@elasticdash/openai 0.0.1 → 0.0.2

package/LICENSE

@@ -0,0 +1,24 @@
+MIT License
+
+This repository is licensed under the MIT License. Portions of this codebase are derived from posthog/posthog-js-lite by PostHog, which is also licensed under the MIT License.
+
+Copyright (c) 2023-2026 Langfuse GmbH
+Copyright (c) 2022 PostHog (part of Hiberly Inc)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -6,8 +6,8 @@ This is the OpenAI integration package of the Langfuse JS SDK containing the `ob
 
 ## Packages
 
-| Package                                     | NPM                                                                                                               | Description                                               | Environments |
-| ------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- | ------------ |
+| Package                                        | NPM                                                                                                                     | Description                                               | Environments |
+| ---------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- | ------------ |
 | [@elasticdash/client](./packages/client)       | [![NPM](https://img.shields.io/npm/v/@elasticdash/client.svg)](https://www.npmjs.com/package/@elasticdash/client)       | Langfuse API client for universal JavaScript environments | Universal JS |
 | [@elasticdash/tracing](./packages/tracing)     | [![NPM](https://img.shields.io/npm/v/@elasticdash/tracing.svg)](https://www.npmjs.com/package/@elasticdash/tracing)     | Langfuse instrumentation methods based on OpenTelemetry   | Node.js 20+  |
 | [@elasticdash/otel](./packages/otel)           | [![NPM](https://img.shields.io/npm/v/@elasticdash/otel.svg)](https://www.npmjs.com/package/@elasticdash/otel)           | Langfuse OpenTelemetry export helpers                     | Node.js 20+  |

package/dist/index.cjs

@@ -0,0 +1,428 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+
+// src/index.ts
+var index_exports = {};
+__export(index_exports, {
+  observeOpenAI: () => observeOpenAI
+});
+module.exports = __toCommonJS(index_exports);
+
+// src/traceMethod.ts
+var import_tracing = require("@elasticdash/tracing");
+
+// src/parseOpenAI.ts
+var parseInputArgs = (args) => {
+  let params = {};
+  params = {
+    frequency_penalty: args.frequency_penalty,
+    logit_bias: args.logit_bias,
+    logprobs: args.logprobs,
+    max_tokens: args.max_tokens,
+    n: args.n,
+    presence_penalty: args.presence_penalty,
+    seed: args.seed,
+    stop: args.stop,
+    stream: args.stream,
+    temperature: args.temperature,
+    top_p: args.top_p,
+    user: args.user,
+    response_format: args.response_format,
+    top_logprobs: args.top_logprobs
+  };
+  let input = args.input;
+  if (args && typeof args === "object" && !Array.isArray(args) && "messages" in args) {
+    input = {};
+    input.messages = args.messages;
+    if ("function_call" in args) {
+      input.function_call = args.function_call;
+    }
+    if ("functions" in args) {
+      input.functions = args.functions;
+    }
+    if ("tools" in args) {
+      input.tools = args.tools;
+    }
+    if ("tool_choice" in args) {
+      input.tool_choice = args.tool_choice;
+    }
+  } else if (!input) {
+    input = args.prompt;
+  }
+  return {
+    model: args.model,
+    input,
+    modelParameters: params
+  };
+};
+var parseCompletionOutput = (res) => {
+  var _a;
+  if (res instanceof Object && "output_text" in res && res["output_text"] !== "") {
+    return res["output_text"];
+  }
+  if (typeof res === "object" && res && "output" in res && Array.isArray(res["output"])) {
+    const output = res["output"];
+    if (output.length > 1) {
+      return output;
+    }
+    if (output.length === 1) {
+      return output[0];
+    }
+    return null;
+  }
+  if (!(res instanceof Object && "choices" in res && Array.isArray(res.choices))) {
+    return "";
+  }
+  return "message" in res.choices[0] ? res.choices[0].message : (_a = res.choices[0].text) != null ? _a : "";
+};
+var parseUsageDetails = (completionUsage) => {
+  if ("prompt_tokens" in completionUsage) {
+    const {
+      prompt_tokens,
+      completion_tokens,
+      total_tokens,
+      completion_tokens_details,
+      prompt_tokens_details
+    } = completionUsage;
+    const flatPromptTokensDetails = Object.fromEntries(
+      Object.entries(prompt_tokens_details != null ? prompt_tokens_details : {}).map(([key, value]) => [
+        `input_${key}`,
+        value
+      ])
+    );
+    const flatCompletionTokensDetails = Object.fromEntries(
+      Object.entries(completion_tokens_details != null ? completion_tokens_details : {}).map(([key, value]) => [
+        `output_${key}`,
+        value
+      ])
+    );
+    let finalInputTokens = prompt_tokens;
+    Object.values(flatPromptTokensDetails).forEach((value) => {
+      finalInputTokens = Math.max(finalInputTokens - value, 0);
+    });
+    let finalOutputTokens = completion_tokens;
+    Object.values(flatCompletionTokensDetails).forEach((value) => {
+      finalOutputTokens = Math.max(finalOutputTokens - value, 0);
+    });
+    return {
+      input: finalInputTokens,
+      output: finalOutputTokens,
+      total: total_tokens,
+      ...flatPromptTokensDetails,
+      ...flatCompletionTokensDetails
+    };
+  } else if ("input_tokens" in completionUsage) {
+    const {
+      input_tokens,
+      output_tokens,
+      total_tokens,
+      input_tokens_details,
+      output_tokens_details
+    } = completionUsage;
+    let finalInputTokens = input_tokens;
+    Object.keys(input_tokens_details != null ? input_tokens_details : {}).forEach((key) => {
+      finalInputTokens = Math.max(
+        finalInputTokens - input_tokens_details[key],
+        0
+      );
+    });
+    let finalOutputTokens = output_tokens;
+    Object.keys(output_tokens_details != null ? output_tokens_details : {}).forEach((key) => {
+      finalOutputTokens = Math.max(
+        finalOutputTokens - output_tokens_details[key],
+        0
+      );
+    });
+    return {
+      input: finalInputTokens,
+      output: finalOutputTokens,
+      total: total_tokens,
+      ...Object.fromEntries(
+        Object.entries(input_tokens_details != null ? input_tokens_details : {}).map(([key, value]) => [
+          `input_${key}`,
+          value
+        ])
+      ),
+      ...Object.fromEntries(
+        Object.entries(output_tokens_details != null ? output_tokens_details : {}).map(([key, value]) => [
+          `output_${key}`,
+          value
+        ])
+      )
+    };
+  }
+};
+var parseUsageDetailsFromResponse = (res) => {
+  if (hasCompletionUsage(res)) {
+    return parseUsageDetails(res.usage);
+  }
+};
+var parseChunk = (rawChunk) => {
+  var _a, _b;
+  let isToolCall = false;
+  const _chunk = rawChunk;
+  const chunkData = (_a = _chunk == null ? void 0 : _chunk.choices) == null ? void 0 : _a[0];
+  try {
+    if ("delta" in chunkData && "tool_calls" in chunkData.delta && Array.isArray(chunkData.delta.tool_calls)) {
+      isToolCall = true;
+      return { isToolCall, data: chunkData.delta.tool_calls[0] };
+    }
+    if ("delta" in chunkData) {
+      return { isToolCall, data: ((_b = chunkData.delta) == null ? void 0 : _b.content) || "" };
+    }
+    if ("text" in chunkData) {
+      return { isToolCall, data: chunkData.text || "" };
+    }
+  } catch {
+  }
+  return { isToolCall: false, data: "" };
+};
+function hasCompletionUsage(obj) {
+  return obj instanceof Object && "usage" in obj && obj.usage instanceof Object && // Completion API Usage format
+  (typeof obj.usage.prompt_tokens === "number" && typeof obj.usage.completion_tokens === "number" && typeof obj.usage.total_tokens === "number" || // Response API Usage format
+  typeof obj.usage.input_tokens === "number" && typeof obj.usage.output_tokens === "number" && typeof obj.usage.total_tokens === "number");
+}
+var getToolCallOutput = (toolCallChunks) => {
+  var _a, _b;
+  let name = "";
+  let toolArguments = "";
+  for (const toolCall of toolCallChunks) {
+    name = ((_a = toolCall.function) == null ? void 0 : _a.name) || name;
+    toolArguments += ((_b = toolCall.function) == null ? void 0 : _b.arguments) || "";
+  }
+  return {
+    tool_calls: [
+      {
+        function: {
+          name,
+          arguments: toolArguments
+        }
+      }
+    ]
+  };
+};
+var parseModelDataFromResponse = (res) => {
+  if (typeof res !== "object" || res === null) {
+    return {
+      model: void 0,
+      modelParameters: void 0,
+      metadata: void 0
+    };
+  }
+  const model = "model" in res ? res["model"] : void 0;
+  const modelParameters = {};
+  const modelParamKeys = [
+    "max_output_tokens",
+    "parallel_tool_calls",
+    "store",
+    "temperature",
+    "tool_choice",
+    "top_p",
+    "truncation",
+    "user"
+  ];
+  const metadata = {};
+  const metadataKeys = [
+    "reasoning",
+    "incomplete_details",
+    "instructions",
+    "previous_response_id",
+    "tools",
+    "metadata",
+    "status",
+    "error"
+  ];
+  for (const key of modelParamKeys) {
+    const val = key in res ? res[key] : null;
+    if (val !== null && val !== void 0) {
+      modelParameters[key] = val;
+    }
+  }
+  for (const key of metadataKeys) {
+    const val = key in res ? res[key] : null;
+    if (val) {
+      metadata[key] = val;
+    }
+  }
+  return {
+    model,
+    modelParameters: Object.keys(modelParameters).length > 0 ? modelParameters : void 0,
+    metadata: Object.keys(metadata).length > 0 ? metadata : void 0
+  };
+};
+
+// src/utils.ts
+var isAsyncIterable = (x) => x != null && typeof x === "object" && typeof x[Symbol.asyncIterator] === "function";
+
+// src/traceMethod.ts
+var withTracing = (tracedMethod, config) => {
+  return (...args) => wrapMethod(tracedMethod, config, ...args);
+};
+var wrapMethod = (tracedMethod, config, ...args) => {
+  var _a, _b;
+  const { model, input, modelParameters } = parseInputArgs((_a = args[0]) != null ? _a : {});
+  const finalModelParams = { ...modelParameters, response_format: "" };
+  const finalMetadata = {
+    ...config == null ? void 0 : config.generationMetadata,
+    response_format: "response_format" in modelParameters ? modelParameters.response_format : void 0
+  };
+  const generation = (0, import_tracing.startObservation)(
+    (_b = config == null ? void 0 : config.generationName) != null ? _b : "OpenAI-completion",
+    {
+      model,
+      input,
+      modelParameters: finalModelParams,
+      prompt: config == null ? void 0 : config.langfusePrompt,
+      metadata: finalMetadata
+    },
+    {
+      asType: "generation",
+      parentSpanContext: config == null ? void 0 : config.parentSpanContext
+    }
+  ).updateTrace({
+    userId: config == null ? void 0 : config.userId,
+    sessionId: config == null ? void 0 : config.sessionId,
+    tags: config == null ? void 0 : config.tags,
+    name: config == null ? void 0 : config.traceName
+  });
+  try {
+    const res = tracedMethod(...args);
+    if (isAsyncIterable(res)) {
+      return wrapAsyncIterable(res, generation);
+    }
+    if (res instanceof Promise) {
+      const wrappedPromise = res.then((result) => {
+        if (isAsyncIterable(result)) {
+          return wrapAsyncIterable(result, generation);
+        }
+        const output = parseCompletionOutput(result);
+        const usageDetails = parseUsageDetailsFromResponse(result);
+        const {
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        } = parseModelDataFromResponse(result);
+        generation.update({
+          output,
+          usageDetails,
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        }).end();
+        return result;
+      }).catch((err) => {
+        generation.update({
+          statusMessage: String(err),
+          level: "ERROR",
+          costDetails: {
+            input: 0,
+            output: 0,
+            total: 0
+          }
+        }).end();
+        throw err;
+      });
+      return wrappedPromise;
+    }
+    return res;
+  } catch (error) {
+    generation.update({
+      statusMessage: String(error),
+      level: "ERROR",
+      costDetails: {
+        input: 0,
+        output: 0,
+        total: 0
