@gammatech/aijsx 0.10.2-dev.2024-06-11 → 0.11.1-dev.2024-06-23

package/dist/index.d.mts

@@ -13,8 +13,9 @@ declare class ChatCompletionError extends Error {
     readonly chatCompletionRequest: LogChatCompletionRequest;
     readonly status: number | undefined;
     readonly shouldRetry: boolean;
+    readonly originalError?: Error | undefined;
     readonly name = "ChatCompletionError";
-    constructor(message: string, chatCompletionRequest: LogChatCompletionRequest, status: number | undefined, shouldRetry?: boolean);
+    constructor(message: string, chatCompletionRequest: LogChatCompletionRequest, status: number | undefined, shouldRetry?: boolean, originalError?: Error | undefined);
 }
 
 declare const SystemMessage: (props: {
@@ -43,6 +44,10 @@ type ChatCompletionClientAndProvider<K> = {
         completion: number;
     }) => number;
 };
+type GetChatCompletionClientAndProvider<Model, Client> = (model: Model, args: {
+    retryCount: number;
+    lastError?: Error | null;
+}) => Promise<ChatCompletionClientAndProvider<Client>>;
 
 type CreateRenderContextOptions = {
     logger?: LogImplementation;
@@ -52,15 +57,25 @@ type CreateRenderContextOptions = {
 };
 declare function createRenderContext({ logger, traceId, processor, contextValues, }?: CreateRenderContextOptions): RenderContext;
 
-declare const RetryCountContext: Context<number>;
+type ExcludeNumber<E extends number, T = number> = T extends E ? never : T;
+type RetryCountContextValue = {
+    retryCount: 0;
+    lastError: null;
+} | {
+    retryCount: ExcludeNumber<0>;
+    lastError: Error;
+};
+declare const RetryCountContext: Context<RetryCountContextValue>;
+declare const RetryLastErrorContext: Context<Error | null>;
 declare const DefaultMaxRetriesContext: Context<number>;
 type RetryProps = {
     shouldRetry: (error: Error) => boolean;
     retries?: number;
+    lastError?: Error;
     maxRetries?: number;
     children: AINode;
 };
-declare function Retry({ shouldRetry, retries, maxRetries, children }: RetryProps, ctx: RenderContext): AsyncGenerator<string, void, unknown>;
+declare function Retry({ shouldRetry, retries, maxRetries, lastError, children }: RetryProps, ctx: RenderContext): AsyncGenerator<string, void, unknown>;
 
 type FallbackProps = {
     fallback: AINode;
@@ -241,7 +256,7 @@ declare module '@gammatech/aijsx' {
 }
 type ValidOpenAIVisionModel = 'gpt-4o' | 'gpt-4o-2024-05-13' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo' | 'gpt-4-vision-preview';
 type ValidOpenAIChatModel = ValidOpenAIVisionModel | 'gpt-4' | 'gpt-4-0314' | 'gpt-4-0613' | 'gpt-4-32k' | 'gpt-4-32k-0314' | 'gpt-4-32k-0613' | 'gpt-4-1106-preview' | 'gpt-4-0125-preview' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-0301' | 'gpt-3.5-turbo-0613' | 'gpt-3.5-turbo-16k' | 'gpt-3.5-turbo-16k-0613' | 'gpt-3.5-turbo-1106' | 'gpt-3.5-turbo-0125';
-declare const OpenAIClientContext: Context<() => ChatCompletionClientAndProvider<OpenAI>>;
+declare const OpenAIClientContext: Context<GetChatCompletionClientAndProvider<ValidOpenAIChatModel, OpenAI>>;
 type OpenAIChatCompletionProps = {
     model: ValidOpenAIChatModel;
     maxTokens?: number;
@@ -265,8 +280,8 @@ declare module '@gammatech/aijsx' {
  * The set of valid Claude models.
  * @see https://docs.anthropic.com/claude/reference/selecting-a-model
  */
-type ValidAnthropicChatModel = 'claude-instant-1.2' | 'claude-2.1' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307';
-declare const AnthropicClientContext: Context<() => ChatCompletionClientAndProvider<AnthropicClient>>;
+type ValidAnthropicChatModel = 'claude-instant-1.2' | 'claude-2.1' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307' | 'claude-3-5-sonnet-20240620';
+declare const AnthropicClientContext: Context<GetChatCompletionClientAndProvider<ValidAnthropicChatModel, AnthropicClient>>;
 type AnthropicChatCompletionProps = {
     model: ValidAnthropicChatModel;
     maxTokens?: number;
@@ -291,7 +306,7 @@ declare module '@gammatech/aijsx' {
     }
 }
 type ValidGoogleChatModel = 'gemini-1.5-pro' | 'gemini-1.5-flash';
-declare const GoogleClientContext: Context<() => ChatCompletionClientAndProvider<VertexAI>>;
+declare const GoogleClientContext: Context<GetChatCompletionClientAndProvider<ValidGoogleChatModel, VertexAI>>;
 type GoogleChatCompletionProps = {
     model: ValidGoogleChatModel;
     maxTokens?: number;
@@ -306,4 +321,4 @@ type GoogleChatCompletionProps = {
 };
 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, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createPrompt, createRenderContext, evaluatePrompt, isPromptParsed, openaiTokenizer, tracing };
+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, type GetChatCompletionClientAndProvider, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, RetryLastErrorContext, SpanAttributes, SpanExporter, SpanProcessor, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createPrompt, createRenderContext, evaluatePrompt, isPromptParsed, openaiTokenizer, tracing };

