llm-exe 3.0.0-beta.2 → 3.0.0

package/dist/index.d.mts

@@ -1142,6 +1142,7 @@ interface GenericEmbeddingOptions extends BaseLlmOptions {
 interface OpenAiEmbeddingOptions extends GenericEmbeddingOptions {
     model?: string;
     openAiApiKey?: string;
+    baseUrl?: string;
 }
 interface AmazonEmbeddingOptions extends GenericEmbeddingOptions {
     model: string;
@@ -1283,6 +1284,9 @@ type AllUseLlmOptions = AllLlm & {
     "openai.o4-mini": {
         input: Omit<OpenAiRequest, "model">;
     };
+    "anthropic.claude-opus-4-8": {
+        input: Omit<AnthropicRequest, "model">;
+    };
     "anthropic.claude-opus-4-7": {
         input: Omit<AnthropicRequest, "model">;
     };
@@ -1549,7 +1553,7 @@ interface Config<Pk = LlmProviderKey> {
      * Embedding configs do not use this — their flow dispatches via
      * `getEmbeddingOutputParser` instead.
      */
-    transformResponse?: (result: any, _config?: Config<any>) => OutputResult;
+    transformResponse?: (result: any, _config?: Config<any>, headers?: Record<string, string>) => OutputResult;
     /**
      * Marks this config as deprecated. When set, useLlm() will emit a one-time
      * deprecation warning to inform users about upcoming model shutdowns.
@@ -2018,6 +2022,7 @@ declare const configs: {
     "anthropic.claude-opus-4-1": Config<any>;
     "anthropic.claude-opus-4-6": Config<any>;
     "anthropic.chat.v1": Config<keyof AllLlm>;
+    "anthropic.claude-opus-4-8": Config<keyof AllLlm>;
     "anthropic.claude-opus-4-7": Config<keyof AllLlm>;
     "anthropic.claude-sonnet-4-6": Config<keyof AllLlm>;
     "anthropic.claude-opus-4-5": Config<keyof AllLlm>;

package/dist/index.d.ts

@@ -1142,6 +1142,7 @@ interface GenericEmbeddingOptions extends BaseLlmOptions {
 interface OpenAiEmbeddingOptions extends GenericEmbeddingOptions {
     model?: string;
     openAiApiKey?: string;
+    baseUrl?: string;
 }
 interface AmazonEmbeddingOptions extends GenericEmbeddingOptions {
     model: string;
@@ -1283,6 +1284,9 @@ type AllUseLlmOptions = AllLlm & {
     "openai.o4-mini": {
         input: Omit<OpenAiRequest, "model">;
     };
+    "anthropic.claude-opus-4-8": {
+        input: Omit<AnthropicRequest, "model">;
+    };
     "anthropic.claude-opus-4-7": {
         input: Omit<AnthropicRequest, "model">;
     };
@@ -1549,7 +1553,7 @@ interface Config<Pk = LlmProviderKey> {
      * Embedding configs do not use this — their flow dispatches via
      * `getEmbeddingOutputParser` instead.
      */
-    transformResponse?: (result: any, _config?: Config<any>) => OutputResult;
+    transformResponse?: (result: any, _config?: Config<any>, headers?: Record<string, string>) => OutputResult;
     /**
      * Marks this config as deprecated. When set, useLlm() will emit a one-time
      * deprecation warning to inform users about upcoming model shutdowns.
@@ -2018,6 +2022,7 @@ declare const configs: {
     "anthropic.claude-opus-4-1": Config<any>;
     "anthropic.claude-opus-4-6": Config<any>;
     "anthropic.chat.v1": Config<keyof AllLlm>;
+    "anthropic.claude-opus-4-8": Config<keyof AllLlm>;
     "anthropic.claude-opus-4-7": Config<keyof AllLlm>;
     "anthropic.claude-sonnet-4-6": Config<keyof AllLlm>;
     "anthropic.claude-opus-4-5": Config<keyof AllLlm>;

package/dist/index.js

@@ -4352,6 +4352,20 @@ function anthropicPromptSanitize(_messages, _inputBodyObj, _outputObj) {
   return mergeConsecutiveSameRole(result);
 }
 