+      }
+    }).end();
+    throw error;
+  }
+};
+function wrapAsyncIterable(iterable, generation) {
+  async function* tracedOutputGenerator() {
+    const response = iterable;
+    const textChunks = [];
+    const toolCallChunks = [];
+    let usage = null;
+    let completionStartTime = void 0;
+    let usageDetails = void 0;
+    let output = null;
+    for await (const rawChunk of response) {
+      completionStartTime = completionStartTime != null ? completionStartTime : /* @__PURE__ */ new Date();
+      if (typeof rawChunk === "object" && rawChunk && "response" in rawChunk) {
+        const result = rawChunk["response"];
+        output = parseCompletionOutput(result);
+        usageDetails = parseUsageDetailsFromResponse(result);
+        const {
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        } = parseModelDataFromResponse(result);
+        generation.update({
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        });
+      }
+      if (typeof rawChunk === "object" && rawChunk != null && "usage" in rawChunk) {
+        usage = rawChunk.usage;
+      }
+      const processedChunk = parseChunk(rawChunk);
+      if (!processedChunk.isToolCall) {
+        textChunks.push(processedChunk.data);
+      } else {
+        toolCallChunks.push(processedChunk.data);
+      }
+      yield rawChunk;
+    }
+    output = output != null ? output : toolCallChunks.length > 0 ? getToolCallOutput(toolCallChunks) : textChunks.join("");
+    generation.update({
+      output,
+      completionStartTime,
+      usageDetails: usageDetails != null ? usageDetails : usage ? parseUsageDetails(usage) : void 0
+    }).end();
+  }
+  return tracedOutputGenerator();
+}
+
+// src/observeOpenAI.ts
+var observeOpenAI = (sdk, langfuseConfig) => {
+  return new Proxy(sdk, {
+    get(wrappedSdk, propKey, proxy) {
+      var _a, _b;
+      const originalProperty = wrappedSdk[propKey];
+      const defaultGenerationName = `${(_a = sdk.constructor) == null ? void 0 : _a.name}.${propKey.toString()}`;
+      const generationName = (_b = langfuseConfig == null ? void 0 : langfuseConfig.generationName) != null ? _b : defaultGenerationName;
+      const config = { ...langfuseConfig, generationName };
+      if (typeof originalProperty === "function") {
+        return withTracing(originalProperty.bind(wrappedSdk), config);
+      }
+      const isNestedOpenAIObject = originalProperty && !Array.isArray(originalProperty) && !(originalProperty instanceof Date) && typeof originalProperty === "object";
+      if (isNestedOpenAIObject) {
+        return observeOpenAI(originalProperty, config);
+      }
+      return Reflect.get(wrappedSdk, propKey, proxy);
+    }
+  });
+};
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  observeOpenAI
+});
+//# sourceMappingURL=index.cjs.map

package/dist/index.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/index.ts","../src/traceMethod.ts","../src/parseOpenAI.ts","../src/utils.ts","../src/observeOpenAI.ts"],"sourcesContent":["export { observeOpenAI } from \"./observeOpenAI.js\";\nexport * from \"./types.js\";\n","import { LangfuseGeneration, startObservation } from \"@elasticdash/tracing\";\nimport type OpenAI from \"openai\";\n\nimport {\n  getToolCallOutput,\n  parseChunk,\n  parseCompletionOutput,\n  parseInputArgs,\n  parseUsageDetails,\n  parseModelDataFromResponse,\n  parseUsageDetailsFromResponse,\n} from \"./parseOpenAI.js\";\nimport type { LangfuseConfig } from \"./types.js\";\nimport { isAsyncIterable } from \"./utils.js\";\n\n/**\n * Generic method type for any function that can be traced.\n * @internal\n */\ntype GenericMethod = (...args: unknown[]) => unknown;\n\n/**\n * Wraps a method with Langfuse tracing functionality.\n *\n * This function creates a wrapper around OpenAI SDK methods that automatically\n * creates Langfuse generations, captures input/output data, handles streaming\n * responses, and records usage metrics and errors.\n *\n * @param tracedMethod - The OpenAI SDK method to wrap with tracing\n * @param config - Configuration for the trace and generation\n * @returns A wrapped version of the method that creates Langfuse traces\n *\n * @internal\n */\nexport const withTracing = <T extends GenericMethod>(\n  tracedMethod: T,\n  config?: LangfuseConfig & Required<{ generationName: string }>,\n): ((...args: Parameters<T>) => Promise<ReturnType<T>>) => {\n  return (...args) => wrapMethod(tracedMethod, config, ...args);\n};\n\n/**\n * Internal method that handles the actual tracing logic for OpenAI SDK methods.\n *\n * This function creates a Langfuse generation, executes the original method,\n * and captures all relevant data including input, output, usage, and errors.\n * It handles both streaming and non-streaming responses appropriately.\n *\n * @param tracedMethod - The original OpenAI SDK method to execute\n * @param config - Langfuse configuration options\n * @param args - Arguments to pass to the original method\n * @returns The result from the original method, potentially wrapped for streaming\n *\n * @internal\n */\nconst wrapMethod = <T extends GenericMethod>(\n  tracedMethod: T,\n  config?: LangfuseConfig,\n  ...args: Parameters<T>\n): ReturnType<T> | any => {\n  const { model, input, modelParameters } = parseInputArgs(args[0] ?? {});\n\n  const finalModelParams = { ...modelParameters, response_format: \"\" };\n  const finalMetadata = {\n    ...config?.generationMetadata,\n    response_format:\n      \"response_format\" in modelParameters\n        ? modelParameters.response_format\n        : undefined,\n  };\n\n  const generation = startObservation(\n    config?.generationName ?? \"OpenAI-completion\",\n    {\n      model,\n      input,\n      modelParameters: finalModelParams,\n      prompt: config?.langfusePrompt,\n      metadata: finalMetadata,\n    },\n    {\n      asType: \"generation\",\n      parentSpanContext: config?.parentSpanContext,\n    },\n  ).updateTrace({\n    userId: config?.userId,\n    sessionId: config?.sessionId,\n    tags: config?.tags,\n    name: config?.traceName,\n  });\n\n  try {\n    const res = tracedMethod(...args);\n\n    // Handle stream responses\n    if (isAsyncIterable(res)) {\n      return wrapAsyncIterable(res, generation);\n    }\n\n    if (res instanceof Promise) {\n      const wrappedPromise = res\n        .then((result) => {\n          if (isAsyncIterable(result)) {\n            return wrapAsyncIterable(result, generation);\n          }\n\n          const output = parseCompletionOutput(result);\n          const usageDetails = parseUsageDetailsFromResponse(result);\n          const {\n            model: modelFromResponse,\n            modelParameters: modelParametersFromResponse,\n            metadata: metadataFromResponse,\n          } = parseModelDataFromResponse(result);\n\n          generation\n            .update({\n              output,\n              usageDetails,\n              model: modelFromResponse,\n              modelParameters: modelParametersFromResponse,\n              metadata: metadataFromResponse,\n            })\n            .end();\n\n          return result;\n        })\n        .catch((err) => {\n          generation\n            .update({\n              statusMessage: String(err),\n              level: \"ERROR\",\n              costDetails: {\n                input: 0,\n                output: 0,\n                total: 0,\n              },\n            })\n            .end();\n\n          throw err;\n        });\n\n      return wrappedPromise;\n    }\n\n    return res;\n  } catch (error) {\n    generation\n      .update({\n        statusMessage: String(error),\n        level: \"ERROR\",\n        costDetails: {\n          input: 0,\n          output: 0,\n          total: 0,\n        },\n      })\n      .end();\n\n    throw error;\n  }\n};\n\n/**\n * Wraps an async iterable (streaming response) with Langfuse tracing.\n *\n * This function handles streaming OpenAI responses by collecting chunks,\n * parsing usage information, and updating the Langfuse generation with\n * the complete output and usage details once the stream is consumed.\n *\n * @param iterable - The async iterable from OpenAI (streaming response)\n * @param generation - The Langfuse generation to update with stream data\n * @returns An async generator that yields original chunks while collecting data\n *\n * @internal\n */\nfunction wrapAsyncIterable<R>(\n  iterable: AsyncIterable<unknown>,\n  generation: LangfuseGeneration,\n): R {\n  async function* tracedOutputGenerator(): AsyncGenerator<\n    unknown,\n    void,\n    unknown\n  > {\n    const response = iterable;\n    const textChunks: string[] = [];\n    const toolCallChunks: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall[] =\n      [];\n    let usage: OpenAI.CompletionUsage | null = null;\n    let completionStartTime: Date | undefined = undefined;\n    let usageDetails: Record<string, number> | undefined = undefined;\n    let output: unknown = null;\n\n    for await (const rawChunk of response as AsyncIterable<unknown>) {\n      completionStartTime = completionStartTime ?? new Date();\n\n      // Handle Response API chunks\n      if (typeof rawChunk === \"object\" && rawChunk && \"response\" in rawChunk) {\n        const result = rawChunk[\"response\"];\n        output = parseCompletionOutput(result);\n        usageDetails = parseUsageDetailsFromResponse(result);\n\n        const {\n          model: modelFromResponse,\n          modelParameters: modelParametersFromResponse,\n          metadata: metadataFromResponse,\n        } = parseModelDataFromResponse(result);\n\n        generation.update({\n          model: modelFromResponse,\n          modelParameters: modelParametersFromResponse,\n          metadata: metadataFromResponse,\n        });\n      }\n\n      if (\n        typeof rawChunk === \"object\" &&\n        rawChunk != null &&\n        \"usage\" in rawChunk\n      ) {\n        usage = rawChunk.usage as OpenAI.CompletionUsage | null;\n      }\n\n      const processedChunk = parseChunk(rawChunk);\n\n      if (!processedChunk.isToolCall) {\n        textChunks.push(processedChunk.data);\n      } else {\n        toolCallChunks.push(processedChunk.data);\n      }\n\n      yield rawChunk;\n    }\n\n    output =\n      output ??\n      (toolCallChunks.length > 0\n        ? getToolCallOutput(toolCallChunks)\n        : textChunks.join(\"\"));\n\n    generation\n      .update({\n        output,\n        completionStartTime,\n        usageDetails:\n          usageDetails ?? (usage ? parseUsageDetails(usage) : undefined),\n      })\n      .end();\n  }\n\n  return tracedOutputGenerator() as R;\n}\n","import type OpenAI from \"openai\";\n\ntype ParsedOpenAIArguments = {\n  model: string;\n  input: Record<string, any> | string;\n  modelParameters: Record<string, any>;\n};\n\nexport const parseInputArgs = (\n  args: Record<string, any>,\n): ParsedOpenAIArguments => {\n  let params: Record<string, any> = {};\n  params = {\n    frequency_penalty: args.frequency_penalty,\n    logit_bias: args.logit_bias,\n    logprobs: args.logprobs,\n    max_tokens: args.max_tokens,\n    n: args.n,\n    presence_penalty: args.presence_penalty,\n    seed: args.seed,\n    stop: args.stop,\n    stream: args.stream,\n    temperature: args.temperature,\n    top_p: args.top_p,\n    user: args.user,\n    response_format: args.response_format,\n    top_logprobs: args.top_logprobs,\n  };\n\n  let input: Record<string, any> | string = args.input;\n\n  if (\n    args &&\n    typeof args === \"object\" &&\n    !Array.isArray(args) &&\n    \"messages\" in args\n  ) {\n    input = {};\n    input.messages = args.messages;\n    if (\"function_call\" in args) {\n      input.function_call = args.function_call;\n    }\n    if (\"functions\" in args) {\n      input.functions = args.functions;\n    }\n    if (\"tools\" in args) {\n      input.tools = args.tools;\n    }\n\n    if (\"tool_choice\" in args) {\n      input.tool_choice = args.tool_choice;\n    }\n  } else if (!input) {\n    input = args.prompt;\n  }\n\n  return {\n    model: args.model,\n    input: input,\n    modelParameters: params,\n  };\n};\n\nexport const parseCompletionOutput = (res: unknown): unknown => {\n  if (\n    res instanceof Object &&\n    \"output_text\" in res &&\n    res[\"output_text\"] !