package/dist/index.d.ts

@@ -13,8 +13,9 @@ declare class ChatCompletionError extends Error {
     readonly chatCompletionRequest: LogChatCompletionRequest;
     readonly status: number | undefined;
     readonly shouldRetry: boolean;
+    readonly originalError?: Error | undefined;
     readonly name = "ChatCompletionError";
-    constructor(message: string, chatCompletionRequest: LogChatCompletionRequest, status: number | undefined, shouldRetry?: boolean);
+    constructor(message: string, chatCompletionRequest: LogChatCompletionRequest, status: number | undefined, shouldRetry?: boolean, originalError?: Error | undefined);
 }
 
 declare const SystemMessage: (props: {
@@ -43,6 +44,10 @@ type ChatCompletionClientAndProvider<K> = {
         completion: number;
     }) => number;
 };
+type GetChatCompletionClientAndProvider<Model, Client> = (model: Model, args: {
+    retryCount: number;
+    lastError?: Error | null;
+}) => Promise<ChatCompletionClientAndProvider<Client>>;
 
 type CreateRenderContextOptions = {
     logger?: LogImplementation;
@@ -52,15 +57,25 @@ type CreateRenderContextOptions = {
 };
 declare function createRenderContext({ logger, traceId, processor, contextValues, }?: CreateRenderContextOptions): RenderContext;
 
-declare const RetryCountContext: Context<number>;
+type ExcludeNumber<E extends number, T = number> = T extends E ? never : T;
+type RetryCountContextValue = {
+    retryCount: 0;
+    lastError: null;
+} | {
+    retryCount: ExcludeNumber<0>;
+    lastError: Error;
+};
+declare const RetryCountContext: Context<RetryCountContextValue>;
+declare const RetryLastErrorContext: Context<Error | null>;
 declare const DefaultMaxRetriesContext: Context<number>;
 type RetryProps = {
     shouldRetry: (error: Error) => boolean;
     retries?: number;
+    lastError?: Error;
     maxRetries?: number;
     children: AINode;
 };
-declare function Retry({ shouldRetry, retries, maxRetries, children }: RetryProps, ctx: RenderContext): AsyncGenerator<string, void, unknown>;
+declare function Retry({ shouldRetry, retries, maxRetries, lastError, children }: RetryProps, ctx: RenderContext): AsyncGenerator<string, void, unknown>;
 
