tbt > 0) {
             span.setAttribute(semantic_convention_1.default.GEN_AI_SERVER_TBT, tbt);
         }
-        // Extract completion text from output
         let completionText = '';
         if (result.output && Array.isArray(result.output)) {
             for (const item of result.output) {
@@ -849,13 +931,11 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 }
             }
         }
-        // Tool calls handling for Responses API
         const toolCalls = result.tools || [];
         if (toolCalls.length > 0) {
             const toolNames = toolCalls.map((t) => t.name || '').filter(Boolean);
             const toolIds = toolCalls.map((t) => t.call_id || '').filter(Boolean);
             const toolArgs = toolCalls.map((t) => t.arguments || '').filter(Boolean);
-            const toolTypes = toolCalls.map((t) => t.type || '').filter(Boolean);
             if (toolNames.length > 0) {
                 span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_NAME, toolNames.join(', '));
             }
@@ -863,21 +943,40 @@ class OpenAIWrapper extends base_wrapper_1.default {
                 span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_CALL_ID, toolIds.join(', '));
             }
             if (toolArgs.length > 0) {
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_CALL_ARGUMENTS, toolArgs.join(', '));
-            }
-            if (toolTypes.length > 0) {
-                const toolTypesStr = toolTypes.join(', ');
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_TYPE, toolTypesStr);
-                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_TYPE_OTEL, toolTypesStr);
+                span.setAttribute(semantic_convention_1.default.GEN_AI_TOOL_ARGS, toolArgs.join(', '));
             }
         }
-        // Content
-        if (traceContent) {
-            span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES, helpers_1.default.buildOutputMessages(completionText, result.status || 'stop'));
+        let inputMessagesJson;
+        let outputMessagesJson;
+        if (captureContent) {
+            outputMessagesJson = helpers_1.default.buildOutputMessages(completionText, result.status || 'stop');
+            span.setAttribute(semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES, outputMessagesJson);
+            inputMessagesJson = helpers_1.default.buildInputMessages(responsesMessages);
+        }
+        if (!config_1.default.disableEvents) {
+            const eventAttrs = {
+                [semantic_convention_1.default.GEN_AI_OPERATION]: semantic_convention_1.default.GEN_AI_OPERATION_TYPE_CHAT,
+                [semantic_convention_1.default.GEN_AI_REQUEST_MODEL]: requestModel,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_MODEL]: responseModel,
+                [semantic_convention_1.default.SERVER_ADDRESS]: OpenAIWrapper.serverAddress,
+                [semantic_convention_1.default.SERVER_PORT]: OpenAIWrapper.serverPort,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_ID]: result.id,
+                [semantic_convention_1.default.GEN_AI_RESPONSE_FINISH_REASON]: [result.status || 'completed'],
+                [semantic_convention_1.default.GEN_AI_OUTPUT_TYPE]: semantic_convention_1.default.GEN_AI_OUTPUT_TYPE_TEXT,
+                [semantic_convention_1.default.GEN_AI_USAGE_INPUT_TOKENS]: inputTokens,
+                [semantic_convention_1.default.GEN_AI_USAGE_OUTPUT_TOKENS]: outputTokens,
+            };
+            if (captureContent) {
+                if (inputMessagesJson)
+                    eventAttrs[semantic_convention_1.default.GEN_AI_INPUT_MESSAGES] = inputMessagesJson;
+                if (outputMessagesJson)
+                    eventAttrs[semantic_convention_1.default.GEN_AI_OUTPUT_MESSAGES] = outputMessagesJson;
+            }
+            helpers_1.default.emitInferenceEvent(span, eventAttrs);
         }
         return {
             genAIEndpoint,
-            model,
+            model: requestModel,
             user: '',
             cost,
             aiSystem: OpenAIWrapper.aiSystem,