== \"\"\n  ) {\n    return res[\"output_text\"] as string;\n  }\n\n  if (\n    typeof res === \"object\" &&\n    res &&\n    \"output\" in res &&\n    Array.isArray(res[\"output\"])\n  ) {\n    const output = res[\"output\"];\n\n    if (output.length > 1) {\n      return output;\n    }\n    if (output.length === 1) {\n      return output[0] as Record<string, unknown>;\n    }\n\n    return null;\n  }\n\n  if (\n    !(res instanceof Object && \"choices\" in res && Array.isArray(res.choices))\n  ) {\n    return \"\";\n  }\n\n  return \"message\" in res.choices[0]\n    ? res.choices[0].message\n    : (res.choices[0].text ?? \"\");\n};\n\nexport const parseUsageDetails = (\n  completionUsage: OpenAI.CompletionUsage,\n): Record<string, number> | undefined => {\n  if (\"prompt_tokens\" in completionUsage) {\n    const {\n      prompt_tokens,\n      completion_tokens,\n      total_tokens,\n      completion_tokens_details,\n      prompt_tokens_details,\n    } = completionUsage;\n\n    const flatPromptTokensDetails = Object.fromEntries(\n      Object.entries(prompt_tokens_details ?? {}).map(([key, value]) => [\n        `input_${key}`,\n        value as number,\n      ]),\n    );\n\n    const flatCompletionTokensDetails = Object.fromEntries(\n      Object.entries(completion_tokens_details ?? {}).map(([key, value]) => [\n        `output_${key}`,\n        value as number,\n      ]),\n    );\n\n    let finalInputTokens = prompt_tokens as number;\n    Object.values(flatPromptTokensDetails).forEach((value) => {\n      finalInputTokens = Math.max(finalInputTokens - value, 0);\n    });\n\n    let finalOutputTokens = completion_tokens as number;\n    Object.values(flatCompletionTokensDetails).forEach((value) => {\n      finalOutputTokens = Math.max(finalOutputTokens - value, 0);\n    });\n\n    return {\n      input: finalInputTokens,\n      output: finalOutputTokens,\n      total: total_tokens,\n      ...flatPromptTokensDetails,\n      ...flatCompletionTokensDetails,\n    };\n  } else if (\"input_tokens\" in completionUsage) {\n    const {\n      input_tokens,\n      output_tokens,\n      total_tokens,\n      input_tokens_details,\n      output_tokens_details,\n    } = completionUsage;\n\n    let finalInputTokens = input_tokens as number;\n    Object.keys(input_tokens_details ?? {}).forEach((key) => {\n      finalInputTokens = Math.max(\n        finalInputTokens -\n          input_tokens_details[key as keyof typeof input_tokens_details],\n        0,\n      );\n    });\n\n    let finalOutputTokens = output_tokens as number;\n    Object.keys(output_tokens_details ?? {}).forEach((key) => {\n      finalOutputTokens = Math.max(\n        finalOutputTokens -\n          output_tokens_details[key as keyof typeof output_tokens_details],\n        0,\n      );\n    });\n\n    return {\n      input: finalInputTokens,\n      output: finalOutputTokens,\n      total: total_tokens,\n      ...Object.fromEntries(\n        Object.entries(input_tokens_details ?? {}).map(([key, value]) => [\n          `input_${key}`,\n          value as number,\n        ]),\n      ),\n      ...Object.fromEntries(\n        Object.entries(output_tokens_details ?? {}).map(([key, value]) => [\n          `output_${key}`,\n          value as number,\n        ]),\n      ),\n    };\n  }\n};\n\nexport const parseUsageDetailsFromResponse = (\n  res: unknown,\n): Record<string, number> | undefined => {\n  if (hasCompletionUsage(res)) {\n    return parseUsageDetails(res.usage);\n  }\n};\n\nexport const parseChunk = (\n  rawChunk: unknown,\n):\n  | { isToolCall: false; data: string }\n  | {\n      isToolCall: true;\n      data: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall;\n    } => {\n  let isToolCall = false;\n  const _chunk = rawChunk as\n    | OpenAI.ChatCompletionChunk\n    | OpenAI.Completions.Completion;\n  const chunkData = _chunk?.choices?.[0];\n\n  try {\n    if (\n      \"delta\" in chunkData &&\n      \"tool_calls\" in chunkData.delta &&\n      Array.isArray(chunkData.delta.tool_calls)\n    ) {\n      isToolCall = true;\n\n      return { isToolCall, data: chunkData.delta.tool_calls[0] };\n    }\n    if (\"delta\" in chunkData) {\n      return { isToolCall, data: chunkData.delta?.content || \"\" };\n    }\n\n    if (\"text\" in chunkData) {\n      return { isToolCall, data: chunkData.text || \"\" };\n    }\n  } catch {}\n\n  return { isToolCall: false, data: \"\" };\n};\n\n// Type guard to check if an unknown object is a UsageResponse\nfunction hasCompletionUsage(\n  obj: any,\n): obj is { usage: OpenAI.CompletionUsage } {\n  return (\n    obj instanceof Object &&\n    \"usage\" in obj &&\n    obj.usage instanceof Object &&\n    // Completion API Usage format\n    ((typeof obj.usage.prompt_tokens === \"number\" &&\n      typeof obj.usage.completion_tokens === \"number\" &&\n      typeof obj.usage.total_tokens === \"number\") ||\n      // Response API Usage format\n      (typeof obj.usage.input_tokens === \"number\" &&\n        typeof obj.usage.output_tokens === \"number\" &&\n        typeof obj.usage.total_tokens === \"number\"))\n  );\n}\n\nexport const getToolCallOutput = (\n  toolCallChunks: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall[],\n): {\n  tool_calls: {\n    function: {\n      name: string;\n      arguments: string;\n    };\n  }[];\n} => {\n  let name = \"\";\n  let toolArguments = \"\";\n\n  for (const toolCall of toolCallChunks) {\n    name = toolCall.function?.name || name;\n    toolArguments += toolCall.function?.arguments || \"\";\n  }\n\n  return {\n    tool_calls: [\n      {\n        function: {\n          name,\n          arguments: toolArguments,\n        },\n      },\n    ],\n  };\n};\n\nexport const parseModelDataFromResponse = (\n  res: unknown,\n): {\n  model: string | undefined;\n  modelParameters: Record<string, string | number> | undefined;\n  metadata: Record<string, unknown> | undefined;\n} => {\n  if (typeof res !== \"object\" || res === null) {\n    return {\n      model: undefined,\n      modelParameters: undefined,\n      metadata: undefined,\n    };\n  }\n\n  const model = \"model\" in res ? (res[\"model\"] as string) : undefined;\n  const modelParameters: Record<string, string | number> = {};\n  const modelParamKeys = [\n    \"max_output_tokens\",\n    \"parallel_tool_calls\",\n    \"store\",\n    \"temperature\",\n    \"tool_choice\",\n    \"top_p\",\n    \"truncation\",\n    \"user\",\n  ];\n\n  const metadata: Record<string, unknown> = {};\n  const metadataKeys = [\n    \"reasoning\",\n    \"incomplete_details\",\n    \"instructions\",\n    \"previous_response_id\",\n    \"tools\",\n    \"metadata\",\n    \"status\",\n    \"error\",\n  ];\n\n  for (const key of modelParamKeys) {\n    const val =\n      key in res ? (res[key as keyof typeof res] as string | number) : null;\n    if (val !== null && val !== undefined) {\n      modelParameters[key as keyof typeof modelParameters] = val;\n    }\n  }\n\n  for (const key of metadataKeys) {\n    const val =\n      key in res ? (res[key as keyof typeof res] as string | number) : null;\n    if (val) {\n      metadata[key as keyof typeof metadata] = val;\n    }\n  }\n\n  return {\n    model,\n    modelParameters:\n      Object.keys(modelParameters).length > 0 ? modelParameters : undefined,\n    metadata: Object.keys(metadata).length > 0 ? metadata : undefined,\n  };\n};\n","/**\n * Type guard to check if a value is an async iterable.\n *\n * This utility function determines whether a given value implements the\n * AsyncIterable interface, which is used to identify streaming responses\n * from the OpenAI SDK.\n *\n * @param x - The value to check\n * @returns True if the value is an async iterable, false otherwise\n *\n * @example\n * ```typescript\n * import { isAsyncIterable } from './utils.js';\n *\n * const response = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [...],\n *   stream: true\n * });\n *\n * if (isAsyncIterable(response)) {\n *   // Handle streaming response\n *   for await (const chunk of response) {\n *     console.log(chunk);\n *   }\n * } else {\n *   // Handle regular response\n *   console.log(response);\n * }\n * ```\n *\n * @public\n */\nexport const isAsyncIterable = (x: unknown): x is AsyncIterable<unknown> =>\n  x != null &&\n  typeof x === \"object\" &&\n  typeof (x as any)[Symbol.asyncIterator] === \"function\";\n","import { withTracing } from \"./traceMethod.js\";\nimport type { LangfuseConfig } from \"./types.js\";\n\n/**\n * Wraps an OpenAI SDK client with automatic Langfuse tracing.\n *\n * This function creates a proxy around the OpenAI SDK that automatically\n * traces all method calls, capturing detailed information about requests,\n * responses, token usage, costs, and performance metrics. It works with\n * both streaming and non-streaming OpenAI API calls.\n *\n * The wrapper recursively traces nested objects in the OpenAI SDK, ensuring\n * that all API calls (chat completions, embeddings, fine-tuning, etc.) are\n * automatically captured as Langfuse generations.\n *\n * @param sdk - The OpenAI SDK client instance to wrap with tracing\n * @param langfuseConfig - Optional configuration for tracing behavior\n * @returns A proxied version of the OpenAI SDK with automatic tracing\n *\n * @example\n * ```typescript\n * import OpenAI from 'openai';\n * import { observeOpenAI } from '@elasticdash/openai';\n *\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY,\n * }));\n *\n * // All OpenAI calls are now automatically traced\n * const response = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Hello!' }],\n *   max_tokens: 100,\n *   temperature: 0.7\n * });\n * ```\n *\n * @example\n * ```typescript\n * // With custom tracing configuration\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY\n * }), {\n *   traceName: 'AI-Assistant-Chat',\n *   userId: 'user-123',\n *   sessionId: 'session-456',\n *   tags: ['production', 'chat-feature'],\n *   generationName: 'gpt-4-chat-completion'\n * });\n *\n * const completion = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Explain quantum computing' }]\n * });\n * ```\n *\n * @example\n * ```typescript\n * // Streaming responses are also automatically traced\n * const stream = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Write a story' }],\n *   stream: true\n * });\n *\n * for await (const chunk of stream) {\n *   process.stdout.write(chunk.choices[0]?.delta?.content || '');\n * }\n * // Final usage details and complete output are captured automatically\n * ```\n *\n * @example\n * ```typescript\n * // Using with Langfuse prompt management\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY\n * }), {\n *   langfusePrompt: {\n *     name: 'chat-assistant-v2',\n *     version: 3,\n *     isFallback: false\n *   },\n *   generationMetadata: {\n *     environment: 'production',\n *     feature: 'chat-assistant'\n *   }\n * });\n * ```\n *\n * @public\n */\nexport const observeOpenAI = <SDKType extends object>(\n  sdk: SDKType,\n  langfuseConfig?: LangfuseConfig,\n): SDKType => {\n  return new Proxy(sdk, {\n    get(wrappedSdk, propKey, proxy) {\n      const originalProperty = wrappedSdk[propKey as keyof SDKType];\n\n      const defaultGenerationName = `${sdk.constructor?.name}.${propKey.toString()}`;\n      const generationName =\n        langfuseConfig?.generationName ?? defaultGenerationName;\n      const config = { ...langfuseConfig, generationName };\n\n      // Trace methods of the OpenAI SDK\n      if (typeof originalProperty === \"function\") {\n        return withTracing(originalProperty.bind(wrappedSdk), config);\n      }\n\n      const isNestedOpenAIObject =\n        originalProperty &&\n        !Array.isArray(originalProperty) &&\n        !