+// src/llm/output/_utils/getBedrockTokenCounts.ts
+function getBedrockTokenCounts(headers) {
+  if (!headers) return void 0;
+  const input = parseInt(headers["x-amzn-bedrock-input-token-count"], 10);
+  const output = parseInt(headers["x-amzn-bedrock-output-token-count"], 10);
+  if (!Number.isFinite(input) && !Number.isFinite(output)) {
+    return void 0;
+  }
+  return {
+    input_tokens: Number.isFinite(input) ? input : 0,
+    output_tokens: Number.isFinite(output) ? output : 0
+  };
+}
+
 // src/llm/output/claude.ts
 function formatResult2(response) {
   const content = response?.content || [];
@@ -4374,15 +4388,18 @@ function formatResult2(response) {
   }
   return out;
 }
-function OutputAnthropicClaude3Chat(result, _config) {
+function OutputAnthropicClaude3Chat(result, _config, headers) {
   const id = result.id;
   const name = result.model || _config?.options.model?.default || "anthropic.unknown";
   const stopReason = result.stop_reason;
   const content = formatResult2(result);
+  const headerUsage = getBedrockTokenCounts(headers);
+  const input_tokens = result?.usage?.input_tokens ?? headerUsage?.input_tokens ?? 0;
+  const output_tokens = result?.usage?.output_tokens ?? headerUsage?.output_tokens ?? 0;
   const usage = {
-    input_tokens: result?.usage?.input_tokens,
-    output_tokens: result?.usage?.output_tokens,
-    total_tokens: result?.usage?.input_tokens + result?.usage?.output_tokens
+    input_tokens,
+    output_tokens,
+    total_tokens: input_tokens + output_tokens
   };
   return {
     id,
@@ -4395,7 +4412,7 @@ function OutputAnthropicClaude3Chat(result, _config) {
 }
 
 // src/llm/output/llama.ts
-function OutputMetaLlama3Chat(result, _config) {
+function OutputMetaLlama3Chat(result, _config, headers) {
   const id = (0, import_uuid.v4)();
   const name = _config?.options?.model?.default || "meta";
   const created = (/* @__PURE__ */ new Date()).getTime();
@@ -4403,10 +4420,13 @@ function OutputMetaLlama3Chat(result, _config) {
   const content = [
     { type: "text", text: result.generation }
   ];
+  const headerUsage = getBedrockTokenCounts(headers);
+  const output_tokens = result?.generation_token_count ?? headerUsage?.output_tokens ?? 0;
+  const input_tokens = result?.prompt_token_count ?? headerUsage?.input_tokens ?? 0;
   const usage = {
-    output_tokens: result?.generation_token_count,
-    input_tokens: result?.prompt_token_count,
-    total_tokens: result?.generation_token_count + result?.prompt_token_count
+    output_tokens,
+    input_tokens,
+    total_tokens: output_tokens + input_tokens
   };
   return {
     id,
@@ -4524,7 +4544,7 @@ var bedrock = {
 
