@gammatech/aijsx 0.9.0-dev.2024-05-28 → 0.9.1-dev.2024-05-28

package/dist/index.d.mts

@@ -6,6 +6,8 @@ export { OpenAI as OpenAIClient } from 'openai';
 import { ChatCompletionSystemMessageParam, ChatCompletionUserMessageParam, ChatCompletionAssistantMessageParam, ChatCompletionCreateParams } from 'openai/resources';
 import AnthropicClient from '@anthropic-ai/sdk';
 export { default as AnthropicClient } from '@anthropic-ai/sdk';
+import { GenerateContentRequest, VertexAI } from '@google-cloud/vertexai';
+export { VertexAI } from '@google-cloud/vertexai';
 
 declare class ChatCompletionError extends Error {
     readonly chatCompletionRequest: LogChatCompletionRequest;
@@ -273,4 +275,22 @@ declare function AnthropicChatCompletion(props: AnthropicChatCompletionProps, ct
 
 declare const anthropicTokenizer: TokenizerFn;
 
-export { AIComponent, AINode, AISpanAttributes, AISpanProcessor, AnthropicChatCompletion, type AnthropicChatCompletionRequest, AnthropicClientContext, AssistantMessage, type ChatCompletionClientAndProvider, ChatCompletionError, type ChatCompletionRequestPayloads, ChatMessage, Context, type CostFn, DebugMessage, DefaultMaxRetriesContext, EnrichingSpanProcessor, EvaluatorFn, EvaluatorResult, Fallback, FunctionChain, LogChatCompletionRequest, LogImplementation, NotAsyncGenerator, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, StreamChain, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createFunctionChain, createPrompt, createRenderContext, createStreamChain, evaluatePrompt, openaiTokenizer, tracing };
+type GoogleChatCompletionRequest = GenerateContentRequest;
+declare module '@gammatech/aijsx' {
+    interface ChatCompletionRequestPayloads {
+        google: GoogleChatCompletionRequest;
+    }
+}
+type ValidGoogleChatModel = 'gemini-1.5-pro' | 'gemini-1.5-flash';
+declare const GoogleClientContext: Context<() => ChatCompletionClientAndProvider<VertexAI>>;
+type GoogleChatCompletionProps = {
+    model: ValidGoogleChatModel;
+    maxTokens?: number;
+    temperature?: number;
+    stop?: string | string[];
+    maxRetries?: number;
+    children: AINode;
+};
+declare function GoogleChatCompletion(props: GoogleChatCompletionProps, ctx: RenderContext): JSX.Element;
+
+export { AIComponent, AINode, AISpanAttributes, AISpanProcessor, AnthropicChatCompletion, type AnthropicChatCompletionRequest, AnthropicClientContext, AssistantMessage, type ChatCompletionClientAndProvider, ChatCompletionError, type ChatCompletionRequestPayloads, ChatMessage, Context, type CostFn, DebugMessage, DefaultMaxRetriesContext, EnrichingSpanProcessor, EvaluatorFn, EvaluatorResult, Fallback, FunctionChain, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, NotAsyncGenerator, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, StreamChain, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createFunctionChain, createPrompt, createRenderContext, createStreamChain, evaluatePrompt, openaiTokenizer, tracing };

package/dist/index.d.ts

@@ -6,6 +6,8 @@ export { OpenAI as OpenAIClient } from 'openai';
 import { ChatCompletionSystemMessageParam, ChatCompletionUserMessageParam, ChatCompletionAssistantMessageParam, ChatCompletionCreateParams } from 'openai/resources';
 import AnthropicClient from '@anthropic-ai/sdk';
 export { default as AnthropicClient } from '@anthropic-ai/sdk';
+import { GenerateContentRequest, VertexAI } from '@google-cloud/vertexai';
+export { VertexAI } from '@google-cloud/vertexai';
 
 declare class ChatCompletionError extends Error {
     readonly chatCompletionRequest: LogChatCompletionRequest;
@@ -273,4 +275,22 @@ declare function AnthropicChatCompletion(props: AnthropicChatCompletionProps, ct
 
 declare const anthropicTokenizer: TokenizerFn;
 