(originalProperty instanceof Date) &&\n        typeof originalProperty === \"object\";\n\n      // Recursively wrap nested objects to ensure all nested properties or methods are also traced\n      if (isNestedOpenAIObject) {\n        return observeOpenAI(originalProperty, config);\n      }\n\n      // Fallback to returning the original value\n      return Reflect.get(wrappedSdk, propKey, proxy);\n    },\n  });\n};\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA,qBAAqD;;;ACQ9C,IAAM,iBAAiB,CAC5B,SAC0B;AAC1B,MAAI,SAA8B,CAAC;AACnC,WAAS;AAAA,IACP,mBAAmB,KAAK;AAAA,IACxB,YAAY,KAAK;AAAA,IACjB,UAAU,KAAK;AAAA,IACf,YAAY,KAAK;AAAA,IACjB,GAAG,KAAK;AAAA,IACR,kBAAkB,KAAK;AAAA,IACvB,MAAM,KAAK;AAAA,IACX,MAAM,KAAK;AAAA,IACX,QAAQ,KAAK;AAAA,IACb,aAAa,KAAK;AAAA,IAClB,OAAO,KAAK;AAAA,IACZ,MAAM,KAAK;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,cAAc,KAAK;AAAA,EACrB;AAEA,MAAI,QAAsC,KAAK;AAE/C,MACE,QACA,OAAO,SAAS,YAChB,CAAC,MAAM,QAAQ,IAAI,KACnB,cAAc,MACd;AACA,YAAQ,CAAC;AACT,UAAM,WAAW,KAAK;AACtB,QAAI,mBAAmB,MAAM;AAC3B,YAAM,gBAAgB,KAAK;AAAA,IAC7B;AACA,QAAI,eAAe,MAAM;AACvB,YAAM,YAAY,KAAK;AAAA,IACzB;AACA,QAAI,WAAW,MAAM;AACnB,YAAM,QAAQ,KAAK;AAAA,IACrB;AAEA,QAAI,iBAAiB,MAAM;AACzB,YAAM,cAAc,KAAK;AAAA,IAC3B;AAAA,EACF,WAAW,CAAC,OAAO;AACjB,YAAQ,KAAK;AAAA,EACf;AAEA,SAAO;AAAA,IACL,OAAO,KAAK;AAAA,IACZ;AAAA,IACA,iBAAiB;AAAA,EACnB;AACF;AAEO,IAAM,wBAAwB,CAAC,QAA0B;AA/DhE;AAgEE,MACE,eAAe,UACf,iBAAiB,OACjB,IAAI,aAAa,MAAM,IACvB;AACA,WAAO,IAAI,aAAa;AAAA,EAC1B;AAEA,MACE,OAAO,QAAQ,YACf,OACA,YAAY,OACZ,MAAM,QAAQ,IAAI,QAAQ,CAAC,GAC3B;AACA,UAAM,SAAS,IAAI,QAAQ;AAE3B,QAAI,OAAO,SAAS,GAAG;AACrB,aAAO;AAAA,IACT;AACA,QAAI,OAAO,WAAW,GAAG;AACvB,aAAO,OAAO,CAAC;AAAA,IACjB;AAEA,WAAO;AAAA,EACT;AAEA,MACE,EAAE,eAAe,UAAU,aAAa,OAAO,MAAM,QAAQ,IAAI,OAAO,IACxE;AACA,WAAO;AAAA,EACT;AAEA,SAAO,aAAa,IAAI,QAAQ,CAAC,IAC7B,IAAI,QAAQ,CAAC,EAAE,WACd,SAAI,QAAQ,CAAC,EAAE,SAAf,YAAuB;AAC9B;AAEO,IAAM,oBAAoB,CAC/B,oBACuC;AACvC,MAAI,mBAAmB,iBAAiB;AACtC,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI;AAEJ,UAAM,0BAA0B,OAAO;AAAA,MACrC,OAAO,QAAQ,wDAAyB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,QAChE,SAAS,GAAG;AAAA,QACZ;AAAA,MACF,CAAC;AAAA,IACH;AAEA,UAAM,8BAA8B,OAAO;AAAA,MACzC,OAAO,QAAQ,gEAA6B,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,QACpE,UAAU,GAAG;AAAA,QACb;AAAA,MACF,CAAC;AAAA,IACH;AAEA,QAAI,mBAAmB;AACvB,WAAO,OAAO,uBAAuB,EAAE,QAAQ,CAAC,UAAU;AACxD,yBAAmB,KAAK,IAAI,mBAAmB,OAAO,CAAC;AAAA,IACzD,CAAC;AAED,QAAI,oBAAoB;AACxB,WAAO,OAAO,2BAA2B,EAAE,QAAQ,CAAC,UAAU;AAC5D,0BAAoB,KAAK,IAAI,oBAAoB,OAAO,CAAC;AAAA,IAC3D,CAAC;AAED,WAAO;AAAA,MACL,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,OAAO;AAAA,MACP,GAAG;AAAA,MACH,GAAG;AAAA,IACL;AAAA,EACF,WAAW,kBAAkB,iBAAiB;AAC5C,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI;AAEJ,QAAI,mBAAmB;AACvB,WAAO,KAAK,sDAAwB,CAAC,CAAC,EAAE,QAAQ,CAAC,QAAQ;AACvD,yBAAmB,KAAK;AAAA,QACtB,mBACE,qBAAqB,GAAwC;AAAA,QAC/D;AAAA,MACF;AAAA,IACF,CAAC;AAED,QAAI,oBAAoB;AACxB,WAAO,KAAK,wDAAyB,CAAC,CAAC,EAAE,QAAQ,CAAC,QAAQ;AACxD,0BAAoB,KAAK;AAAA,QACvB,oBACE,sBAAsB,GAAyC;AAAA,QACjE;AAAA,MACF;AAAA,IACF,CAAC;AAED,WAAO;AAAA,MACL,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,OAAO;AAAA,MACP,GAAG,OAAO;AAAA,QACR,OAAO,QAAQ,sDAAwB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,UAC/D,SAAS,GAAG;AAAA,UACZ;AAAA,QACF,CAAC;AAAA,MACH;AAAA,MACA,GAAG,OAAO;AAAA,QACR,OAAO,QAAQ,wDAAyB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,UAChE,UAAU,GAAG;AAAA,UACb;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AACF;AAEO,IAAM,gCAAgC,CAC3C,QACuC;AACvC,MAAI,mBAAmB,GAAG,GAAG;AAC3B,WAAO,kBAAkB,IAAI,KAAK;AAAA,EACpC;AACF;AAEO,IAAM,aAAa,CACxB,aAMO;AA9MT;AA+ME,MAAI,aAAa;AACjB,QAAM,SAAS;AAGf,QAAM,aAAY,sCAAQ,YAAR,mBAAkB;AAEpC,MAAI;AACF,QACE,WAAW,aACX,gBAAgB,UAAU,SAC1B,MAAM,QAAQ,UAAU,MAAM,UAAU,GACxC;AACA,mBAAa;AAEb,aAAO,EAAE,YAAY,MAAM,UAAU,MAAM,WAAW,CAAC,EAAE;AAAA,IAC3D;AACA,QAAI,WAAW,WAAW;AACxB,aAAO,EAAE,YAAY,QAAM,eAAU,UAAV,mBAAiB,YAAW,GAAG;AAAA,IAC5D;AAEA,QAAI,UAAU,WAAW;AACvB,aAAO,EAAE,YAAY,MAAM,UAAU,QAAQ,GAAG;AAAA,IAClD;AAAA,EACF,QAAQ;AAAA,EAAC;AAET,SAAO,EAAE,YAAY,OAAO,MAAM,GAAG;AACvC;AAGA,SAAS,mBACP,KAC0C;AAC1C,SACE,eAAe,UACf,WAAW,OACX,IAAI,iBAAiB;AAAA,GAEnB,OAAO,IAAI,MAAM,kBAAkB,YACnC,OAAO,IAAI,MAAM,sBAAsB,YACvC,OAAO,IAAI,MAAM,iBAAiB;AAAA,EAEjC,OAAO,IAAI,MAAM,iBAAiB,YACjC,OAAO,IAAI,MAAM,kBAAkB,YACnC,OAAO,IAAI,MAAM,iBAAiB;AAE1C;AAEO,IAAM,oBAAoB,CAC/B,mBAQG;AAvQL;AAwQE,MAAI,OAAO;AACX,MAAI,gBAAgB;AAEpB,aAAW,YAAY,gBAAgB;AACrC,aAAO,cAAS,aAAT,mBAAmB,SAAQ;AAClC,uBAAiB,cAAS,aAAT,mBAAmB,cAAa;AAAA,EACnD;AAEA,SAAO;AAAA,IACL,YAAY;AAAA,MACV;AAAA,QACE,UAAU;AAAA,UACR;AAAA,UACA,WAAW;AAAA,QACb;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF;AAEO,IAAM,6BAA6B,CACxC,QAKG;AACH,MAAI,OAAO,QAAQ,YAAY,QAAQ,MAAM;AAC3C,WAAO;AAAA,MACL,OAAO;AAAA,MACP,iBAAiB;AAAA,MACjB,UAAU;AAAA,IACZ;AAAA,EACF;AAEA,QAAM,QAAQ,WAAW,MAAO,IAAI,OAAO,IAAe;AAC1D,QAAM,kBAAmD,CAAC;AAC1D,QAAM,iBAAiB;AAAA,IACrB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,QAAM,WAAoC,CAAC;AAC3C,QAAM,eAAe;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,aAAW,OAAO,gBAAgB;AAChC,UAAM,MACJ,OAAO,MAAO,IAAI,GAAuB,IAAwB;AACnE,QAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,sBAAgB,GAAmC,IAAI;AAAA,IACzD;AAAA,EACF;AAEA,aAAW,OAAO,cAAc;AAC9B,UAAM,MACJ,OAAO,MAAO,IAAI,GAAuB,IAAwB;AACnE,QAAI,KAAK;AACP,eAAS,GAA4B,IAAI;AAAA,IAC3C;AAAA,EACF;AAEA,SAAO;AAAA,IACL;AAAA,IACA,iBACE,OAAO,KAAK,eAAe,EAAE,SAAS,IAAI,kBAAkB;AAAA,IAC9D,UAAU,OAAO,KAAK,QAAQ,EAAE,SAAS,IAAI,WAAW;AAAA,EAC1D;AACF;;;ACzTO,IAAM,kBAAkB,CAAC,MAC9B,KAAK,QACL,OAAO,MAAM,YACb,OAAQ,EAAU,OAAO,aAAa,MAAM;;;AFFvC,IAAM,cAAc,CACzB,cACA,WACyD;AACzD,SAAO,IAAI,SAAS,WAAW,cAAc,QAAQ,GAAG,IAAI;AAC9D;AAgBA,IAAM,aAAa,CACjB,cACA,WACG,SACqB;AA3D1B;AA4DE,QAAM,EAAE,OAAO,OAAO,gBAAgB,IAAI,gBAAe,UAAK,CAAC,MAAN,YAAW,CAAC,CAAC;AAEtE,QAAM,mBAAmB,EAAE,GAAG,iBAAiB,iBAAiB,GAAG;AACnE,QAAM,gBAAgB;AAAA,IACpB,GAAG,iCAAQ;AAAA,IACX,iBACE,qBAAqB,kBACjB,gBAAgB,kBAChB;AAAA,EACR;AAEA,QAAM,iBAAa;AAAA,KACjB,sCAAQ,mBAAR,YAA0B;AAAA,IAC1B;AAAA,MACE;AAAA,MACA;AAAA,MACA,iBAAiB;AAAA,MACjB,QAAQ,iCAAQ;AAAA,MAChB,UAAU;AAAA,IACZ;AAAA,IACA;AAAA,MACE,QAAQ;AAAA,MACR,mBAAmB,iCAAQ;AAAA,IAC7B;AAAA,EACF,EAAE,YAAY;AAAA,IACZ,QAAQ,iCAAQ;AAAA,IAChB,WAAW,iCAAQ;AAAA,IACnB,MAAM,iCAAQ;AAAA,IACd,MAAM,iCAAQ;AAAA,EAChB,CAAC;AAED,MAAI;AACF,UAAM,MAAM,aAAa,GAAG,IAAI;AAGhC,QAAI,gBAAgB,GAAG,GAAG;AACxB,aAAO,kBAAkB,KAAK,UAAU;AAAA,IAC1C;AAEA,QAAI,eAAe,SAAS;AAC1B,YAAM,iBAAiB,IACpB,KAAK,CAAC,WAAW;AAChB,YAAI,gBAAgB,MAAM,GAAG;AAC3B,iBAAO,kBAAkB,QAAQ,UAAU;AAAA,QAC7C;AAEA,cAAM,SAAS,sBAAsB,MAAM;AAC3C,cAAM,eAAe,8BAA8B,MAAM;AACzD,cAAM;AAAA,UACJ,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,IAAI,2BAA2B,MAAM;AAErC,mBACG,OAAO;AAAA,UACN;AAAA,UACA;AAAA,UACA,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,CAAC,EACA,IAAI;AAEP,eAAO;AAAA,MACT,CAAC,EACA,MAAM,CAAC,QAAQ;AACd,mBACG,OAAO;AAAA,UACN,eAAe,OAAO,GAAG;AAAA,UACzB,OAAO;AAAA,UACP,aAAa;AAAA,YACX,OAAO;AAAA,YACP,QAAQ;AAAA,YACR,OAAO;AAAA,UACT;AAAA,QACF,CAAC,EACA,IAAI;AAEP,cAAM;AAAA,MACR,CAAC;AAEH,aAAO;AAAA,IACT;AAEA,WAAO;AAAA,EACT,SAAS,OAAO;AACd,eACG,OAAO;AAAA,MACN,eAAe,OAAO,KAAK;AAAA,MAC3B,OAAO;AAAA,MACP,aAAa;AAAA,QACX,OAAO;AAAA,QACP,QAAQ;AAAA,QACR,OAAO;AAAA,MACT;AAAA,IACF,CAAC,EACA,IAAI;AAEP,UAAM;AAAA,EACR;AACF;AAeA,SAAS,kBACP,UACA,YACG;AACH,kBAAgB,wBAId;AACA,UAAM,WAAW;AACjB,UAAM,aAAuB,CAAC;AAC9B,UAAM,iBACJ,CAAC;AACH,QAAI,QAAuC;AAC3C,QAAI,sBAAwC;AAC5C,QAAI,eAAmD;AACvD,QAAI,SAAkB;AAEtB,qBAAiB,YAAY,UAAoC;AAC/D,4BAAsB,oDAAuB,oBAAI,KAAK;AAGtD,UAAI,OAAO,aAAa,YAAY,YAAY,cAAc,UAAU;AACtE,cAAM,SAAS,SAAS,UAAU;AAClC,iBAAS,sBAAsB,MAAM;AACrC,uBAAe,8BAA8B,MAAM;AAEnD,cAAM;AAAA,UACJ,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,IAAI,2BAA2B,MAAM;AAErC,mBAAW,OAAO;AAAA,UAChB,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,CAAC;AAAA,MACH;AAEA,UACE,OAAO,aAAa,YACpB,YAAY,QACZ,WAAW,UACX;AACA,gBAAQ,SAAS;AAAA,MACnB;AAEA,YAAM,iBAAiB,WAAW,QAAQ;AAE1C,UAAI,CAAC,eAAe,YAAY;AAC9B,mBAAW,KAAK,eAAe,IAAI;AAAA,MACrC,OAAO;AACL,uBAAe,KAAK,eAAe,IAAI;AAAA,MACzC;AAEA,YAAM;AAAA,IACR;AAEA,aACE,0BACC,eAAe,SAAS,IACrB,kBAAkB,cAAc,IAChC,WAAW,KAAK,EAAE;AAExB,eACG,OAAO;AAAA,MACN;AAAA,MACA;AAAA,MACA,cACE,sCAAiB,QAAQ,kBAAkB,KAAK,IAAI;AAAA,IACxD,CAAC,EACA,IAAI;AAAA,EACT;AAEA,SAAO,sBAAsB;AAC/B;;;AGjKO,IAAM,gBAAgB,CAC3B,KACA,mBACY;AACZ,SAAO,IAAI,MAAM,KAAK;AAAA,IACpB,IAAI,YAAY,SAAS,OAAO;AAhGpC;AAiGM,YAAM,mBAAmB,WAAW,OAAwB;AAE5D,YAAM,wBAAwB,IAAG,SAAI,gBAAJ,mBAAiB,IAAI,IAAI,QAAQ,SAAS,CAAC;AAC5E,YAAM,kBACJ,sDAAgB,mBAAhB,YAAkC;AACpC,YAAM,SAAS,EAAE,GAAG,gBAAgB,eAAe;AAGnD,UAAI,OAAO,qBAAqB,YAAY;AAC1C,eAAO,YAAY,iBAAiB,KAAK,UAAU,GAAG,MAAM;AAAA,MAC9D;AAEA,YAAM,uBACJ,oBACA,CAAC,MAAM,QAAQ,gBAAgB,KAC/B,EAAE,4BAA4B,SAC9B,OAAO,qBAAqB;AAG9B,UAAI,sBAAsB;AACxB,eAAO,cAAc,kBAAkB,MAAM;AAAA,MAC/C;AAGA,aAAO,QAAQ,IAAI,YAAY,SAAS,KAAK;AAAA,IAC/C;AAAA,EACF,CAAC;AACH;","names":[]}

package/dist/index.d.cts

@@ -0,0 +1,128 @@
+import { SpanContext } from '@opentelemetry/api';
+
+/**
+ * Configuration options for Langfuse OpenAI tracing.
+ *
+ * This interface defines all available options for customizing how OpenAI
+ * SDK calls are traced and stored in Langfuse. It includes both trace-level
+ * metadata and generation-specific configuration.
+ *
+ * @public
+ */
+type LangfuseConfig = {
+    /** OpenTelemetry span context to use as parent for the generated span */
+    parentSpanContext?: SpanContext;
+    /** Name for the trace that will contain this generation */
+    traceName?: string;
+    /** Session identifier to group related interactions */
+    sessionId?: string;
+    /** User identifier for associating the trace with a specific user */
+    userId?: string;
+    /** Tags for categorizing and filtering traces */
+    tags?: string[];
+    /** Custom name for the generation observation (defaults to SDK method name) */
+    generationName?: string;
+    /** Additional metadata to attach to the generation */
+    generationMetadata?: Record<string, unknown>;
+    /** Information about the Langfuse prompt used for this generation */
+    langfusePrompt?: {
+        /** Name of the prompt template in Langfuse */
+        name: string;
+        /** Version number of the prompt template */
+        version: number;
+        /** Whether this is a fallback prompt due to retrieval failure */
+        isFallback: boolean;
+    };
+};
+
+/**
+ * Wraps an OpenAI SDK client with automatic Langfuse tracing.
+ *
+ * This function creates a proxy around the OpenAI SDK that automatically
+ * traces all method calls, capturing detailed information about requests,
+ * responses, token usage, costs, and performance metrics. It works with
+ * both streaming and non-streaming OpenAI API calls.
+ *
+ * The wrapper recursively traces nested objects in the OpenAI SDK, ensuring
+ * that all API calls (chat completions, embeddings, fine-tuning, etc.) are
+ * automatically captured as Langfuse generations.
+ *
+ * @param sdk - The OpenAI SDK client instance to wrap with tracing
+ * @param langfuseConfig - Optional configuration for tracing behavior
+ * @returns A proxied version of the OpenAI SDK with automatic tracing
+ *
+ * @example
+ * ```typescript
+ * import OpenAI from 'openai';
+ * import { observeOpenAI } from '@elasticdash/openai';
+ *
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY,
+ * }));
+ *
+ * // All OpenAI calls are now automatically traced
+ * const response = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Hello!' }],
+ *   max_tokens: 100,
+ *   temperature: 0.7
+ * });
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // With custom tracing configuration
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY
+ * }), {
+ *   traceName: 'AI-Assistant-Chat',
+ *   userId: 'user-123',
+ *   sessionId: 'session-456',
+ *   tags: ['production', 'chat-feature'],
+ *   generationName: 'gpt-4-chat-completion'
+ * });
+ *
+ * const completion = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Explain quantum computing' }]
+ * });
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // Streaming responses are also automatically traced
+ * const stream = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Write a story' }],
+ *   stream: true
+ * });
+ *
+ * for await (const chunk of stream) {
+ *   process.stdout.write(chunk.choices[0]?.delta?.content || '');
+ * }
+ * // Final usage details and complete output are captured automatically
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // Using with Langfuse prompt management
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY
+ * }), {
+ *   langfusePrompt: {
+ *     name: 'chat-assistant-v2',
+ *     version: 3,
+ *     isFallback: false
+ *   },
+ *   generationMetadata: {
+ *     environment: 'production',
+ *     feature: 'chat-assistant'
+ *   }
+ * });
+ * ```
+ *
+ * @public
+ */
+declare const observeOpenAI: <SDKType extends object>(sdk: SDKType, langfuseConfig?: LangfuseConfig) => SDKType;
+
+export { type LangfuseConfig, observeOpenAI };

package/dist/index.d.ts

@@ -0,0 +1,128 @@
+import { SpanContext } from '@opentelemetry/api';
+
+/**
+ * Configuration options for Langfuse OpenAI tracing.