 // src/llm/config/anthropic/index.ts
 var ANTHROPIC_VERSION = "2023-06-01";
-var MODELS_REJECTING_SAMPLING_PARAMS = ["claude-opus-4-7"];
+var MODELS_REJECTING_SAMPLING_PARAMS = ["claude-opus-4-7", "claude-opus-4-8"];
 var isClaude4x = (model) => /^claude-(opus|sonnet|haiku)-4-/.test(model);
 var dropIfModelRejectsSamplingParams = (v, body) => MODELS_REJECTING_SAMPLING_PARAMS.includes(body.model) ? void 0 : v;
 var topPTransform = (v, body) => {
@@ -4609,6 +4629,11 @@ var anthropicChatV1 = {
 };
 var anthropic = {
   "anthropic.chat.v1": anthropicChatV1,
+  // Claude 4.8 models
+  "anthropic.claude-opus-4-8": withDefaultModel(
+    anthropicChatV1,
+    "claude-opus-4-8"
+  ),
   // Claude 4.7 models
   "anthropic.claude-opus-4-7": withDefaultModel(
     anthropicChatV1,
@@ -4792,7 +4817,11 @@ var ollamaChatV1 = {
   provider: "ollama.chat",
   endpoint: `${getEnvironmentVariable("OLLAMA_ENDPOINT") || `http://localhost:11434`}/api/chat`,
   options: {
-    prompt: {}
+    prompt: {},
+    temperature: {},
+    topP: {},
+    maxTokens: {},
+    stopSequences: {}
   },
   method: "POST",
   headers: `{"Content-Type": "application/json" }`,
@@ -4808,6 +4837,18 @@ var ollamaChatV1 = {
     },
     model: {
       key: "model"
+    },
+    temperature: {
+      key: "options.temperature"
+    },
+    topP: {
+      key: "options.top_p"
+    },
+    maxTokens: {
+      key: "options.num_predict"
+    },
+    stopSequences: {
+      key: "options.stop"
     }
   },
   transformResponse: OutputOllamaChat
@@ -5003,7 +5044,10 @@ var googleGeminiChatV1 = {
   options: {
     effort: {},
     prompt: {},
-    // topP: {},
+    temperature: {},
+    topP: {},
+    maxTokens: {},
+    stopSequences: {},
     geminiApiKey: {
       default: getEnvironmentVariable("GEMINI_API_KEY")
     }
@@ -5015,6 +5059,18 @@ var googleGeminiChatV1 = {
       key: "contents",
       transform: googleGeminiPromptSanitize
     },
+    temperature: {
+      key: "generationConfig.temperature"
+    },
+    topP: {
+      key: "generationConfig.topP"
+    },
+    maxTokens: {
+      key: "generationConfig.maxOutputTokens"
+    },
+    stopSequences: {
+      key: "generationConfig.stopSequences"
+    },
     effort: {
       key: "config.thinkingConfig.thinkingBudget",
       transform: (v, _s) => {
@@ -5156,6 +5212,13 @@ function isValidUrl(input) {
 }
 
 // src/utils/modules/request.ts
+function headersToRecord(headers) {
+  const record = {};
+  headers.forEach((value, key) => {
+    record[key.toLowerCase()] = value;
+  });
+  return record;
+}
 async function apiRequest(url, options) {
   const finalOptions = {
     ...options
@@ -5241,13 +5304,16 @@ async function apiRequest(url, options) {
     });
   }
   const contentType = response.headers.get("content-type");
+  const headers = headersToRecord(response.headers);
   if (contentType?.includes("application/json")) {
-    const responseData = await response.json();
-    return responseData;
-  } else {
-    const responseData = await response.text();
-    return responseData;
+    const responseData2 = await response.json();
+    return { data: responseData2, headers };
   }
+  const responseData = await response.text();
+  return {
+    data: responseData,
+    headers
+  };
 }
 
 // src/utils/modules/convertDotNotation.ts
@@ -5619,7 +5685,7 @@ async function useLlm_call(state, messages, _options) {
         }
       ]
     };
-    return BaseLlmOutput(transformResponse(mockResponse, config));
+    return BaseLlmOutput(transformResponse(mockResponse, config, {}));
   }
   try {
     const response = await apiRequest(url, {
@@ -5627,7 +5693,9 @@ async function useLlm_call(state, messages, _options) {
       body,
       headers
     });
-    return BaseLlmOutput(transformResponse(response, config));
+    return BaseLlmOutput(
+      transformResponse(response.data, config, response.headers)
+    );
   } catch (e) {
     if (!isLlmExeError(e, "request.http_error")) throw e;
     const ctx = e.context ?? {};
@@ -5808,7 +5876,9 @@ var embeddingConfigs = {
   "openai.embedding.v1": {
     key: "openai.embedding.v1",
     provider: "openai.embedding",
-    endpoint: `https://api.openai.com/v1/embeddings`,
+    // Templated host: `baseUrl` defaults to OpenAI. Override to point at any
+    // OpenAI-compatible embeddings server (Baseten, vLLM, TEI, Together, etc.).
+    endpoint: `{{baseUrl}}/embeddings`,
     method: "POST",
     headers: `{"Authorization":"Bearer {{openAiApiKey}}", "Content-Type": "application/json" }`,
     options: {
@@ -5819,6 +5889,9 @@ var embeddingConfigs = {
       encodingFormat: {},
       openAiApiKey: {
         default: getEnvironmentVariable("OPENAI_API_KEY")
+      },
+      baseUrl: {
+        default: "https://api.openai.com/v1"
       }
     },
     mapBody: {
@@ -6005,15 +6078,34 @@ function AmazonTitanEmbedding(result, config) {
 }
 