 type FallbackProps = {
     fallback: AINode;
@@ -241,7 +256,7 @@ declare module '@gammatech/aijsx' {
 }
 type ValidOpenAIVisionModel = 'gpt-4o' | 'gpt-4o-2024-05-13' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo' | 'gpt-4-vision-preview';
 type ValidOpenAIChatModel = ValidOpenAIVisionModel | 'gpt-4' | 'gpt-4-0314' | 'gpt-4-0613' | 'gpt-4-32k' | 'gpt-4-32k-0314' | 'gpt-4-32k-0613' | 'gpt-4-1106-preview' | 'gpt-4-0125-preview' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-0301' | 'gpt-3.5-turbo-0613' | 'gpt-3.5-turbo-16k' | 'gpt-3.5-turbo-16k-0613' | 'gpt-3.5-turbo-1106' | 'gpt-3.5-turbo-0125';
-declare const OpenAIClientContext: Context<() => ChatCompletionClientAndProvider<OpenAI>>;
+declare const OpenAIClientContext: Context<GetChatCompletionClientAndProvider<ValidOpenAIChatModel, OpenAI>>;
 type OpenAIChatCompletionProps = {
     model: ValidOpenAIChatModel;
     maxTokens?: number;
@@ -265,8 +280,8 @@ declare module '@gammatech/aijsx' {
  * The set of valid Claude models.
  * @see https://docs.anthropic.com/claude/reference/selecting-a-model
  */
-type ValidAnthropicChatModel = 'claude-instant-1.2' | 'claude-2.1' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307';
-declare const AnthropicClientContext: Context<() => ChatCompletionClientAndProvider<AnthropicClient>>;
+type ValidAnthropicChatModel = 'claude-instant-1.2' | 'claude-2.1' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307' | 'claude-3-5-sonnet-20240620';
+declare const AnthropicClientContext: Context<GetChatCompletionClientAndProvider<ValidAnthropicChatModel, AnthropicClient>>;
 type AnthropicChatCompletionProps = {
     model: ValidAnthropicChatModel;
     maxTokens?: number;
@@ -291,7 +306,7 @@ declare module '@gammatech/aijsx' {
     }
 }
 type ValidGoogleChatModel = 'gemini-1.5-pro' | 'gemini-1.5-flash';
-declare const GoogleClientContext: Context<() => ChatCompletionClientAndProvider<VertexAI>>;
+declare const GoogleClientContext: Context<GetChatCompletionClientAndProvider<ValidGoogleChatModel, VertexAI>>;
 type GoogleChatCompletionProps = {
     model: ValidGoogleChatModel;
     maxTokens?: number;
@@ -306,4 +321,4 @@ type GoogleChatCompletionProps = {
 };
 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, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, SpanAttributes, SpanExporter, SpanProcessor, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createPrompt, createRenderContext, evaluatePrompt, isPromptParsed, openaiTokenizer, tracing };
+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, type GetChatCompletionClientAndProvider, GoogleChatCompletion, type GoogleChatCompletionRequest, GoogleClientContext, LogChatCompletionRequest, LogImplementation, OpenAIChatCompletion, type OpenAIChatCompletionRequest, type OpenAIChatMessage, OpenAIClientContext, ParseVariablesError, ProcessedAISpanAttributes, type Prompt, PromptInvalidOutputError, PromptParsed, ReadableSpan, RenderContext, Retry, RetryCountContext, RetryLastErrorContext, SpanAttributes, SpanExporter, SpanProcessor, SystemMessage, type TokenizerFn, Trace, Tracer, UserMessage, type ValidAnthropicChatModel, type ValidGoogleChatModel, type ValidOpenAIChatModel, type ValidOpenAIVisionModel, anthropicTokenizer, computeUsage, createPrompt, createRenderContext, evaluatePrompt, isPromptParsed, openaiTokenizer, tracing };

package/dist/index.js

@@ -56,6 +56,7 @@ __export(src_exports, {
   PromptInvalidOutputError: () => PromptInvalidOutputError,
   Retry: () => Retry,
   RetryCountContext: () => RetryCountContext,
+  RetryLastErrorContext: () => RetryLastErrorContext,
   SystemMessage: () => SystemMessage,
   Trace: () => Trace,
   UserMessage: () => UserMessage,
@@ -77,11 +78,12 @@ module.exports = __toCommonJS(src_exports);
 
 // src/chat/errors.ts
 var ChatCompletionError = class extends Error {
-  constructor(message, chatCompletionRequest, status, shouldRetry4 = false) {
+  constructor(message, chatCompletionRequest, status, shouldRetry4 = false, originalError) {
     super(message);
     this.chatCompletionRequest = chatCompletionRequest;
     this.status = status;
     this.shouldRetry = shouldRetry4;
+    this.originalError = originalError;
   }
   name = "ChatCompletionError";
 };
@@ -1537,27 +1539,67 @@ function renderCloseTag(element) {
   return `</${element.tag.name}>`;
 }
 