+ *
+ * This interface defines all available options for customizing how OpenAI
+ * SDK calls are traced and stored in Langfuse. It includes both trace-level
+ * metadata and generation-specific configuration.
+ *
+ * @public
+ */
+type LangfuseConfig = {
+    /** OpenTelemetry span context to use as parent for the generated span */
+    parentSpanContext?: SpanContext;
+    /** Name for the trace that will contain this generation */
+    traceName?: string;
+    /** Session identifier to group related interactions */
+    sessionId?: string;
+    /** User identifier for associating the trace with a specific user */
+    userId?: string;
+    /** Tags for categorizing and filtering traces */
+    tags?: string[];
+    /** Custom name for the generation observation (defaults to SDK method name) */
+    generationName?: string;
+    /** Additional metadata to attach to the generation */
+    generationMetadata?: Record<string, unknown>;
+    /** Information about the Langfuse prompt used for this generation */
+    langfusePrompt?: {
+        /** Name of the prompt template in Langfuse */
+        name: string;
+        /** Version number of the prompt template */
+        version: number;
+        /** Whether this is a fallback prompt due to retrieval failure */
+        isFallback: boolean;
+    };
+};
+
+/**
+ * Wraps an OpenAI SDK client with automatic Langfuse tracing.
+ *
+ * This function creates a proxy around the OpenAI SDK that automatically
+ * traces all method calls, capturing detailed information about requests,
+ * responses, token usage, costs, and performance metrics. It works with
+ * both streaming and non-streaming OpenAI API calls.
+ *
+ * The wrapper recursively traces nested objects in the OpenAI SDK, ensuring
+ * that all API calls (chat completions, embeddings, fine-tuning, etc.) are
+ * automatically captured as Langfuse generations.
+ *
+ * @param sdk - The OpenAI SDK client instance to wrap with tracing
+ * @param langfuseConfig - Optional configuration for tracing behavior
+ * @returns A proxied version of the OpenAI SDK with automatic tracing
+ *
+ * @example
+ * ```typescript
+ * import OpenAI from 'openai';
+ * import { observeOpenAI } from '@elasticdash/openai';
+ *
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY,
+ * }));
+ *
+ * // All OpenAI calls are now automatically traced
+ * const response = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Hello!' }],
+ *   max_tokens: 100,
+ *   temperature: 0.7
+ * });
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // With custom tracing configuration
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY
+ * }), {
+ *   traceName: 'AI-Assistant-Chat',
+ *   userId: 'user-123',
+ *   sessionId: 'session-456',
+ *   tags: ['production', 'chat-feature'],
+ *   generationName: 'gpt-4-chat-completion'
+ * });
+ *
+ * const completion = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Explain quantum computing' }]
+ * });
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // Streaming responses are also automatically traced
+ * const stream = await openai.chat.completions.create({
+ *   model: 'gpt-4',
+ *   messages: [{ role: 'user', content: 'Write a story' }],
+ *   stream: true
+ * });
+ *
+ * for await (const chunk of stream) {
+ *   process.stdout.write(chunk.choices[0]?.delta?.content || '');
+ * }
+ * // Final usage details and complete output are captured automatically
+ * ```
+ *
+ * @example
+ * ```typescript
+ * // Using with Langfuse prompt management
+ * const openai = observeOpenAI(new OpenAI({
+ *   apiKey: process.env.OPENAI_API_KEY
+ * }), {
+ *   langfusePrompt: {
+ *     name: 'chat-assistant-v2',
+ *     version: 3,
+ *     isFallback: false
+ *   },
+ *   generationMetadata: {
+ *     environment: 'production',
+ *     feature: 'chat-assistant'
+ *   }
+ * });
+ * ```
+ *
+ * @public
+ */
+declare const observeOpenAI: <SDKType extends object>(sdk: SDKType, langfuseConfig?: LangfuseConfig) => SDKType;
+
+export { type LangfuseConfig, observeOpenAI };

package/dist/index.mjs

@@ -0,0 +1,401 @@
+// src/traceMethod.ts
+import { startObservation } from "@elasticdash/tracing";
+
+// src/parseOpenAI.ts
+var parseInputArgs = (args) => {
+  let params = {};
+  params = {
+    frequency_penalty: args.frequency_penalty,
+    logit_bias: args.logit_bias,
+    logprobs: args.logprobs,
+    max_tokens: args.max_tokens,
+    n: args.n,
+    presence_penalty: args.presence_penalty,
+    seed: args.seed,
+    stop: args.stop,
+    stream: args.stream,
+    temperature: args.temperature,
+    top_p: args.top_p,
+    user: args.user,
+    response_format: args.response_format,
+    top_logprobs: args.top_logprobs
+  };
+  let input = args.input;
+  if (args && typeof args === "object" && !Array.isArray(args) && "messages" in args) {
+    input = {};
+    input.messages = args.messages;
+    if ("function_call" in args) {
+      input.function_call = args.function_call;
+    }
+    if ("functions" in args) {
+      input.functions = args.functions;
+    }
+    if ("tools" in args) {
+      input.tools = args.tools;
+    }
+    if ("tool_choice" in args) {
+      input.tool_choice = args.tool_choice;
+    }
+  } else if (!input) {
+    input = args.prompt;
+  }
+  return {
+    model: args.model,
+    input,
+    modelParameters: params
+  };
+};
+var parseCompletionOutput = (res) => {
+  var _a;
+  if (res instanceof Object && "output_text" in res && res["output_text"] !== "") {
+    return res["output_text"];
+  }
+  if (typeof res === "object" && res && "output" in res && Array.isArray(res["output"])) {
+    const output = res["output"];
+    if (output.length > 1) {
+      return output;
+    }
+    if (output.length === 1) {
+      return output[0];
+    }
+    return null;
+  }
+  if (!(res instanceof Object && "choices" in res && Array.isArray(res.choices))) {
+    return "";
+  }
+  return "message" in res.choices[0] ? res.choices[0].message : (_a = res.choices[0].text) != null ? _a : "";
+};
+var parseUsageDetails = (completionUsage) => {
+  if ("prompt_tokens" in completionUsage) {
+    const {
+      prompt_tokens,
+      completion_tokens,
+      total_tokens,
+      completion_tokens_details,
+      prompt_tokens_details
+    } = completionUsage;
+    const flatPromptTokensDetails = Object.fromEntries(
+      Object.entries(prompt_tokens_details != null ? prompt_tokens_details : {}).map(([key, value]) => [
+        `input_${key}`,
+        value
+      ])
+    );
+    const flatCompletionTokensDetails = Object.fromEntries(
+      Object.entries(completion_tokens_details != null ? completion_tokens_details : {}).map(([key, value]) => [
+        `output_${key}`,
+        value
+      ])
+    );
+    let finalInputTokens = prompt_tokens;
+    Object.values(flatPromptTokensDetails).forEach((value) => {
+      finalInputTokens = Math.max(finalInputTokens - value, 0);
+    });
+    let finalOutputTokens = completion_tokens;
+    Object.values(flatCompletionTokensDetails).forEach((value) => {
+      finalOutputTokens = Math.max(finalOutputTokens - value, 0);
+    });
+    return {
+      input: finalInputTokens,
+      output: finalOutputTokens,
+      total: total_tokens,
+      ...flatPromptTokensDetails,
+      ...flatCompletionTokensDetails
+    };
+  } else if ("input_tokens" in completionUsage) {
+    const {
+      input_tokens,
+      output_tokens,
+      total_tokens,
+      input_tokens_details,
+      output_tokens_details
+    } = completionUsage;
+    let finalInputTokens = input_tokens;
+    Object.keys(input_tokens_details != null ? input_tokens_details : {}).forEach((key) => {
+      finalInputTokens = Math.max(
+        finalInputTokens - input_tokens_details[key],
+        0
+      );
+    });
+    let finalOutputTokens = output_tokens;
+    Object.keys(output_tokens_details != null ? output_tokens_details : {}).forEach((key) => {
+      finalOutputTokens = Math.max(
+        finalOutputTokens - output_tokens_details[key],
+        0
+      );
+    });
+    return {
+      input: finalInputTokens,
+      output: finalOutputTokens,
+      total: total_tokens,
+      ...Object.fromEntries(
+        Object.entries(input_tokens_details != null ? input_tokens_details : {}).map(([key, value]) => [
+          `input_${key}`,
+          value
+        ])
+      ),
+      ...Object.fromEntries(
+        Object.entries(output_tokens_details != null ? output_tokens_details : {}).map(([key, value]) => [
+          `output_${key}`,
+          value
+        ])
+      )
+    };
+  }
+};
+var parseUsageDetailsFromResponse = (res) => {
+  if (hasCompletionUsage(res)) {
+    return parseUsageDetails(res.usage);
+  }
+};
+var parseChunk = (rawChunk) => {
+  var _a, _b;
+  let isToolCall = false;
+  const _chunk = rawChunk;
+  const chunkData = (_a = _chunk == null ? void 0 : _chunk.choices) == null ? void 0 : _a[0];
+  try {
+    if ("delta" in chunkData && "tool_calls" in chunkData.delta && Array.isArray(chunkData.delta.tool_calls)) {
+      isToolCall = true;
+      return { isToolCall, data: chunkData.delta.tool_calls[0] };
+    }
+    if ("delta" in chunkData) {
+      return { isToolCall, data: ((_b = chunkData.delta) == null ? void 0 : _b.content) || "" };
+    }
+    if ("text" in chunkData) {
+      return { isToolCall, data: chunkData.text || "" };
+    }
+  } catch {
+  }
+  return { isToolCall: false, data: "" };
+};
+function hasCompletionUsage(obj) {
+  return obj instanceof Object && "usage" in obj && obj.usage instanceof Object && // Completion API Usage format
+  (typeof obj.usage.prompt_tokens === "number" && typeof obj.usage.completion_tokens === "number" && typeof obj.usage.total_tokens === "number" || // Response API Usage format
+  typeof obj.usage.input_tokens === "number" && typeof obj.usage.output_tokens === "number" && typeof obj.usage.total_tokens === "number");
+}
+var getToolCallOutput = (toolCallChunks) => {
+  var _a, _b;
+  let name = "";
+  let toolArguments = "";
+  for (const toolCall of toolCallChunks) {
+    name = ((_a = toolCall.function) == null ? void 0 : _a.name) || name;
+    toolArguments += ((_b = toolCall.function) == null ? void 0 : _b.arguments) || "";
+  }
+  return {
+    tool_calls: [
+      {
+        function: {
+          name,
+          arguments: toolArguments
+        }
+      }
+    ]
+  };
+};
+var parseModelDataFromResponse = (res) => {
+  if (typeof res !== "object" || res === null) {
+    return {
+      model: void 0,
+      modelParameters: void 0,
+      metadata: void 0
+    };
+  }
+  const model = "model" in res ? res["model"] : void 0;
+  const modelParameters = {};
+  const modelParamKeys = [
+    "max_output_tokens",
+    "parallel_tool_calls",
+    "store",
+    "temperature",
+    "tool_choice",
+    "top_p",
+    "truncation",
+    "user"
+  ];
+  const metadata = {};
+  const metadataKeys = [
+    "reasoning",
+    "incomplete_details",
+    "instructions",
+    "previous_response_id",
+    "tools",
+    "metadata",
+    "status",
+    "error"
+  ];
+  for (const key of modelParamKeys) {
+    const val = key in res ? res[key] : null;
+    if (val !== null && val !== void 0) {
+      modelParameters[key] = val;
+    }
+  }
+  for (const key of metadataKeys) {
+    const val = key in res ? res[key] : null;
+    if (val) {
+      metadata[key] = val;
+    }
+  }
+  return {
+    model,
+    modelParameters: Object.keys(modelParameters).length > 0 ? modelParameters : void 0,
+    metadata: Object.keys(metadata).length > 0 ? metadata : void 0
+  };
+};
+
+// src/utils.ts
+var isAsyncIterable = (x) => x != null && typeof x === "object" && typeof x[Symbol.asyncIterator] === "function";
+
+// src/traceMethod.ts
+var withTracing = (tracedMethod, config) => {
+  return (...args) => wrapMethod(tracedMethod, config, ...args);
+};
+var wrapMethod = (tracedMethod, config, ...args) => {
+  var _a, _b;
+  const { model, input, modelParameters } = parseInputArgs((_a = args[0]) != null ? _a : {});