 // src/embedding/output/CohereBedrockEmbedding.ts
-function CohereBedrockEmbedding(result, config) {
+function resolveEmbeddings(embeddings, model) {
+  if (Array.isArray(embeddings)) {
+    return embeddings;
+  }
+  if (embeddings && typeof embeddings === "object") {
+    if (Array.isArray(embeddings.float)) {
+      return embeddings.float;
+    }
+    throw new Error(
+      `Unexpected embeddings in Cohere Bedrock response (model: "${model}"). Expected float embeddings, received object with keys: [${Object.keys(embeddings).join(", ")}]`
+    );
+  }
+  throw new Error(
+    `Unexpected embeddings shape in Cohere Bedrock response (model: "${model}"). Expected an array (Embed v3) or an object with float embeddings (Embed v4), received: ${embeddings === null ? "null" : typeof embeddings}`
+  );
+}
+function CohereBedrockEmbedding(result, config, headers) {
+  const headerUsage = getBedrockTokenCounts(headers);
   const __result = deepClone(result);
   const model = config.model || "cohere.unknown";
   const created = (/* @__PURE__ */ new Date()).getTime();
-  const embedding = Array.isArray(__result.embeddings) ? __result.embeddings : [];
+  const embedding = resolveEmbeddings(__result.embeddings, model);
+  const input_tokens = headerUsage?.input_tokens ?? 0;
+  const output_tokens = headerUsage?.output_tokens ?? 0;
   const usage = {
-    output_tokens: 0,
-    input_tokens: 0,
-    total_tokens: 0
+    output_tokens,
+    input_tokens,
+    total_tokens: input_tokens + output_tokens
   };
   return BaseEmbeddingOutput({
     id: __result.id,
@@ -6045,14 +6137,14 @@ function OpenAiEmbedding(result, config) {
 }
 
 // src/embedding/output/getEmbeddingOutputParser.ts
-function getEmbeddingOutputParser(config, response) {
+function getEmbeddingOutputParser(config, response, headers) {
   switch (config.key) {
     case "openai.embedding.v1":
       return OpenAiEmbedding(response, config);
     case "amazon.embedding.v1":
       return AmazonTitanEmbedding(response, config);
     case "amazon:cohere.embedding.v1":
-      return CohereBedrockEmbedding(response, config);
+      return CohereBedrockEmbedding(response, config, headers);
     default:
       throw new LlmExeError("Unsupported provider", {
         code: "embedding.invalid_response_shape",
@@ -6089,7 +6181,7 @@ async function createEmbedding_call(state, _input, _options) {
       body,
       headers
     });
-    return getEmbeddingOutputParser(state, request);
+    return getEmbeddingOutputParser(state, request.data, request.headers);
   } catch (e) {
     if (!isLlmExeError(e, "request.http_error")) throw e;
     const ctx = e.context ?? {};

package/dist/index.mjs

@@ -4286,6 +4286,20 @@ function anthropicPromptSanitize(_messages, _inputBodyObj, _outputObj) {
   return mergeConsecutiveSameRole(result);
 }
 