+// src/utils.ts
+function getEnvVar(name, shouldThrow = true) {
+  let env = globalThis.process?.env ?? void 0;
+  if (env === void 0) {
+    try {
+      env = process.env;
+    } catch {
+    }
+  }
+  const result = env?.[name];
+  if (result === void 0 && shouldThrow) {
+    throw new Error(`Please specify env var '${name}'`);
+  }
+  return result;
+}
+var castToError = (e) => {
+  if (e instanceof Error) {
+    return e;
+  }
+  if (typeof e === "string") {
+    return new Error(e);
+  }
+  return new Error("Unknown error");
+};
+
 // src/retry.tsx
-var RetryCountContext = createContext(0);
+var RetryCountContext = createContext({
+  retryCount: 0,
+  lastError: null
+});
+var RetryLastErrorContext = createContext(null);
 var DefaultMaxRetriesContext = createContext(0);
-async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, children }, ctx) {
+async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, lastError, children }, ctx) {
   const { render } = ctx;
   let hasYieldedData = false;
   try {
+    const ctxValue = retries === 0 ? { retryCount: 0, lastError: null } : {
+      retryCount: retries,
+      lastError
+    };
     const result = render(
-      /* @__PURE__ */ jsx(RetryCountContext.Provider, { value: retries, children })
+      /* @__PURE__ */ jsx(RetryCountContext.Provider, { value: ctxValue, children })
     );
     for await (const value of result) {
       hasYieldedData = true;
       yield value;
     }
   } catch (e) {
-    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(e)) {
+    const err = castToError(e);
+    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(err)) {
       throw e;
     }
     await backoff(retries);
     yield* Retry(
-      { shouldRetry: shouldRetry4, retries: retries + 1, maxRetries, children },
+      {
+        shouldRetry: shouldRetry4,
+        retries: retries + 1,
+        maxRetries,
+        lastError: err,
+        children
+      },
       ctx
     );
   }
@@ -1903,25 +1945,6 @@ function isPromptParsed2(prompt) {
 // src/lib/openai/OpenAI.tsx
 var import_openai2 = require("openai");
 
-// src/lib/openai/shouldRetryOpenAI.ts
-var import_openai = require("openai");
-var shouldRetryOpenAI = (error) => {
-  if (error instanceof import_openai.OpenAI.APIConnectionError) {
-    return true;
-  }
-  if (error instanceof import_openai.OpenAI.APIError) {
-    if ("status" in error && typeof error.status === "number") {
-      if (error.status === 409)
-        return true;
-      if (error.status === 429)
-        return true;
-      if (error.status >= 500)
-        return true;
-    }
-  }
-  return false;
-};
-
 // src/lib/openai/tokenizer.ts
 var import_js_tiktoken = require("js-tiktoken");
 var cl100kTokenizer = (0, import_js_tiktoken.getEncoding)("cl100k_base");
@@ -2033,25 +2056,39 @@ async function buildChatMessages(ctx, children, opts) {
   });
 }
 
-// src/utils.ts
-function getEnvVar(name, shouldThrow = true) {
-  let env = globalThis.process?.env ?? void 0;
-  if (env === void 0) {
-    try {
-      env = process.env;
-    } catch {
-    }
+// src/lib/openai/errors.ts
+var import_openai = require("openai");
+var extractStatusFromError = (error) => {
+  if (error instanceof import_openai.OpenAI.APIError) {
+    return error.status;
+  } else if (error instanceof import_openai.OpenAI.APIConnectionError) {
+    return void 0;
+  } else {
+    return void 0;
   }
-  const result = env?.[name];
-  if (result === void 0 && shouldThrow) {
-    throw new Error(`Please specify env var '${name}'`);
+};
+var errorToChatCompletionError = (error, requestData) => {
+  const castError = castToError(error);
+  const status = extractStatusFromError(castError);
+  let messagePrefix = "";
+  if (error instanceof import_openai.OpenAI.APIError) {
+    messagePrefix = "OpenAIClient.APIError: ";
+  } else if (error instanceof import_openai.OpenAI.APIConnectionError) {
+    messagePrefix = "OpenAIClient.APIConnectionError: ";
   }
-  return result;
-}
+  const shouldRetry4 = status !== 400;
+  return new ChatCompletionError(
+    `${messagePrefix}${castError.message}`,
+    requestData,
+    status,
+    shouldRetry4,
+    error instanceof Error ? error : void 0
+  );
+};
 