+  const finalModelParams = { ...modelParameters, response_format: "" };
+  const finalMetadata = {
+    ...config == null ? void 0 : config.generationMetadata,
+    response_format: "response_format" in modelParameters ? modelParameters.response_format : void 0
+  };
+  const generation = startObservation(
+    (_b = config == null ? void 0 : config.generationName) != null ? _b : "OpenAI-completion",
+    {
+      model,
+      input,
+      modelParameters: finalModelParams,
+      prompt: config == null ? void 0 : config.langfusePrompt,
+      metadata: finalMetadata
+    },
+    {
+      asType: "generation",
+      parentSpanContext: config == null ? void 0 : config.parentSpanContext
+    }
+  ).updateTrace({
+    userId: config == null ? void 0 : config.userId,
+    sessionId: config == null ? void 0 : config.sessionId,
+    tags: config == null ? void 0 : config.tags,
+    name: config == null ? void 0 : config.traceName
+  });
+  try {
+    const res = tracedMethod(...args);
+    if (isAsyncIterable(res)) {
+      return wrapAsyncIterable(res, generation);
+    }
+    if (res instanceof Promise) {
+      const wrappedPromise = res.then((result) => {
+        if (isAsyncIterable(result)) {
+          return wrapAsyncIterable(result, generation);
+        }
+        const output = parseCompletionOutput(result);
+        const usageDetails = parseUsageDetailsFromResponse(result);
+        const {
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        } = parseModelDataFromResponse(result);
+        generation.update({
+          output,
+          usageDetails,
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        }).end();
+        return result;
+      }).catch((err) => {
+        generation.update({
+          statusMessage: String(err),
+          level: "ERROR",
+          costDetails: {
+            input: 0,
+            output: 0,
+            total: 0
+          }
+        }).end();
+        throw err;
+      });
+      return wrappedPromise;
+    }
+    return res;
+  } catch (error) {
+    generation.update({
+      statusMessage: String(error),
+      level: "ERROR",
+      costDetails: {
+        input: 0,
+        output: 0,
+        total: 0
+      }
+    }).end();
+    throw error;
+  }
+};
+function wrapAsyncIterable(iterable, generation) {
+  async function* tracedOutputGenerator() {
+    const response = iterable;
+    const textChunks = [];
+    const toolCallChunks = [];
+    let usage = null;
+    let completionStartTime = void 0;
+    let usageDetails = void 0;
+    let output = null;
+    for await (const rawChunk of response) {
+      completionStartTime = completionStartTime != null ? completionStartTime : /* @__PURE__ */ new Date();
+      if (typeof rawChunk === "object" && rawChunk && "response" in rawChunk) {
+        const result = rawChunk["response"];
+        output = parseCompletionOutput(result);
+        usageDetails = parseUsageDetailsFromResponse(result);
+        const {
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        } = parseModelDataFromResponse(result);
+        generation.update({
+          model: modelFromResponse,
+          modelParameters: modelParametersFromResponse,
+          metadata: metadataFromResponse
+        });
+      }
+      if (typeof rawChunk === "object" && rawChunk != null && "usage" in rawChunk) {
+        usage = rawChunk.usage;
+      }
+      const processedChunk = parseChunk(rawChunk);
+      if (!processedChunk.isToolCall) {
+        textChunks.push(processedChunk.data);
+      } else {
+        toolCallChunks.push(processedChunk.data);
+      }
+      yield rawChunk;
+    }
+    output = output != null ? output : toolCallChunks.length > 0 ? getToolCallOutput(toolCallChunks) : textChunks.join("");
+    generation.update({
+      output,
+      completionStartTime,
+      usageDetails: usageDetails != null ? usageDetails : usage ? parseUsageDetails(usage) : void 0
+    }).end();
+  }
+  return tracedOutputGenerator();
+}
+
+// src/observeOpenAI.ts
+var observeOpenAI = (sdk, langfuseConfig) => {
+  return new Proxy(sdk, {
+    get(wrappedSdk, propKey, proxy) {
+      var _a, _b;
+      const originalProperty = wrappedSdk[propKey];
+      const defaultGenerationName = `${(_a = sdk.constructor) == null ? void 0 : _a.name}.${propKey.toString()}`;
+      const generationName = (_b = langfuseConfig == null ? void 0 : langfuseConfig.generationName) != null ? _b : defaultGenerationName;
+      const config = { ...langfuseConfig, generationName };
+      if (typeof originalProperty === "function") {
+        return withTracing(originalProperty.bind(wrappedSdk), config);
+      }
+      const isNestedOpenAIObject = originalProperty && !Array.isArray(originalProperty) && !(originalProperty instanceof Date) && typeof originalProperty === "object";
+      if (isNestedOpenAIObject) {
+        return observeOpenAI(originalProperty, config);
+      }
+      return Reflect.get(wrappedSdk, propKey, proxy);
+    }
+  });
+};
+export {
+  observeOpenAI
+};
+//# sourceMappingURL=index.mjs.map

package/dist/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/traceMethod.ts","../src/parseOpenAI.ts","../src/utils.ts","../src/observeOpenAI.ts"],"sourcesContent":["import { LangfuseGeneration, startObservation } from \"@elasticdash/tracing\";\nimport type OpenAI from \"openai\";\n\nimport {\n  getToolCallOutput,\n  parseChunk,\n  parseCompletionOutput,\n  parseInputArgs,\n  parseUsageDetails,\n  parseModelDataFromResponse,\n  parseUsageDetailsFromResponse,\n} from \"./parseOpenAI.js\";\nimport type { LangfuseConfig } from \"./types.js\";\nimport { isAsyncIterable } from \"./utils.js\";\n\n/**\n * Generic method type for any function that can be traced.\n * @internal\n */\ntype GenericMethod = (...args: unknown[]) => unknown;\n\n/**\n * Wraps a method with Langfuse tracing functionality.\n *\n * This function creates a wrapper around OpenAI SDK methods that automatically\n * creates Langfuse generations, captures input/output data, handles streaming\n * responses, and records usage metrics and errors.\n *\n * @param tracedMethod - The OpenAI SDK method to wrap with tracing\n * @param config - Configuration for the trace and generation\n * @returns A wrapped version of the method that creates Langfuse traces\n *\n * @internal\n */\nexport const withTracing = <T extends GenericMethod>(\n  tracedMethod: T,\n  config?: LangfuseConfig & Required<{ generationName: string }>,\n): ((...args: Parameters<T>) => Promise<ReturnType<T>>) => {\n  return (...args) => wrapMethod(tracedMethod, config, ...args);\n};\n\n/**\n * Internal method that handles the actual tracing logic for OpenAI SDK methods.\n *\n * This function creates a Langfuse generation, executes the original method,\n * and captures all relevant data including input, output, usage, and errors.\n * It handles both streaming and non-streaming responses appropriately.\n *\n * @param tracedMethod - The original OpenAI SDK method to execute\n * @param config - Langfuse configuration options\n * @param args - Arguments to pass to the original method\n * @returns The result from the original method, potentially wrapped for streaming\n *\n * @internal\n */\nconst wrapMethod = <T extends GenericMethod>(\n  tracedMethod: T,\n  config?: LangfuseConfig,\n  ...args: Parameters<T>\n): ReturnType<T> | any => {\n  const { model, input, modelParameters } = parseInputArgs(args[0] ?? {});\n\n  const finalModelParams = { ...modelParameters, response_format: \"\" };\n  const finalMetadata = {\n    ...config?.generationMetadata,\n    response_format:\n      \"response_format\" in modelParameters\n        ? modelParameters.response_format\n        : undefined,\n  };\n\n  const generation = startObservation(\n    config?.generationName ?? \"OpenAI-completion\",\n    {\n      model,\n      input,\n      modelParameters: finalModelParams,\n      prompt: config?.langfusePrompt,\n      metadata: finalMetadata,\n    },\n    {\n      asType: \"generation\",\n      parentSpanContext: config?.parentSpanContext,\n    },\n  ).updateTrace({\n    userId: config?.userId,\n    sessionId: config?.sessionId,\n    tags: config?.tags,\n    name: config?.traceName,\n  });\n\n  try {\n    const res = tracedMethod(...args);\n\n    // Handle stream responses\n    if (isAsyncIterable(res)) {\n      return wrapAsyncIterable(res, generation);\n    }\n\n    if (res instanceof Promise) {\n      const wrappedPromise = res\n        .then((result) => {\n          if (isAsyncIterable(result)) {\n            return wrapAsyncIterable(result, generation);\n          }\n\n          const output = parseCompletionOutput(result);\n          const usageDetails = parseUsageDetailsFromResponse(result);\n          const {\n            model: modelFromResponse,\n            modelParameters: modelParametersFromResponse,\n            metadata: metadataFromResponse,\n          } = parseModelDataFromResponse(result);\n\n          generation\n            .update({\n              output,\n              usageDetails,\n              model: modelFromResponse,\n              modelParameters: modelParametersFromResponse,\n              metadata: metadataFromResponse,\n            })\n            .end();\n\n          return result;\n        })\n        .catch((err) => {\n          generation\n            .update({\n              statusMessage: String(err),\n              level: \"ERROR\",\n              costDetails: {\n                input: 0,\n                output: 0,\n                total: 0,\n              },\n            })\n            .end();\n\n          throw err;\n        });\n\n      return wrappedPromise;\n    }\n\n    return res;\n  } catch (error) {\n    generation\n      .update({\n        statusMessage: String(error),\n        level: \"ERROR\",\n        costDetails: {\n          input: 0,\n          output: 0,\n          total: 0,\n        },\n      })\n      .end();\n\n    throw error;\n  }\n};\n\n/**\n * Wraps an async iterable (streaming response) with Langfuse tracing.\n *\n * This function handles streaming OpenAI responses by collecting chunks,\n * parsing usage information, and updating the Langfuse generation with\n * the complete output and usage details once the stream is consumed.\n *\n * @param iterable - The async iterable from OpenAI (streaming response)\n * @param generation - The Langfuse generation to update with stream data\n * @returns An async generator that yields original chunks while collecting data\n *\n * @internal\n */\nfunction wrapAsyncIterable<R>(\n  iterable: AsyncIterable<unknown>,\n  generation: LangfuseGeneration,\n): R {\n  async function* tracedOutputGenerator(): AsyncGenerator<\n    unknown,\n    void,\n    unknown\n  > {\n    const response = iterable;\n    const textChunks: string[] = [];\n    const toolCallChunks: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall[] =\n      [];\n    let usage: OpenAI.CompletionUsage | null = null;\n    let completionStartTime: Date | undefined = undefined;\n    let usageDetails: Record<string, number> | undefined = undefined;\n    let output: unknown = null;\n\n    for await (const rawChunk of response as AsyncIterable<unknown>) {\n      completionStartTime = completionStartTime ?? new Date();\n\n      // Handle Response API chunks\n      if (typeof rawChunk === \"object\" && rawChunk && \"response\" in rawChunk) {\n        const result = rawChunk[\"response\"];\n        output = parseCompletionOutput(result);\n        usageDetails = parseUsageDetailsFromResponse(result);\n\n        const {\n          model: modelFromResponse,\n          modelParameters: modelParametersFromResponse,\n          metadata: metadataFromResponse,\n        } = parseModelDataFromResponse(result);\n\n        generation.update({\n          model: modelFromResponse,\n          modelParameters: modelParametersFromResponse,\n          metadata: metadataFromResponse,\n        });\n      }\n\n      if (\n        typeof rawChunk === \"object\" &&\n        rawChunk != null &&\n        \"usage\" in rawChunk\n      ) {\n        usage = rawChunk.usage as OpenAI.CompletionUsage | null;\n      }\n\n      const processedChunk = parseChunk(rawChunk);\n\n      if (!processedChunk.isToolCall) {\n        textChunks.push(processedChunk.data);\n      } else {\n        toolCallChunks.push(processedChunk.data);\n      }\n\n      yield rawChunk;\n    }\n\n    output =\n      output ??