+// src/llm/output/_utils/getBedrockTokenCounts.ts
+function getBedrockTokenCounts(headers) {
+  if (!headers) return void 0;
+  const input = parseInt(headers["x-amzn-bedrock-input-token-count"], 10);
+  const output = parseInt(headers["x-amzn-bedrock-output-token-count"], 10);
+  if (!Number.isFinite(input) && !Number.isFinite(output)) {
+    return void 0;
+  }
+  return {
+    input_tokens: Number.isFinite(input) ? input : 0,
+    output_tokens: Number.isFinite(output) ? output : 0
+  };
+}
+
 // src/llm/output/claude.ts
 function formatResult2(response) {
   const content = response?.content || [];
@@ -4308,15 +4322,18 @@ function formatResult2(response) {
   }
   return out;
 }
-function OutputAnthropicClaude3Chat(result, _config) {
+function OutputAnthropicClaude3Chat(result, _config, headers) {
   const id = result.id;
   const name = result.model || _config?.options.model?.default || "anthropic.unknown";
   const stopReason = result.stop_reason;
   const content = formatResult2(result);
+  const headerUsage = getBedrockTokenCounts(headers);
+  const input_tokens = result?.usage?.input_tokens ?? headerUsage?.input_tokens ?? 0;
+  const output_tokens = result?.usage?.output_tokens ?? headerUsage?.output_tokens ?? 0;
   const usage = {
-    input_tokens: result?.usage?.input_tokens,
-    output_tokens: result?.usage?.output_tokens,
-    total_tokens: result?.usage?.input_tokens + result?.usage?.output_tokens
+    input_tokens,
+    output_tokens,
+    total_tokens: input_tokens + output_tokens
   };
   return {
     id,
@@ -4329,7 +4346,7 @@ function OutputAnthropicClaude3Chat(result, _config) {
 }
 
 // src/llm/output/llama.ts
-function OutputMetaLlama3Chat(result, _config) {
+function OutputMetaLlama3Chat(result, _config, headers) {
   const id = uuidv4();
   const name = _config?.options?.model?.default || "meta";
   const created = (/* @__PURE__ */ new Date()).getTime();
@@ -4337,10 +4354,13 @@ function OutputMetaLlama3Chat(result, _config) {
   const content = [
     { type: "text", text: result.generation }
   ];
+  const headerUsage = getBedrockTokenCounts(headers);
+  const output_tokens = result?.generation_token_count ?? headerUsage?.output_tokens ?? 0;
+  const input_tokens = result?.prompt_token_count ?? headerUsage?.input_tokens ?? 0;
   const usage = {
-    output_tokens: result?.generation_token_count,
-    input_tokens: result?.prompt_token_count,
-    total_tokens: result?.generation_token_count + result?.prompt_token_count
+    output_tokens,
+    input_tokens,
+    total_tokens: output_tokens + input_tokens
   };
   return {
     id,
@@ -4458,7 +4478,7 @@ var bedrock = {
 
 // src/llm/config/anthropic/index.ts
 var ANTHROPIC_VERSION = "2023-06-01";
-var MODELS_REJECTING_SAMPLING_PARAMS = ["claude-opus-4-7"];
+var MODELS_REJECTING_SAMPLING_PARAMS = ["claude-opus-4-7", "claude-opus-4-8"];
 var isClaude4x = (model) => /^claude-(opus|sonnet|haiku)-4-/.test(model);
 var dropIfModelRejectsSamplingParams = (v, body) => MODELS_REJECTING_SAMPLING_PARAMS.includes(body.model) ? void 0 : v;
 var topPTransform = (v, body) => {
@@ -4543,6 +4563,11 @@ var anthropicChatV1 = {
 };
 var anthropic = {
   "anthropic.chat.v1": anthropicChatV1,
+  // Claude 4.8 models
+  "anthropic.claude-opus-4-8": withDefaultModel(
+    anthropicChatV1,
+    "claude-opus-4-8"
+  ),
   // Claude 4.7 models
   "anthropic.claude-opus-4-7": withDefaultModel(
     anthropicChatV1,
@@ -4726,7 +4751,11 @@ var ollamaChatV1 = {
   provider: "ollama.chat",
   endpoint: `${getEnvironmentVariable("OLLAMA_ENDPOINT") || `http://localhost:11434`}/api/chat`,
   options: {
-    prompt: {}
+    prompt: {},
+    temperature: {},
+    topP: {},
+    maxTokens: {},
+    stopSequences: {}
   },
   method: "POST",
   headers: `{"Content-Type": "application/json" }`,
@@ -4742,6 +4771,18 @@ var ollamaChatV1 = {
     },
     model: {
       key: "model"
+    },
+    temperature: {
+      key: "options.temperature"
+    },
+    topP: {
+      key: "options.top_p"
+    },
+    maxTokens: {
+      key: "options.num_predict"
+    },
+    stopSequences: {
+      key: "options.stop"
     }
   },
   transformResponse: OutputOllamaChat
@@ -4937,7 +4978,10 @@ var googleGeminiChatV1 = {
   options: {
     effort: {},
     prompt: {},
-    // topP: {},
+    temperature: {},
+    topP: {},
+    maxTokens: {},
+    stopSequences: {},
     geminiApiKey: {
       default: getEnvironmentVariable("GEMINI_API_KEY")
     }
@@ -4949,6 +4993,18 @@ var googleGeminiChatV1 = {
       key: "contents",
       transform: googleGeminiPromptSanitize
     },
+    temperature: {
+      key: "generationConfig.temperature"
+    },
+    topP: {
+      key: "generationConfig.topP"
+    },
+    maxTokens: {
+      key: "generationConfig.maxOutputTokens"
+    },
+    stopSequences: {
+      key: "generationConfig.stopSequences"
+    },
     effort: {
       key: "config.thinkingConfig.thinkingBudget",
       transform: (v, _s) => {
@@ -5090,6 +5146,13 @@ function isValidUrl(input) {
 }
 