 // src/lib/openai/OpenAI.tsx
 var defaultClient = null;
-var OpenAIClientContext = createContext(() => {
+var OpenAIClientContext = createContext(async () => {
   if (defaultClient) {
     return defaultClient;
   }
@@ -2123,9 +2160,16 @@ function OpenAIChatCompletion(props, ctx) {
 async function* OpenAIChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(OpenAIClientContext)();
+  const getClientFn = getContext(OpenAIClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error("[OpenAI] must supply OpenAI model via context");
   }
@@ -2170,38 +2214,32 @@ async function* OpenAIChatCompletionInner(props, ctx) {
   try {
     chatResponse = await client.chat.completions.create(chatCompletionRequest);
   } catch (ex) {
-    const retry = shouldRetryOpenAI(ex);
-    if (ex instanceof import_openai2.OpenAI.APIError) {
-      throw new ChatCompletionError(
-        `OpenAIClient.APIError: ${ex.message}`,
-        logRequestData,
-        ex.status,
-        retry
-      );
-    } else if (ex instanceof Error) {
-      throw new ChatCompletionError(
-        ex.message,
-        logRequestData,
-        void 0,
-        retry
-      );
-    }
-    throw ex;
+    throw errorToChatCompletionError(ex, logRequestData);
   }
   let finishReason = void 0;
   let content = "";
-  for await (const message of chatResponse) {
-    if (!message.choices || !message.choices[0]) {
-      continue;
-    }
-    const delta = message.choices[0].delta;
-    if (message.choices[0].finish_reason) {
-      finishReason = message.choices[0].finish_reason;
-    }
-    if (delta.content) {
-      content += delta.content;
-      yield delta.content;
+  try {
+    for await (const message of chatResponse) {
+      if (!message.choices || !message.choices[0]) {
+        continue;
+      }
+      const delta = message.choices[0].delta;
+      if (message.choices[0].finish_reason) {
+        finishReason = message.choices[0].finish_reason;
+        span.setAttributes({
+          finishReason
+        });
+      }
+      if (delta.content) {
+        content += delta.content;
+        yield delta.content;
+      }
     }
+  } catch (e) {
+    span.setAttributes({
+      output: content
+    });
+    throw errorToChatCompletionError(e, logRequestData);
   }
   const outputMessage = {
     role: "assistant",
@@ -2295,7 +2333,7 @@ var anthropicTokenizer = (message) => {
 
 // src/lib/anthropic/Anthropic.tsx
 var defaultClient2 = null;
-var AnthropicClientContext = createContext(() => {
+var AnthropicClientContext = createContext(async () => {
   if (defaultClient2) {
     return defaultClient2;
   }
@@ -2374,7 +2412,7 @@ var shouldRetry2 = (error) => {
 var shouldRetryFromStatus = (status) => Boolean(status && [424, 429, 500].includes(status));
 var RE_INTERNAL_SERVER_MESSASGE = /The system encountered an unexpected error during processing/i;
 var RE_RATE_LIMIT_MESSAGE = /Too many requests, please wait before trying again/;
-var extractStatusFromError = (error) => {
+var extractStatusFromError2 = (error) => {
   if (typeof error !== "object" || !(error instanceof Error)) {
     return;
   }
@@ -2413,11 +2451,16 @@ function AnthropicChatCompletion(props, ctx) {
 async function* AnthropicChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(
-    AnthropicClientContext
-  )();
+  const getClientFn = getContext(AnthropicClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error(
       "[AnthropicChatCompletion] must supply AnthropicClient via context"
@@ -2468,7 +2511,7 @@ async function* AnthropicChatCompletionInner(props, ctx) {
     response = client.messages.stream(anthropicCompletionRequest);
   } catch (err) {
     if (err instanceof import_sdk.default.APIError) {
-      const status = extractStatusFromError(err);
+      const status = extractStatusFromError2(err);
       const retry = shouldRetryFromStatus(status);
       throw new ChatCompletionError(
         `AnthropicClient.APIError: ${err.message}`,
@@ -2477,7 +2520,7 @@ async function* AnthropicChatCompletionInner(props, ctx) {
         retry
       );
     } else if (err instanceof Error) {
-      const status = extractStatusFromError(err);
+      const status = extractStatusFromError2(err);
       const retry = shouldRetryFromStatus(status);
       throw new ChatCompletionError(err.message, logRequestData, status, retry);
     }
@@ -2500,10 +2543,16 @@ async function* AnthropicChatCompletionInner(props, ctx) {
       if (event.type === "message_delta") {
         finishReason = event.delta.stop_reason;
         outputUsage = event.usage?.output_tokens;
+        span.setAttributes({
+          finishReason
+        });
       }
     }
   } catch (e) {
-    const status = extractStatusFromError(e);
+    span.setAttributes({
+      output: content
+    });
+    const status = extractStatusFromError2(e);
     const retry = shouldRetryFromStatus(status);
     throw new ChatCompletionError(e.message, logRequestData, status, retry);
   }
@@ -2590,14 +2639,15 @@ var extractStatusFromMessage = (message) => {
   }
   return 500;
 };
-var errorToChatCompletionError = (error, requestData) => {
+var errorToChatCompletionError2 = (error, requestData) => {
   const status = extractStatusFromMessage(error.message);
   const shouldRetry4 = status !== 400;
   return new ChatCompletionError(
     error.message,
     requestData,
     status,
-    shouldRetry4
+    shouldRetry4,
+    error
   );
 };
 