-export { AIComponent, AINode, AISpanAttributes, AISpanProcessor, AnthropicChatCompletion, type AnthropicChatCompletionRequest, AnthropicClientContext, AssistantMessage, type ChatCompletionClientAndProvider, ChatCompletionError, type ChatCompletionRequestPayloads, ChatMessage, Context, type CostFn, DebugMessage, DefaultMaxRetriesContext, EnrichingSpanProcessor, EvaluatorFn, EvaluatorResult, Fallback, FunctionChain, LogChatCompletionRequest, LogImplementation, NotAsyncGenerator, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, StreamChain, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createFunctionChain, createPrompt, createRenderContext, createStreamChain, evaluatePrompt, openaiTokenizer, tracing };
+type GoogleChatCompletionRequest = GenerateContentRequest;
+declare module '@gammatech/aijsx' {
+    interface ChatCompletionRequestPayloads {
+        google: GoogleChatCompletionRequest;
+    }
+}
+type ValidGoogleChatModel = 'gemini-1.5-pro' | 'gemini-1.5-flash';
+declare const GoogleClientContext: Context<() => ChatCompletionClientAndProvider<VertexAI>>;
+type GoogleChatCompletionProps = {
+    model: ValidGoogleChatModel;
+    maxTokens?: number;
+    temperature?: number;
+    stop?: string | string[];
+    maxRetries?: number;
+    children: AINode;
+};
+declare function GoogleChatCompletion(props: GoogleChatCompletionProps, ctx: RenderContext): JSX.Element;
+
+export { AIComponent, AINode, AISpanAttributes, AISpanProcessor, AnthropicChatCompletion, type AnthropicChatCompletionRequest, AnthropicClientContext, AssistantMessage, type ChatCompletionClientAndProvider, ChatCompletionError, type ChatCompletionRequestPayloads, ChatMessage, Context, type CostFn, DebugMessage, DefaultMaxRetriesContext, EnrichingSpanProcessor, EvaluatorFn, EvaluatorResult, Fallback, FunctionChain, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, NotAsyncGenerator, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, StreamChain, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createFunctionChain, createPrompt, createRenderContext, createStreamChain, evaluatePrompt, openaiTokenizer, tracing };

package/dist/index.js

@@ -42,6 +42,8 @@ __export(src_exports, {
   DefaultMaxRetriesContext: () => DefaultMaxRetriesContext,
   EnrichingSpanProcessor: () => EnrichingSpanProcessor,
   Fallback: () => Fallback,
+  GoogleChatCompletion: () => GoogleChatCompletion,
+  GoogleClientContext: () => GoogleClientContext,
   ImagePart: () => ImagePart,
   LogImplementation: () => LogImplementation,
   NoopLogImplementation: () => NoopLogImplementation,
@@ -55,6 +57,7 @@ __export(src_exports, {
   SystemMessage: () => SystemMessage,
   Trace: () => Trace,
   UserMessage: () => UserMessage,
+  VertexAI: () => import_vertexai2.VertexAI,
   anthropicTokenizer: () => anthropicTokenizer,
   attachedContextSymbol: () => attachedContextSymbol,
   computeUsage: () => computeUsage,
@@ -73,11 +76,11 @@ module.exports = __toCommonJS(src_exports);
 