 // src/utils/modules/request.ts
+function headersToRecord(headers) {
+  const record = {};
+  headers.forEach((value, key) => {
+    record[key.toLowerCase()] = value;
+  });
+  return record;
+}
 async function apiRequest(url, options) {
   const finalOptions = {
     ...options
@@ -5175,13 +5238,16 @@ async function apiRequest(url, options) {
     });
   }
   const contentType = response.headers.get("content-type");
+  const headers = headersToRecord(response.headers);
   if (contentType?.includes("application/json")) {
-    const responseData = await response.json();
-    return responseData;
-  } else {
-    const responseData = await response.text();
-    return responseData;
+    const responseData2 = await response.json();
+    return { data: responseData2, headers };
   }
+  const responseData = await response.text();
+  return {
+    data: responseData,
+    headers
+  };
 }
 
 // src/utils/modules/convertDotNotation.ts
@@ -5553,7 +5619,7 @@ async function useLlm_call(state, messages, _options) {
         }
       ]
     };
-    return BaseLlmOutput(transformResponse(mockResponse, config));
+    return BaseLlmOutput(transformResponse(mockResponse, config, {}));
   }
   try {
     const response = await apiRequest(url, {
@@ -5561,7 +5627,9 @@ async function useLlm_call(state, messages, _options) {
       body,
       headers
     });
-    return BaseLlmOutput(transformResponse(response, config));
+    return BaseLlmOutput(
+      transformResponse(response.data, config, response.headers)
+    );
   } catch (e) {
     if (!isLlmExeError(e, "request.http_error")) throw e;
     const ctx = e.context ?? {};
@@ -5742,7 +5810,9 @@ var embeddingConfigs = {
   "openai.embedding.v1": {
     key: "openai.embedding.v1",
     provider: "openai.embedding",
-    endpoint: `https://api.openai.com/v1/embeddings`,
+    // Templated host: `baseUrl` defaults to OpenAI. Override to point at any
+    // OpenAI-compatible embeddings server (Baseten, vLLM, TEI, Together, etc.).
+    endpoint: `{{baseUrl}}/embeddings`,
     method: "POST",
     headers: `{"Authorization":"Bearer {{openAiApiKey}}", "Content-Type": "application/json" }`,
     options: {
@@ -5753,6 +5823,9 @@ var embeddingConfigs = {
       encodingFormat: {},
       openAiApiKey: {
         default: getEnvironmentVariable("OPENAI_API_KEY")
+      },
+      baseUrl: {
+        default: "https://api.openai.com/v1"
       }
     },
     mapBody: {
@@ -5939,15 +6012,34 @@ function AmazonTitanEmbedding(result, config) {
 }
 