@@ -2624,7 +2674,7 @@ var DEFAULT_SAFETY_SETTINGS = [
     threshold: import_vertexai.HarmBlockThreshold.BLOCK_ONLY_HIGH
   }
 ];
-var GoogleClientContext = createContext(() => {
+var GoogleClientContext = createContext(async () => {
   if (defaultClient3) {
     return defaultClient3;
   }
@@ -2705,9 +2755,16 @@ function GoogleChatCompletion(props, ctx) {
 async function* GoogleChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(GoogleClientContext)();
+  const getClientFn = getContext(GoogleClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error(
       "[GoogleChatCompletion] must supply GoogleClient via context"
@@ -2762,7 +2819,7 @@ async function* GoogleChatCompletionInner(props, ctx) {
   try {
     response = await model.generateContentStream(googleCompletionRequest);
   } catch (err) {
-    throw errorToChatCompletionError(err, logRequestData);
+    throw errorToChatCompletionError2(err, logRequestData);
   }
   let content = "";
   let outputUsage = 0;
@@ -2773,6 +2830,9 @@ async function* GoogleChatCompletionInner(props, ctx) {
       if (event.candidates) {
         if (event.candidates[0]?.finishReason) {
           finishReason = event.candidates[0].finishReason;
+          span.setAttributes({
+            finishReason
+          });
         }
         if (event.usageMetadata) {
           if (event.usageMetadata.promptTokenCount) {
@@ -2808,7 +2868,10 @@ async function* GoogleChatCompletionInner(props, ctx) {
       }
     }
   } catch (err) {
-    throw errorToChatCompletionError(err, logRequestData);
+    span.setAttributes({
+      output: content
+    });
+    throw errorToChatCompletionError2(err, logRequestData);
   }
   const outputMessage = {
     role: "assistant",
@@ -2832,8 +2895,7 @@ async function* GoogleChatCompletionInner(props, ctx) {
   span.setAttributes({
     tokensUsed,
     output: content,
-    cost,
-    finishReason
+    cost
   });
 }
 function cleanChatCompletionRequest3(chatCompletionRequest) {
@@ -2896,6 +2958,7 @@ var import_vertexai2 = require("@google-cloud/vertexai");
   PromptInvalidOutputError,
   Retry,
   RetryCountContext,
+  RetryLastErrorContext,
   SystemMessage,
   Trace,
   UserMessage,

package/dist/index.mjs

@@ -8,11 +8,12 @@ import {
 
 // src/chat/errors.ts
 var ChatCompletionError = class extends Error {
-  constructor(message, chatCompletionRequest, status, shouldRetry4 = false) {
+  constructor(message, chatCompletionRequest, status, shouldRetry4 = false, originalError) {
     super(message);
     this.chatCompletionRequest = chatCompletionRequest;
     this.status = status;
     this.shouldRetry = shouldRetry4;
+    this.originalError = originalError;
   }
   name = "ChatCompletionError";
 };
@@ -1435,27 +1436,67 @@ function renderCloseTag(element) {
   return `</${element.tag.name}>`;
 }
 