 // src/chat/errors.ts
 var ChatCompletionError = class extends Error {
-  constructor(message, chatCompletionRequest, status, shouldRetry3 = false) {
+  constructor(message, chatCompletionRequest, status, shouldRetry4 = false) {
     super(message);
     this.chatCompletionRequest = chatCompletionRequest;
     this.status = status;
-    this.shouldRetry = shouldRetry3;
+    this.shouldRetry = shouldRetry4;
   }
   name = "ChatCompletionError";
 };
@@ -1599,7 +1602,7 @@ function renderCloseTag(element) {
 // src/retry.tsx
 var RetryCountContext = createContext(0);
 var DefaultMaxRetriesContext = createContext(0);
-async function* Retry({ shouldRetry: shouldRetry3, retries = 0, maxRetries = 3, children }, ctx) {
+async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, children }, ctx) {
   const { render } = ctx;
   let hasYieldedData = false;
   try {
@@ -1611,12 +1614,12 @@ async function* Retry({ shouldRetry: shouldRetry3, retries = 0, maxRetries = 3,
       yield value;
     }
   } catch (e) {
-    if (hasYieldedData || retries >= maxRetries || !shouldRetry3(e)) {
+    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(e)) {
       throw e;
     }
     await backoff(retries);
     yield* Retry(
-      { shouldRetry: shouldRetry3, retries: retries + 1, maxRetries, children },
+      { shouldRetry: shouldRetry4, retries: retries + 1, maxRetries, children },
       ctx
     );
   }
@@ -2527,6 +2530,259 @@ function cleanChatCompletionRequest2(chatCompletionRequest) {
 