\n      (toolCallChunks.length > 0\n        ? getToolCallOutput(toolCallChunks)\n        : textChunks.join(\"\"));\n\n    generation\n      .update({\n        output,\n        completionStartTime,\n        usageDetails:\n          usageDetails ?? (usage ? parseUsageDetails(usage) : undefined),\n      })\n      .end();\n  }\n\n  return tracedOutputGenerator() as R;\n}\n","import type OpenAI from \"openai\";\n\ntype ParsedOpenAIArguments = {\n  model: string;\n  input: Record<string, any> | string;\n  modelParameters: Record<string, any>;\n};\n\nexport const parseInputArgs = (\n  args: Record<string, any>,\n): ParsedOpenAIArguments => {\n  let params: Record<string, any> = {};\n  params = {\n    frequency_penalty: args.frequency_penalty,\n    logit_bias: args.logit_bias,\n    logprobs: args.logprobs,\n    max_tokens: args.max_tokens,\n    n: args.n,\n    presence_penalty: args.presence_penalty,\n    seed: args.seed,\n    stop: args.stop,\n    stream: args.stream,\n    temperature: args.temperature,\n    top_p: args.top_p,\n    user: args.user,\n    response_format: args.response_format,\n    top_logprobs: args.top_logprobs,\n  };\n\n  let input: Record<string, any> | string = args.input;\n\n  if (\n    args &&\n    typeof args === \"object\" &&\n    !Array.isArray(args) &&\n    \"messages\" in args\n  ) {\n    input = {};\n    input.messages = args.messages;\n    if (\"function_call\" in args) {\n      input.function_call = args.function_call;\n    }\n    if (\"functions\" in args) {\n      input.functions = args.functions;\n    }\n    if (\"tools\" in args) {\n      input.tools = args.tools;\n    }\n\n    if (\"tool_choice\" in args) {\n      input.tool_choice = args.tool_choice;\n    }\n  } else if (!input) {\n    input = args.prompt;\n  }\n\n  return {\n    model: args.model,\n    input: input,\n    modelParameters: params,\n  };\n};\n\nexport const parseCompletionOutput = (res: unknown): unknown => {\n  if (\n    res instanceof Object &&\n    \"output_text\" in res &&\n    res[\"output_text\"] !== \"\"\n  ) {\n    return res[\"output_text\"] as string;\n  }\n\n  if (\n    typeof res === \"object\" &&\n    res &&\n    \"output\" in res &&\n    Array.isArray(res[\"output\"])\n  ) {\n    const output = res[\"output\"];\n\n    if (output.length > 1) {\n      return output;\n    }\n    if (output.length === 1) {\n      return output[0] as Record<string, unknown>;\n    }\n\n    return null;\n  }\n\n  if (\n    !(res instanceof Object && \"choices\" in res && Array.isArray(res.choices))\n  ) {\n    return \"\";\n  }\n\n  return \"message\" in res.choices[0]\n    ? res.choices[0].message\n    : (res.choices[0].text ?? \"\");\n};\n\nexport const parseUsageDetails = (\n  completionUsage: OpenAI.CompletionUsage,\n): Record<string, number> | undefined => {\n  if (\"prompt_tokens\" in completionUsage) {\n    const {\n      prompt_tokens,\n      completion_tokens,\n      total_tokens,\n      completion_tokens_details,\n      prompt_tokens_details,\n    } = completionUsage;\n\n    const flatPromptTokensDetails = Object.fromEntries(\n      Object.entries(prompt_tokens_details ?? {}).map(([key, value]) => [\n        `input_${key}`,\n        value as number,\n      ]),\n    );\n\n    const flatCompletionTokensDetails = Object.fromEntries(\n      Object.entries(completion_tokens_details ?? {}).map(([key, value]) => [\n        `output_${key}`,\n        value as number,\n      ]),\n    );\n\n    let finalInputTokens = prompt_tokens as number;\n    Object.values(flatPromptTokensDetails).forEach((value) => {\n      finalInputTokens = Math.max(finalInputTokens - value, 0);\n    });\n\n    let finalOutputTokens = completion_tokens as number;\n    Object.values(flatCompletionTokensDetails).forEach((value) => {\n      finalOutputTokens = Math.max(finalOutputTokens - value, 0);\n    });\n\n    return {\n      input: finalInputTokens,\n      output: finalOutputTokens,\n      total: total_tokens,\n      ...flatPromptTokensDetails,\n      ...flatCompletionTokensDetails,\n    };\n  } else if (\"input_tokens\" in completionUsage) {\n    const {\n      input_tokens,\n      output_tokens,\n      total_tokens,\n      input_tokens_details,\n      output_tokens_details,\n    } = completionUsage;\n\n    let finalInputTokens = input_tokens as number;\n    Object.keys(input_tokens_details ?? {}).forEach((key) => {\n      finalInputTokens = Math.max(\n        finalInputTokens -\n          input_tokens_details[key as keyof typeof input_tokens_details],\n        0,\n      );\n    });\n\n    let finalOutputTokens = output_tokens as number;\n    Object.keys(output_tokens_details ?? {}).forEach((key) => {\n      finalOutputTokens = Math.max(\n        finalOutputTokens -\n          output_tokens_details[key as keyof typeof output_tokens_details],\n        0,\n      );\n    });\n\n    return {\n      input: finalInputTokens,\n      output: finalOutputTokens,\n      total: total_tokens,\n      ...Object.fromEntries(\n        Object.entries(input_tokens_details ?? {}).map(([key, value]) => [\n          `input_${key}`,\n          value as number,\n        ]),\n      ),\n      ...Object.fromEntries(\n        Object.entries(output_tokens_details ?? {}).map(([key, value]) => [\n          `output_${key}`,\n          value as number,\n        ]),\n      ),\n    };\n  }\n};\n\nexport const parseUsageDetailsFromResponse = (\n  res: unknown,\n): Record<string, number> | undefined => {\n  if (hasCompletionUsage(res)) {\n    return parseUsageDetails(res.usage);\n  }\n};\n\nexport const parseChunk = (\n  rawChunk: unknown,\n):\n  | { isToolCall: false; data: string }\n  | {\n      isToolCall: true;\n      data: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall;\n    } => {\n  let isToolCall = false;\n  const _chunk = rawChunk as\n    | OpenAI.ChatCompletionChunk\n    | OpenAI.Completions.Completion;\n  const chunkData = _chunk?.choices?.[0];\n\n  try {\n    if (\n      \"delta\" in chunkData &&\n      \"tool_calls\" in chunkData.delta &&\n      Array.isArray(chunkData.delta.tool_calls)\n    ) {\n      isToolCall = true;\n\n      return { isToolCall, data: chunkData.delta.tool_calls[0] };\n    }\n    if (\"delta\" in chunkData) {\n      return { isToolCall, data: chunkData.delta?.content || \"\" };\n    }\n\n    if (\"text\" in chunkData) {\n      return { isToolCall, data: chunkData.text || \"\" };\n    }\n  } catch {}\n\n  return { isToolCall: false, data: \"\" };\n};\n\n// Type guard to check if an unknown object is a UsageResponse\nfunction hasCompletionUsage(\n  obj: any,\n): obj is { usage: OpenAI.CompletionUsage } {\n  return (\n    obj instanceof Object &&\n    \"usage\" in obj &&\n    obj.usage instanceof Object &&\n    // Completion API Usage format\n    ((typeof obj.usage.prompt_tokens === \"number\" &&\n      typeof obj.usage.completion_tokens === \"number\" &&\n      typeof obj.usage.total_tokens === \"number\") ||\n      // Response API Usage format\n      (typeof obj.usage.input_tokens === \"number\" &&\n        typeof obj.usage.output_tokens === \"number\" &&\n        typeof obj.usage.total_tokens === \"number\"))\n  );\n}\n\nexport const getToolCallOutput = (\n  toolCallChunks: OpenAI.Chat.Completions.ChatCompletionChunk.Choice.Delta.ToolCall[],\n): {\n  tool_calls: {\n    function: {\n      name: string;\n      arguments: string;\n    };\n  }[];\n} => {\n  let name = \"\";\n  let toolArguments = \"\";\n\n  for (const toolCall of toolCallChunks) {\n    name = toolCall.function?.name || name;\n    toolArguments += toolCall.function?.arguments || \"\";\n  }\n\n  return {\n    tool_calls: [\n      {\n        function: {\n          name,\n          arguments: toolArguments,\n        },\n      },\n    ],\n  };\n};\n\nexport const parseModelDataFromResponse = (\n  res: unknown,\n): {\n  model: string | undefined;\n  modelParameters: Record<string, string | number> | undefined;\n  metadata: Record<string, unknown> | undefined;\n} => {\n  if (typeof res !== \"object\" || res === null) {\n    return {\n      model: undefined,\n      modelParameters: undefined,\n      metadata: undefined,\n    };\n  }\n\n  const model = \"model\" in res ? (res[\"model\"] as string) : undefined;\n  const modelParameters: Record<string, string | number> = {};\n  const modelParamKeys = [\n    \"max_output_tokens\",\n    \"parallel_tool_calls\",\n    \"store\",\n    \"temperature\",\n    \"tool_choice\",\n    \"top_p\",\n    \"truncation\",\n    \"user\",\n  ];\n\n  const metadata: Record<string, unknown> = {};\n  const metadataKeys = [\n    \"reasoning\",\n    \"incomplete_details\",\n    \"instructions\",\n    \"previous_response_id\",\n    \"tools\",\n    \"metadata\",\n    \"status\",\n    \"error\",\n  ];\n\n  for (const key of modelParamKeys) {\n    const val =\n      key in res ? (res[key as keyof typeof res] as string | number) : null;\n    if (val !== null && val !== undefined) {\n      modelParameters[key as keyof typeof modelParameters] = val;\n    }\n  }\n\n  for (const key of metadataKeys) {\n    const val =\n      key in res ? (res[key as keyof typeof res] as string | number) : null;\n    if (val) {\n      metadata[key as keyof typeof metadata] = val;\n    }\n  }\n\n  return {\n    model,\n    modelParameters:\n      Object.keys(modelParameters).length > 0 ? modelParameters : undefined,\n    metadata: Object.keys(metadata).length > 0 ? metadata : undefined,\n  };\n};\n","/**\n * Type guard to check if a value is an async iterable.\n *\n * This utility function determines whether a given value implements the\n * AsyncIterable interface, which is used to identify streaming responses\n * from the OpenAI SDK.\n *\n * @param x - The value to check\n * @returns True if the value is an async iterable, false otherwise\n *\n * @example\n * ```typescript\n * import { isAsyncIterable } from './utils.js';\n *\n * const response = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [...],\n *   stream: true\n * });\n *\n * if (isAsyncIterable(response)) {\n *   // Handle streaming response\n *   for await (const chunk of response) {\n *     console.log(chunk);\n *   }\n * } else {\n *   // Handle regular response\n *   console.log(response);\n * }\n * ```\n *\n * @public\n */\nexport const isAsyncIterable = (x: unknown): x is AsyncIterable<unknown> =>\n  x != null &&\n  typeof x === \"object\" &&\n  typeof (x as any)[Symbol.asyncIterator] === \"function\";\n","import { withTracing } from \"./traceMethod.js\";\nimport type { LangfuseConfig } from \"./types.js\";\n\n/**\n * Wraps an OpenAI SDK client with automatic Langfuse tracing.\n *\n * This function creates a proxy around the OpenAI SDK that automatically\n * traces all method calls, capturing detailed information about requests,\n * responses, token usage, costs, and performance metrics. It works with\n * both streaming and non-streaming OpenAI API calls.\n *\n * The wrapper recursively traces nested objects in the OpenAI SDK, ensuring\n * that all API calls (chat completions, embeddings, fine-tuning, etc.) are\n * automatically captured as Langfuse generations.