 // src/embedding/output/CohereBedrockEmbedding.ts
-function CohereBedrockEmbedding(result, config) {
+function resolveEmbeddings(embeddings, model) {
+  if (Array.isArray(embeddings)) {
+    return embeddings;
+  }
+  if (embeddings && typeof embeddings === "object") {
+    if (Array.isArray(embeddings.float)) {
+      return embeddings.float;
+    }
+    throw new Error(
+      `Unexpected embeddings in Cohere Bedrock response (model: "${model}"). Expected float embeddings, received object with keys: [${Object.keys(embeddings).join(", ")}]`
+    );
+  }
+  throw new Error(
+    `Unexpected embeddings shape in Cohere Bedrock response (model: "${model}"). Expected an array (Embed v3) or an object with float embeddings (Embed v4), received: ${embeddings === null ? "null" : typeof embeddings}`
+  );
+}
+function CohereBedrockEmbedding(result, config, headers) {
+  const headerUsage = getBedrockTokenCounts(headers);
   const __result = deepClone(result);
   const model = config.model || "cohere.unknown";
   const created = (/* @__PURE__ */ new Date()).getTime();
-  const embedding = Array.isArray(__result.embeddings) ? __result.embeddings : [];
+  const embedding = resolveEmbeddings(__result.embeddings, model);
+  const input_tokens = headerUsage?.input_tokens ?? 0;
+  const output_tokens = headerUsage?.output_tokens ?? 0;
   const usage = {
-    output_tokens: 0,
-    input_tokens: 0,
-    total_tokens: 0
+    output_tokens,
+    input_tokens,
+    total_tokens: input_tokens + output_tokens
   };
   return BaseEmbeddingOutput({
     id: __result.id,
@@ -5979,14 +6071,14 @@ function OpenAiEmbedding(result, config) {
 }
 
 // src/embedding/output/getEmbeddingOutputParser.ts
-function getEmbeddingOutputParser(config, response) {
+function getEmbeddingOutputParser(config, response, headers) {
   switch (config.key) {
     case "openai.embedding.v1":
       return OpenAiEmbedding(response, config);
     case "amazon.embedding.v1":
       return AmazonTitanEmbedding(response, config);
     case "amazon:cohere.embedding.v1":
-      return CohereBedrockEmbedding(response, config);
+      return CohereBedrockEmbedding(response, config, headers);
     default:
       throw new LlmExeError("Unsupported provider", {
         code: "embedding.invalid_response_shape",
@@ -6023,7 +6115,7 @@ async function createEmbedding_call(state, _input, _options) {
       body,
       headers
     });
-    return getEmbeddingOutputParser(state, request);
+    return getEmbeddingOutputParser(state, request.data, request.headers);
   } catch (e) {
     if (!isLlmExeError(e, "request.http_error")) throw e;
     const ctx = e.context ?? {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llm-exe",
-  "version": "3.0.0-beta.2",
+  "version": "3.0.0",
   "description": "Simplify building LLM-powered apps with easy-to-use base components, supporting text and chat-based prompts with handlebars template engine, output parsers, and flexible function calling capabilities.",
   "keywords": [
     "ai",

package/readme.md

@@ -9,7 +9,7 @@ A package that provides simplified base components to make building and maintain
 - Write functions powered by LLM's with easy to use building blocks.
 - Pure Javascript and Typescript. Allows you to pass and infer types.
 - Supercharge your prompts by using handlebars within prompt template.
-- Support for text-based (llama-3) and chat-based prompts. (gpt-4o, claude-3.5, grok-3, Gemini, Bedrock, Ollama, etc)
+- Support for text-based and chat-based prompts. (ChatGPT, Claude, Grok, Gemini, Bedrock, Ollama, etc)
 - Call LLM's from different providers without changing your code. (OpenAi/Anthropic/xAI/Google/AWS Bedrock/Ollama/Deepseek)
 - Allow LLM's to call functions (or call other LLM executors).
 - Not very opinionated. You have control on how you use it.