 // src/lib/anthropic/index.ts
 var import_sdk2 = __toESM(require("@anthropic-ai/sdk"));
+
+// src/lib/google/Google.tsx
+var import_vertexai = require("@google-cloud/vertexai");
+var GoogleClientContext = createContext(() => {
+  if (defaultClient3) {
+    return defaultClient3;
+  }
+  const project = process.env.GOOGLE_PROJECT_ID || "";
+  const client = new import_vertexai.VertexAI({ project });
+  defaultClient3 = {
+    client,
+    provider: "google"
+  };
+  return defaultClient3;
+});
+var defaultClient3 = null;
+var buildGoogleMessages = (chatMesssages) => {
+  let systemInstruction;
+  const messages = [];
+  chatMesssages.forEach(({ role, content }) => {
+    if (role === "system") {
+      systemInstruction = content;
+      return;
+    }
+    if (role === "user") {
+      const userContent = content;
+      if (userContent.length === 1 && userContent[0].type === "text") {
+        messages.push({
+          role,
+          parts: [{ text: userContent[0].text }]
+        });
+        return;
+      }
+      const c = userContent.map((part) => {
+        if (part.type === "text") {
+          return { text: part.text };
+        } else if (part.type === "image") {
+          const imagePart = {
+            inlineData: {
+              mimeType: part.image.mediaType,
+              data: part.image.data
+            }
+          };
+          return imagePart;
+        }
+        throw new Error("Invalid part");
+      });
+      messages.push({
+        role,
+        parts: c
+      });
+      return;
+    }
+    if (role === "assistant") {
+      messages.push({
+        role,
+        parts: [{ text: content }]
+      });
+      return;
+    }
+    throw new Error(`Invalid role: ${role}`);
+  });
+  return {
+    systemInstruction,
+    messages
+  };
+};
+var shouldRetry3 = (error) => {
+  return error instanceof ChatCompletionError && error.shouldRetry;
+};
+function GoogleChatCompletion(props, ctx) {
+  const defaultMaxRetries = ctx.getContext(DefaultMaxRetriesContext);
+  return /* @__PURE__ */ jsx(
+    Retry,
+    {
+      maxRetries: props.maxRetries || defaultMaxRetries,
+      shouldRetry: shouldRetry3,
+      children: /* @__PURE__ */ jsx(Trace, { name: "ai.chatCompletion", children: /* @__PURE__ */ jsx(GoogleChatCompletionInner, { ...props }) })
+    }
+  );
+}
+async function* GoogleChatCompletionInner(props, ctx) {
+  const startTime = performance.now();
+  const { logger, tracer, getContext } = ctx;
+  const retryCount = getContext(RetryCountContext);
+  const span = tracer.getActiveSpan();
+  const { client, provider, providerRegion, costFn } = getContext(GoogleClientContext)();
+  if (!client) {
+    throw new Error(
+      "[GoogleChatCompletion] must supply GoogleClient via context"
+    );
+  }
+  const chatMessages = await buildChatMessages(ctx, props.children, {
+    useBase64Images: true
+  });
+  const { systemInstruction, messages } = buildGoogleMessages(chatMessages);
+  const inputMessages = chatMessages.map((m) => toDebugMessage(m));
+  let stopSequences;
+  if (props.stop && typeof props.stop === "string") {
+    stopSequences = [props.stop];
+  } else if (Array.isArray(props.stop)) {
+    stopSequences = props.stop;
+  }
+  const googleCompletionRequest = {
+    systemInstruction,
+    contents: messages,
+    generationConfig: {
+      maxOutputTokens: props.maxTokens,
+      temperature: props.temperature,
+      stopSequences
+    }
+  };
+  const chatCompletionRequestToLog = cleanChatCompletionRequest3(
+    googleCompletionRequest
+  );
+  const logRequestData = {
+    startTime,
+    model: props.model,
+    provider,
+    providerRegion,
+    inputMessages,
+    request: chatCompletionRequestToLog
+  };
+  logger.chatCompletionRequest("google", logRequestData);
+  span.setAttributes({
+    model: props.model,
+    provider,
+    providerRegion,
+    requestType: "google",
+    chatCompletionRequest: chatCompletionRequestToLog,
+    inputMessages,
+    retryCount
+  });
+  const model = client.getGenerativeModel({ model: props.model });
+  let response;
+  try {
+    response = await model.generateContentStream(googleCompletionRequest);
+  } catch (err) {
+    if (err instanceof import_vertexai.GoogleGenerativeAIError) {
+      throw new ChatCompletionError(
+        err.message,
+        logRequestData,
+        void 0,
+        // always retry on GoogleGenerativeAIError errors, they represent non 4xx error cases
+        true
+      );
+    } else if (err instanceof Error) {
+      const status = void 0;
+      const retry = false;
+      throw new ChatCompletionError(err.message, logRequestData, status, retry);
+    }
+    throw err;
+  }
+  let content = "";
+  let outputUsage = 0;
+  let inputUsage = 0;
+  let finishReason = null;
+  try {
+    for await (const event of response.stream) {
+      if (event.candidates) {
+        const chunk = event.candidates[0].content;
+        if (event.candidates[0]?.finishReason) {
+          finishReason = event.candidates[0].finishReason;
+        }
+        if (event.usageMetadata) {
+          if (event.usageMetadata.promptTokenCount) {
+            inputUsage = event.usageMetadata.promptTokenCount;
+          }
+          if (event.usageMetadata.candidatesTokenCount) {
+            outputUsage = event.usageMetadata.candidatesTokenCount;
+          }
+        }
+        for (const block of chunk.parts) {
+          if (block.text) {
+            content += block.text;
+            yield block.text;
+          }
+        }
+      }
+    }
+  } catch (err) {
+    if (err instanceof import_vertexai.GoogleGenerativeAIError) {
+      throw new ChatCompletionError(
+        err.message,
+        logRequestData,
+        void 0,
+        // always retry these errors they represent non 4xx error cases
+        true
+      );
+    } else if (err instanceof Error) {
+      const status = void 0;
+      const retry = false;
+      throw new ChatCompletionError(err.message, logRequestData, status, retry);
+    }
+    throw err;
+  }
+  const outputMessage = {
+    role: "assistant",
+    content
+  };
+  const tokensUsed = {
+    prompt: inputUsage,
+    completion: outputUsage,
+    total: inputUsage + outputUsage
+  };
+  const cost = costFn?.(props.model, tokensUsed) ?? void 0;
+  const responseData = {
+    ...logRequestData,
+    finishReason,
+    latency: performance.now() - startTime,
+    inputMessages,
+    outputMessage,
+    tokensUsed
+  };
+  logger.chatCompletionResponse("google", responseData);
+  span.setAttributes({
+    tokensUsed,
+    output: content,
+    cost,
+    finishReason
+  });
+}
+function cleanChatCompletionRequest3(chatCompletionRequest) {
+  const { contents, ...rest } = chatCompletionRequest;
+  return {
+    ...rest,
+    contents: contents.map((message) => {
+      if (message.role !== "user") {
+        return message;
+      }
+      return {
+        ...message,
+        parts: message.parts.map((part) => {
+          if ("text" in part) {
+            return part;
+          } else if ("inlineData" in part) {
+            return {
+              inlineData: {
+                ...part.inlineData,
+                mimeType: part.inlineData?.mimeType || "",
+                data: part.inlineData?.data?.slice(0, 22) + "..."
+              }
+            };
+          }
+          return part;
+        })
+      };
+    })
+  };
+}
+
+// src/lib/google/index.ts
+var import_vertexai2 = require("@google-cloud/vertexai");
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   AIFragment,
@@ -2542,6 +2798,8 @@ var import_sdk2 = __toESM(require("@anthropic-ai/sdk"));
   DefaultMaxRetriesContext,
   EnrichingSpanProcessor,
   Fallback,
+  GoogleChatCompletion,
+  GoogleClientContext,
   ImagePart,
   LogImplementation,
   NoopLogImplementation,
@@ -2555,6 +2813,7 @@ var import_sdk2 = __toESM(require("@anthropic-ai/sdk"));
   SystemMessage,
   Trace,
   UserMessage,
+  VertexAI,
   anthropicTokenizer,
   attachedContextSymbol,
   computeUsage,