\n *\n * @param sdk - The OpenAI SDK client instance to wrap with tracing\n * @param langfuseConfig - Optional configuration for tracing behavior\n * @returns A proxied version of the OpenAI SDK with automatic tracing\n *\n * @example\n * ```typescript\n * import OpenAI from 'openai';\n * import { observeOpenAI } from '@elasticdash/openai';\n *\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY,\n * }));\n *\n * // All OpenAI calls are now automatically traced\n * const response = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Hello!' }],\n *   max_tokens: 100,\n *   temperature: 0.7\n * });\n * ```\n *\n * @example\n * ```typescript\n * // With custom tracing configuration\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY\n * }), {\n *   traceName: 'AI-Assistant-Chat',\n *   userId: 'user-123',\n *   sessionId: 'session-456',\n *   tags: ['production', 'chat-feature'],\n *   generationName: 'gpt-4-chat-completion'\n * });\n *\n * const completion = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Explain quantum computing' }]\n * });\n * ```\n *\n * @example\n * ```typescript\n * // Streaming responses are also automatically traced\n * const stream = await openai.chat.completions.create({\n *   model: 'gpt-4',\n *   messages: [{ role: 'user', content: 'Write a story' }],\n *   stream: true\n * });\n *\n * for await (const chunk of stream) {\n *   process.stdout.write(chunk.choices[0]?.delta?.content || '');\n * }\n * // Final usage details and complete output are captured automatically\n * ```\n *\n * @example\n * ```typescript\n * // Using with Langfuse prompt management\n * const openai = observeOpenAI(new OpenAI({\n *   apiKey: process.env.OPENAI_API_KEY\n * }), {\n *   langfusePrompt: {\n *     name: 'chat-assistant-v2',\n *     version: 3,\n *     isFallback: false\n *   },\n *   generationMetadata: {\n *     environment: 'production',\n *     feature: 'chat-assistant'\n *   }\n * });\n * ```\n *\n * @public\n */\nexport const observeOpenAI = <SDKType extends object>(\n  sdk: SDKType,\n  langfuseConfig?: LangfuseConfig,\n): SDKType => {\n  return new Proxy(sdk, {\n    get(wrappedSdk, propKey, proxy) {\n      const originalProperty = wrappedSdk[propKey as keyof SDKType];\n\n      const defaultGenerationName = `${sdk.constructor?.name}.${propKey.toString()}`;\n      const generationName =\n        langfuseConfig?.generationName ?? defaultGenerationName;\n      const config = { ...langfuseConfig, generationName };\n\n      // Trace methods of the OpenAI SDK\n      if (typeof originalProperty === \"function\") {\n        return withTracing(originalProperty.bind(wrappedSdk), config);\n      }\n\n      const isNestedOpenAIObject =\n        originalProperty &&\n        !Array.isArray(originalProperty) &&\n        !(originalProperty instanceof Date) &&\n        typeof originalProperty === \"object\";\n\n      // Recursively wrap nested objects to ensure all nested properties or methods are also traced\n      if (isNestedOpenAIObject) {\n        return observeOpenAI(originalProperty, config);\n      }\n\n      // Fallback to returning the original value\n      return Reflect.get(wrappedSdk, propKey, proxy);\n    },\n  });\n};\n"],"mappings":";AAAA,SAA6B,wBAAwB;;;ACQ9C,IAAM,iBAAiB,CAC5B,SAC0B;AAC1B,MAAI,SAA8B,CAAC;AACnC,WAAS;AAAA,IACP,mBAAmB,KAAK;AAAA,IACxB,YAAY,KAAK;AAAA,IACjB,UAAU,KAAK;AAAA,IACf,YAAY,KAAK;AAAA,IACjB,GAAG,KAAK;AAAA,IACR,kBAAkB,KAAK;AAAA,IACvB,MAAM,KAAK;AAAA,IACX,MAAM,KAAK;AAAA,IACX,QAAQ,KAAK;AAAA,IACb,aAAa,KAAK;AAAA,IAClB,OAAO,KAAK;AAAA,IACZ,MAAM,KAAK;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,cAAc,KAAK;AAAA,EACrB;AAEA,MAAI,QAAsC,KAAK;AAE/C,MACE,QACA,OAAO,SAAS,YAChB,CAAC,MAAM,QAAQ,IAAI,KACnB,cAAc,MACd;AACA,YAAQ,CAAC;AACT,UAAM,WAAW,KAAK;AACtB,QAAI,mBAAmB,MAAM;AAC3B,YAAM,gBAAgB,KAAK;AAAA,IAC7B;AACA,QAAI,eAAe,MAAM;AACvB,YAAM,YAAY,KAAK;AAAA,IACzB;AACA,QAAI,WAAW,MAAM;AACnB,YAAM,QAAQ,KAAK;AAAA,IACrB;AAEA,QAAI,iBAAiB,MAAM;AACzB,YAAM,cAAc,KAAK;AAAA,IAC3B;AAAA,EACF,WAAW,CAAC,OAAO;AACjB,YAAQ,KAAK;AAAA,EACf;AAEA,SAAO;AAAA,IACL,OAAO,KAAK;AAAA,IACZ;AAAA,IACA,iBAAiB;AAAA,EACnB;AACF;AAEO,IAAM,wBAAwB,CAAC,QAA0B;AA/DhE;AAgEE,MACE,eAAe,UACf,iBAAiB,OACjB,IAAI,aAAa,MAAM,IACvB;AACA,WAAO,IAAI,aAAa;AAAA,EAC1B;AAEA,MACE,OAAO,QAAQ,YACf,OACA,YAAY,OACZ,MAAM,QAAQ,IAAI,QAAQ,CAAC,GAC3B;AACA,UAAM,SAAS,IAAI,QAAQ;AAE3B,QAAI,OAAO,SAAS,GAAG;AACrB,aAAO;AAAA,IACT;AACA,QAAI,OAAO,WAAW,GAAG;AACvB,aAAO,OAAO,CAAC;AAAA,IACjB;AAEA,WAAO;AAAA,EACT;AAEA,MACE,EAAE,eAAe,UAAU,aAAa,OAAO,MAAM,QAAQ,IAAI,OAAO,IACxE;AACA,WAAO;AAAA,EACT;AAEA,SAAO,aAAa,IAAI,QAAQ,CAAC,IAC7B,IAAI,QAAQ,CAAC,EAAE,WACd,SAAI,QAAQ,CAAC,EAAE,SAAf,YAAuB;AAC9B;AAEO,IAAM,oBAAoB,CAC/B,oBACuC;AACvC,MAAI,mBAAmB,iBAAiB;AACtC,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI;AAEJ,UAAM,0BAA0B,OAAO;AAAA,MACrC,OAAO,QAAQ,wDAAyB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,QAChE,SAAS,GAAG;AAAA,QACZ;AAAA,MACF,CAAC;AAAA,IACH;AAEA,UAAM,8BAA8B,OAAO;AAAA,MACzC,OAAO,QAAQ,gEAA6B,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,QACpE,UAAU,GAAG;AAAA,QACb;AAAA,MACF,CAAC;AAAA,IACH;AAEA,QAAI,mBAAmB;AACvB,WAAO,OAAO,uBAAuB,EAAE,QAAQ,CAAC,UAAU;AACxD,yBAAmB,KAAK,IAAI,mBAAmB,OAAO,CAAC;AAAA,IACzD,CAAC;AAED,QAAI,oBAAoB;AACxB,WAAO,OAAO,2BAA2B,EAAE,QAAQ,CAAC,UAAU;AAC5D,0BAAoB,KAAK,IAAI,oBAAoB,OAAO,CAAC;AAAA,IAC3D,CAAC;AAED,WAAO;AAAA,MACL,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,OAAO;AAAA,MACP,GAAG;AAAA,MACH,GAAG;AAAA,IACL;AAAA,EACF,WAAW,kBAAkB,iBAAiB;AAC5C,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI;AAEJ,QAAI,mBAAmB;AACvB,WAAO,KAAK,sDAAwB,CAAC,CAAC,EAAE,QAAQ,CAAC,QAAQ;AACvD,yBAAmB,KAAK;AAAA,QACtB,mBACE,qBAAqB,GAAwC;AAAA,QAC/D;AAAA,MACF;AAAA,IACF,CAAC;AAED,QAAI,oBAAoB;AACxB,WAAO,KAAK,wDAAyB,CAAC,CAAC,EAAE,QAAQ,CAAC,QAAQ;AACxD,0BAAoB,KAAK;AAAA,QACvB,oBACE,sBAAsB,GAAyC;AAAA,QACjE;AAAA,MACF;AAAA,IACF,CAAC;AAED,WAAO;AAAA,MACL,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,OAAO;AAAA,MACP,GAAG,OAAO;AAAA,QACR,OAAO,QAAQ,sDAAwB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,UAC/D,SAAS,GAAG;AAAA,UACZ;AAAA,QACF,CAAC;AAAA,MACH;AAAA,MACA,GAAG,OAAO;AAAA,QACR,OAAO,QAAQ,wDAAyB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,MAAM;AAAA,UAChE,UAAU,GAAG;AAAA,UACb;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AACF;AAEO,IAAM,gCAAgC,CAC3C,QACuC;AACvC,MAAI,mBAAmB,GAAG,GAAG;AAC3B,WAAO,kBAAkB,IAAI,KAAK;AAAA,EACpC;AACF;AAEO,IAAM,aAAa,CACxB,aAMO;AA9MT;AA+ME,MAAI,aAAa;AACjB,QAAM,SAAS;AAGf,QAAM,aAAY,sCAAQ,YAAR,mBAAkB;AAEpC,MAAI;AACF,QACE,WAAW,aACX,gBAAgB,UAAU,SAC1B,MAAM,QAAQ,UAAU,MAAM,UAAU,GACxC;AACA,mBAAa;AAEb,aAAO,EAAE,YAAY,MAAM,UAAU,MAAM,WAAW,CAAC,EAAE;AAAA,IAC3D;AACA,QAAI,WAAW,WAAW;AACxB,aAAO,EAAE,YAAY,QAAM,eAAU,UAAV,mBAAiB,YAAW,GAAG;AAAA,IAC5D;AAEA,QAAI,UAAU,WAAW;AACvB,aAAO,EAAE,YAAY,MAAM,UAAU,QAAQ,GAAG;AAAA,IAClD;AAAA,EACF,QAAQ;AAAA,EAAC;AAET,SAAO,EAAE,YAAY,OAAO,MAAM,GAAG;AACvC;AAGA,SAAS,mBACP,KAC0C;AAC1C,SACE,eAAe,UACf,WAAW,OACX,IAAI,iBAAiB;AAAA,GAEnB,OAAO,IAAI,MAAM,kBAAkB,YACnC,OAAO,IAAI,MAAM,sBAAsB,YACvC,OAAO,IAAI,MAAM,iBAAiB;AAAA,EAEjC,OAAO,IAAI,MAAM,iBAAiB,YACjC,OAAO,IAAI,MAAM,kBAAkB,YACnC,OAAO,IAAI,MAAM,iBAAiB;AAE1C;AAEO,IAAM,oBAAoB,CAC/B,mBAQG;AAvQL;AAwQE,MAAI,OAAO;AACX,MAAI,gBAAgB;AAEpB,aAAW,YAAY,gBAAgB;AACrC,aAAO,cAAS,aAAT,mBAAmB,SAAQ;AAClC,uBAAiB,cAAS,aAAT,mBAAmB,cAAa;AAAA,EACnD;AAEA,SAAO;AAAA,IACL,YAAY;AAAA,MACV;AAAA,QACE,UAAU;AAAA,UACR;AAAA,UACA,WAAW;AAAA,QACb;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF;AAEO,IAAM,6BAA6B,CACxC,QAKG;AACH,MAAI,OAAO,QAAQ,YAAY,QAAQ,MAAM;AAC3C,WAAO;AAAA,MACL,OAAO;AAAA,MACP,iBAAiB;AAAA,MACjB,UAAU;AAAA,IACZ;AAAA,EACF;AAEA,QAAM,QAAQ,WAAW,MAAO,IAAI,OAAO,IAAe;AAC1D,QAAM,kBAAmD,CAAC;AAC1D,QAAM,iBAAiB;AAAA,IACrB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,QAAM,WAAoC,CAAC;AAC3C,QAAM,eAAe;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,aAAW,OAAO,gBAAgB;AAChC,UAAM,MACJ,OAAO,MAAO,IAAI,GAAuB,IAAwB;AACnE,QAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,sBAAgB,GAAmC,IAAI;AAAA,IACzD;AAAA,EACF;AAEA,aAAW,OAAO,cAAc;AAC9B,UAAM,MACJ,OAAO,MAAO,IAAI,GAAuB,IAAwB;AACnE,QAAI,KAAK;AACP,eAAS,GAA4B,IAAI;AAAA,IAC3C;AAAA,EACF;AAEA,SAAO;AAAA,IACL;AAAA,IACA,iBACE,OAAO,KAAK,eAAe,EAAE,SAAS,IAAI,kBAAkB;AAAA,IAC9D,UAAU,OAAO,KAAK,QAAQ,EAAE,SAAS,IAAI,WAAW;AAAA,EAC1D;AACF;;;ACzTO,IAAM,kBAAkB,CAAC,MAC9B,KAAK,QACL,OAAO,MAAM,YACb,OAAQ,EAAU,OAAO,aAAa,MAAM;;;AFFvC,IAAM,cAAc,CACzB,cACA,WACyD;AACzD,SAAO,IAAI,SAAS,WAAW,cAAc,QAAQ,GAAG,IAAI;AAC9D;AAgBA,IAAM,aAAa,CACjB,cACA,WACG,SACqB;AA3D1B;AA4DE,QAAM,EAAE,OAAO,OAAO,gBAAgB,IAAI,gBAAe,UAAK,CAAC,MAAN,YAAW,CAAC,CAAC;AAEtE,QAAM,mBAAmB,EAAE,GAAG,iBAAiB,iBAAiB,GAAG;AACnE,QAAM,gBAAgB;AAAA,IACpB,GAAG,iCAAQ;AAAA,IACX,iBACE,qBAAqB,kBACjB,gBAAgB,kBAChB;AAAA,EACR;AAEA,QAAM,aAAa;AAAA,KACjB,sCAAQ,mBAAR,YAA0B;AAAA,IAC1B;AAAA,MACE;AAAA,MACA;AAAA,MACA,iBAAiB;AAAA,MACjB,QAAQ,iCAAQ;AAAA,MAChB,UAAU;AAAA,IACZ;AAAA,IACA;AAAA,MACE,QAAQ;AAAA,MACR,mBAAmB,iCAAQ;AAAA,IAC7B;AAAA,EACF,EAAE,YAAY;AAAA,IACZ,QAAQ,iCAAQ;AAAA,IAChB,WAAW,iCAAQ;AAAA,IACnB,MAAM,iCAAQ;AAAA,IACd,MAAM,iCAAQ;AAAA,EAChB,CAAC;AAED,MAAI;AACF,UAAM,MAAM,aAAa,GAAG,IAAI;AAGhC,QAAI,gBAAgB,GAAG,GAAG;AACxB,aAAO,kBAAkB,KAAK,UAAU;AAAA,IAC1C;AAEA,QAAI,eAAe,SAAS;AAC1B,YAAM,iBAAiB,IACpB,KAAK,CAAC,WAAW;AAChB,YAAI,gBAAgB,MAAM,GAAG;AAC3B,iBAAO,kBAAkB,QAAQ,UAAU;AAAA,QAC7C;AAEA,cAAM,SAAS,sBAAsB,MAAM;AAC3C,cAAM,eAAe,8BAA8B,MAAM;AACzD,cAAM;AAAA,UACJ,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,IAAI,2BAA2B,MAAM;AAErC,mBACG,OAAO;AAAA,UACN;AAAA,UACA;AAAA,UACA,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,CAAC,EACA,IAAI;AAEP,eAAO;AAAA,MACT,CAAC,EACA,MAAM,CAAC,QAAQ;AACd,mBACG,OAAO;AAAA,UACN,eAAe,OAAO,GAAG;AAAA,UACzB,OAAO;AAAA,UACP,aAAa;AAAA,YACX,OAAO;AAAA,YACP,QAAQ;AAAA,YACR,OAAO;AAAA,UACT;AAAA,QACF,CAAC,EACA,IAAI;AAEP,cAAM;AAAA,MACR,CAAC;AAEH,aAAO;AAAA,IACT;AAEA,WAAO;AAAA,EACT,SAAS,OAAO;AACd,eACG,OAAO;AAAA,MACN,eAAe,OAAO,KAAK;AAAA,MAC3B,OAAO;AAAA,MACP,aAAa;AAAA,QACX,OAAO;AAAA,QACP,QAAQ;AAAA,QACR,OAAO;AAAA,MACT;AAAA,IACF,CAAC,EACA,IAAI;AAEP,UAAM;AAAA,EACR;AACF;AAeA,SAAS,kBACP,UACA,YACG;AACH,kBAAgB,wBAId;AACA,UAAM,WAAW;AACjB,UAAM,aAAuB,CAAC;AAC9B,UAAM,iBACJ,CAAC;AACH,QAAI,QAAuC;AAC3C,QAAI,sBAAwC;AAC5C,QAAI,eAAmD;AACvD,QAAI,SAAkB;AAEtB,qBAAiB,YAAY,UAAoC;AAC/D,4BAAsB,oDAAuB,oBAAI,KAAK;AAGtD,UAAI,OAAO,aAAa,YAAY,YAAY,cAAc,UAAU;AACtE,cAAM,SAAS,SAAS,UAAU;AAClC,iBAAS,sBAAsB,MAAM;AACrC,uBAAe,8BAA8B,MAAM;AAEnD,cAAM;AAAA,UACJ,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,IAAI,2BAA2B,MAAM;AAErC,mBAAW,OAAO;AAAA,UAChB,OAAO;AAAA,UACP,iBAAiB;AAAA,UACjB,UAAU;AAAA,QACZ,CAAC;AAAA,MACH;AAEA,UACE,OAAO,aAAa,YACpB,YAAY,QACZ,WAAW,UACX;AACA,gBAAQ,SAAS;AAAA,MACnB;AAEA,YAAM,iBAAiB,WAAW,QAAQ;AAE1C,UAAI,CAAC,eAAe,YAAY;AAC9B,mBAAW,KAAK,eAAe,IAAI;AAAA,MACrC,OAAO;AACL,uBAAe,KAAK,eAAe,IAAI;AAAA,MACzC;AAEA,YAAM;AAAA,IACR;AAEA,aACE,0BACC,eAAe,SAAS,IACrB,kBAAkB,cAAc,IAChC,WAAW,KAAK,EAAE;AAExB,eACG,OAAO;AAAA,MACN;AAAA,MACA;AAAA,MACA,cACE,sCAAiB,QAAQ,kBAAkB,KAAK,IAAI;AAAA,IACxD,CAAC,EACA,IAAI;AAAA,EACT;AAEA,SAAO,sBAAsB;AAC/B;;;AGjKO,IAAM,gBAAgB,CAC3B,KACA,mBACY;AACZ,SAAO,IAAI,MAAM,KAAK;AAAA,IACpB,IAAI,YAAY,SAAS,OAAO;AAhGpC;AAiGM,YAAM,mBAAmB,WAAW,OAAwB;AAE5D,YAAM,wBAAwB,IAAG,SAAI,gBAAJ,mBAAiB,IAAI,IAAI,QAAQ,SAAS,CAAC;AAC5E,YAAM,kBACJ,sDAAgB,mBAAhB,YAAkC;AACpC,YAAM,SAAS,EAAE,GAAG,gBAAgB,eAAe;AAGnD,UAAI,OAAO,qBAAqB,YAAY;AAC1C,eAAO,YAAY,iBAAiB,KAAK,UAAU,GAAG,MAAM;AAAA,MAC9D;AAEA,YAAM,uBACJ,oBACA,CAAC,MAAM,QAAQ,gBAAgB,KAC/B,EAAE,4BAA4B,SAC9B,OAAO,qBAAqB;AAG9B,UAAI,sBAAsB;AACxB,eAAO,cAAc,kBAAkB,MAAM;AAAA,MAC/C;AAGA,aAAO,QAAQ,IAAI,YAAY,SAAS,KAAK;AAAA,IAC/C;AAAA,EACF,CAAC;AACH;","names":[]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@elasticdash/openai",
-  "version": "0.0.1",
+  "version": "0.0.2",
   "description": "Langfuse integration for OpenAI SDK",
   "type": "module",
   "sideEffects": false,
@@ -14,14 +14,6 @@
       "require": "./dist/index.cjs"
     }
   },
-  "scripts": {
-    "build": "tsup",
-    "test": "vitest run",
-    "test:watch": "vitest",
-    "format": "prettier --write \"src/**/*.ts\"",
-    "format:check": "prettier --check \"src/**/*.ts\"",
-    "clean": "rm -rf dist"
-  },
   "author": "Langfuse",
   "license": "MIT",
   "repository": {
@@ -33,10 +25,18 @@
     "dist"
   ],
   "dependencies": {
-    "@elasticdash/core": "workspace:^",
-    "@elasticdash/tracing": "workspace:^"
+    "@elasticdash/core": "^0.0.2",
+    "@elasticdash/tracing": "^0.0.2"
   },
   "devDependencies": {
     "openai": "^5.0.0"
+  },
+  "scripts": {
+    "build": "tsup",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "format": "prettier --write \"src/**/*.ts\"",
+    "format:check": "prettier --check \"src/**/*.ts\"",
+    "clean": "rm -rf dist"
   }
-}
+}