package/dist/index.mjs

@@ -8,11 +8,11 @@ import {
 
 // src/chat/errors.ts
 var ChatCompletionError = class extends Error {
-  constructor(message, chatCompletionRequest, status, shouldRetry3 = false) {
+  constructor(message, chatCompletionRequest, status, shouldRetry4 = false) {
     super(message);
     this.chatCompletionRequest = chatCompletionRequest;
     this.status = status;
-    this.shouldRetry = shouldRetry3;
+    this.shouldRetry = shouldRetry4;
   }
   name = "ChatCompletionError";
 };
@@ -1501,7 +1501,7 @@ function renderCloseTag(element) {
 // src/retry.tsx
 var RetryCountContext = createContext(0);
 var DefaultMaxRetriesContext = createContext(0);
-async function* Retry({ shouldRetry: shouldRetry3, retries = 0, maxRetries = 3, children }, ctx) {
+async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, children }, ctx) {
   const { render } = ctx;
   let hasYieldedData = false;
   try {
@@ -1513,12 +1513,12 @@ async function* Retry({ shouldRetry: shouldRetry3, retries = 0, maxRetries = 3,
       yield value;
     }
   } catch (e) {
-    if (hasYieldedData || retries >= maxRetries || !shouldRetry3(e)) {
+    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(e)) {
       throw e;
     }
     await backoff(retries);
     yield* Retry(
-      { shouldRetry: shouldRetry3, retries: retries + 1, maxRetries, children },
+      { shouldRetry: shouldRetry4, retries: retries + 1, maxRetries, children },
       ctx
     );
   }
@@ -2429,6 +2429,262 @@ function cleanChatCompletionRequest2(chatCompletionRequest) {
 
 // src/lib/anthropic/index.ts
 import AnthropicClient2 from "@anthropic-ai/sdk";
+
+// src/lib/google/Google.tsx
+import {
+  VertexAI,
+  GoogleGenerativeAIError
+} from "@google-cloud/vertexai";
+var GoogleClientContext = createContext(() => {
+  if (defaultClient3) {
+    return defaultClient3;
+  }
+  const project = process.env.GOOGLE_PROJECT_ID || "";
+  const client = new VertexAI({ project });
+  defaultClient3 = {
+    client,
+    provider: "google"
+  };
+  return defaultClient3;
+});
+var defaultClient3 = null;
+var buildGoogleMessages = (chatMesssages) => {
+  let systemInstruction;
+  const messages = [];
+  chatMesssages.forEach(({ role, content }) => {
+    if (role === "system") {
+      systemInstruction = content;
+      return;
+    }
+    if (role === "user") {
+      const userContent = content;
+      if (userContent.length === 1 && userContent[0].type === "text") {
+        messages.push({
+          role,
+          parts: [{ text: userContent[0].text }]
+        });
+        return;
+      }
+      const c = userContent.map((part) => {
+        if (part.type === "text") {
+          return { text: part.text };
+        } else if (part.type === "image") {
+          const imagePart = {
+            inlineData: {
+              mimeType: part.image.mediaType,
+              data: part.image.data
+            }
+          };
+          return imagePart;
+        }
+        throw new Error("Invalid part");
+      });
+      messages.push({
+        role,
+        parts: c
+      });
+      return;
+    }
+    if (role === "assistant") {
+      messages.push({
+        role,
+        parts: [{ text: content }]
+      });
+      return;
+    }
+    throw new Error(`Invalid role: ${role}`);
+  });
+  return {
+    systemInstruction,
+    messages
+  };
+};
+var shouldRetry3 = (error) => {
+  return error instanceof ChatCompletionError && error.shouldRetry;
+};
+function GoogleChatCompletion(props, ctx) {
+  const defaultMaxRetries = ctx.getContext(DefaultMaxRetriesContext);
+  return /* @__PURE__ */ jsx(
+    Retry,
+    {
+      maxRetries: props.maxRetries || defaultMaxRetries,
+      shouldRetry: shouldRetry3,
+      children: /* @__PURE__ */ jsx(Trace, { name: "ai.chatCompletion", children: /* @__PURE__ */ jsx(GoogleChatCompletionInner, { ...props }) })
+    }
+  );
+}
+async function* GoogleChatCompletionInner(props, ctx) {
+  const startTime = performance.now();
+  const { logger, tracer, getContext } = ctx;
+  const retryCount = getContext(RetryCountContext);
+  const span = tracer.getActiveSpan();
+  const { client, provider, providerRegion, costFn } = getContext(GoogleClientContext)();
+  if (!client) {
+    throw new Error(
+      "[GoogleChatCompletion] must supply GoogleClient via context"
+    );
+  }
+  const chatMessages = await buildChatMessages(ctx, props.children, {
+    useBase64Images: true
+  });
+  const { systemInstruction, messages } = buildGoogleMessages(chatMessages);
+  const inputMessages = chatMessages.map((m) => toDebugMessage(m));
+  let stopSequences;
+  if (props.stop && typeof props.stop === "string") {
+    stopSequences = [props.stop];
+  } else if (Array.isArray(props.stop)) {
+    stopSequences = props.stop;
+  }
+  const googleCompletionRequest = {
+    systemInstruction,
+    contents: messages,
+    generationConfig: {
+      maxOutputTokens: props.maxTokens,
+      temperature: props.temperature,
+      stopSequences
+    }
+  };
+  const chatCompletionRequestToLog = cleanChatCompletionRequest3(
+    googleCompletionRequest
+  );
+  const logRequestData = {
+    startTime,
+    model: props.model,
+    provider,
+    providerRegion,
+    inputMessages,
+    request: chatCompletionRequestToLog
+  };
+  logger.chatCompletionRequest("google", logRequestData);
+  span.setAttributes({
+    model: props.model,
+    provider,
+    providerRegion,
+    requestType: "google",
+    chatCompletionRequest: chatCompletionRequestToLog,
+    inputMessages,
+    retryCount
+  });
+  const model = client.getGenerativeModel({ model: props.model });
+  let response;
+  try {
+    response = await model.generateContentStream(googleCompletionRequest);
+  } catch (err) {
+    if (err instanceof GoogleGenerativeAIError) {
+      throw new ChatCompletionError(
+        err.message,
+        logRequestData,
+        void 0,
+        // always retry on GoogleGenerativeAIError errors, they represent non 4xx error cases
+        true
+      );
+    } else if (err instanceof Error) {
+      const status = void 0;
+      const retry = false;
+      throw new ChatCompletionError(err.message, logRequestData, status, retry);
+    }
+    throw err;
+  }
+  let content = "";
+  let outputUsage = 0;
+  let inputUsage = 0;
+  let finishReason = null;
+  try {
+    for await (const event of response.stream) {
+      if (event.candidates) {
+        const chunk = event.candidates[0].content;
+        if (event.candidates[0]?.finishReason) {
+          finishReason = event.candidates[0].finishReason;
+        }
+        if (event.usageMetadata) {
+          if (event.usageMetadata.promptTokenCount) {
+            inputUsage = event.usageMetadata.promptTokenCount;
+          }
+          if (event.usageMetadata.candidatesTokenCount) {
+            outputUsage = event.usageMetadata.candidatesTokenCount;
+          }
+        }
+        for (const block of chunk.parts) {
+          if (block.text) {
+            content += block.text;
+            yield block.text;
+          }
+        }
+      }
+    }
+  } catch (err) {
+    if (err instanceof GoogleGenerativeAIError) {
+      throw new ChatCompletionError(
+        err.message,
+        logRequestData,
+        void 0,
+        // always retry these errors they represent non 4xx error cases
+        true
+      );
+    } else if (err instanceof Error) {
+      const status = void 0;
+      const retry = false;
+      throw new ChatCompletionError(err.message, logRequestData, status, retry);
+    }
+    throw err;
+  }
+  const outputMessage = {
+    role: "assistant",
+    content
+  };
+  const tokensUsed = {
+    prompt: inputUsage,
+    completion: outputUsage,
+    total: inputUsage + outputUsage
+  };
+  const cost = costFn?.(props.model, tokensUsed) ?? void 0;
+  const responseData = {
+    ...logRequestData,
+    finishReason,
+    latency: performance.now() - startTime,
+    inputMessages,
+    outputMessage,
+    tokensUsed
+  };
+  logger.chatCompletionResponse("google", responseData);
+  span.setAttributes({
+    tokensUsed,
+    output: content,
+    cost,
+    finishReason
+  });
+}
+function cleanChatCompletionRequest3(chatCompletionRequest) {
+  const { contents, ...rest } = chatCompletionRequest;
+  return {
+    ...rest,
+    contents: contents.map((message) => {
+      if (message.role !== "user") {
+        return message;
+      }
+      return {
+        ...message,
+        parts: message.parts.map((part) => {
+          if ("text" in part) {
+            return part;
+          } else if ("inlineData" in part) {
+            return {
+              inlineData: {
+                ...part.inlineData,
+                mimeType: part.inlineData?.mimeType || "",
+                data: part.inlineData?.data?.slice(0, 22) + "..."
+              }
+            };
+          }
+          return part;
+        })
+      };
+    })
+  };
+}
+
+// src/lib/google/index.ts
+import { VertexAI as VertexAI2 } from "@google-cloud/vertexai";
 export {
   AIFragment,
   AISpanProcessor,
@@ -2443,6 +2699,8 @@ export {
   DefaultMaxRetriesContext,
   EnrichingSpanProcessor,
   Fallback,
+  GoogleChatCompletion,
+  GoogleClientContext,
   ImagePart,
   LogImplementation,
   NoopLogImplementation,
@@ -2456,6 +2714,7 @@ export {
   SystemMessage,
   Trace,
   UserMessage,
+  VertexAI2 as VertexAI,
   anthropicTokenizer,
   attachedContextSymbol,
   computeUsage,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gammatech/aijsx",
-  "version": "0.9.0-dev.2024-05-28",
+  "version": "0.9.1-dev.2024-05-28",
   "description": "Rewrite of aijsx",
   "author": "Jordan Garcia",
   "license": "MIT",
@@ -20,7 +20,7 @@
   "dependencies": {
     "@anthropic-ai/sdk": "0.19.1",
     "@anthropic-ai/tokenizer": "^0.0.4",
-    "exif-reader": "^2.0.1",
+    "@google-cloud/vertexai": "^1.2.0",
     "fast-xml-parser": "^4.3.4",
     "js-tiktoken": "^1.0.8",
     "nanoid": "^3.1.23",