+// src/utils.ts
+function getEnvVar(name, shouldThrow = true) {
+  let env = globalThis.process?.env ?? void 0;
+  if (env === void 0) {
+    try {
+      env = process.env;
+    } catch {
+    }
+  }
+  const result = env?.[name];
+  if (result === void 0 && shouldThrow) {
+    throw new Error(`Please specify env var '${name}'`);
+  }
+  return result;
+}
+var castToError = (e) => {
+  if (e instanceof Error) {
+    return e;
+  }
+  if (typeof e === "string") {
+    return new Error(e);
+  }
+  return new Error("Unknown error");
+};
+
 // src/retry.tsx
-var RetryCountContext = createContext(0);
+var RetryCountContext = createContext({
+  retryCount: 0,
+  lastError: null
+});
+var RetryLastErrorContext = createContext(null);
 var DefaultMaxRetriesContext = createContext(0);
-async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, children }, ctx) {
+async function* Retry({ shouldRetry: shouldRetry4, retries = 0, maxRetries = 3, lastError, children }, ctx) {
   const { render } = ctx;
   let hasYieldedData = false;
   try {
+    const ctxValue = retries === 0 ? { retryCount: 0, lastError: null } : {
+      retryCount: retries,
+      lastError
+    };
     const result = render(
-      /* @__PURE__ */ jsx(RetryCountContext.Provider, { value: retries, children })
+      /* @__PURE__ */ jsx(RetryCountContext.Provider, { value: ctxValue, children })
     );
     for await (const value of result) {
       hasYieldedData = true;
       yield value;
     }
   } catch (e) {
-    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(e)) {
+    const err = castToError(e);
+    if (hasYieldedData || retries >= maxRetries || !shouldRetry4(err)) {
       throw e;
     }
     await backoff(retries);
     yield* Retry(
-      { shouldRetry: shouldRetry4, retries: retries + 1, maxRetries, children },
+      {
+        shouldRetry: shouldRetry4,
+        retries: retries + 1,
+        maxRetries,
+        lastError: err,
+        children
+      },
       ctx
     );
   }
@@ -1801,25 +1842,6 @@ function isPromptParsed2(prompt) {
 // src/lib/openai/OpenAI.tsx
 import { OpenAI as OpenAIClient2 } from "openai";
 
-// src/lib/openai/shouldRetryOpenAI.ts
-import { OpenAI as OpenAIClient } from "openai";
-var shouldRetryOpenAI = (error) => {
-  if (error instanceof OpenAIClient.APIConnectionError) {
-    return true;
-  }
-  if (error instanceof OpenAIClient.APIError) {
-    if ("status" in error && typeof error.status === "number") {
-      if (error.status === 409)
-        return true;
-      if (error.status === 429)
-        return true;
-      if (error.status >= 500)
-        return true;
-    }
-  }
-  return false;
-};
-
 // src/lib/openai/tokenizer.ts
 import { getEncoding } from "js-tiktoken";
 var cl100kTokenizer = getEncoding("cl100k_base");
@@ -1931,25 +1953,39 @@ async function buildChatMessages(ctx, children, opts) {
   });
 }
 
-// src/utils.ts
-function getEnvVar(name, shouldThrow = true) {
-  let env = globalThis.process?.env ?? void 0;
-  if (env === void 0) {
-    try {
-      env = process.env;
-    } catch {
-    }
+// src/lib/openai/errors.ts
+import { OpenAI as OpenAIClient } from "openai";
+var extractStatusFromError = (error) => {
+  if (error instanceof OpenAIClient.APIError) {
+    return error.status;
+  } else if (error instanceof OpenAIClient.APIConnectionError) {
+    return void 0;
+  } else {
+    return void 0;
   }
-  const result = env?.[name];
-  if (result === void 0 && shouldThrow) {
-    throw new Error(`Please specify env var '${name}'`);
+};
+var errorToChatCompletionError = (error, requestData) => {
+  const castError = castToError(error);
+  const status = extractStatusFromError(castError);
+  let messagePrefix = "";
+  if (error instanceof OpenAIClient.APIError) {
+    messagePrefix = "OpenAIClient.APIError: ";
+  } else if (error instanceof OpenAIClient.APIConnectionError) {
+    messagePrefix = "OpenAIClient.APIConnectionError: ";
   }
-  return result;
-}
+  const shouldRetry4 = status !== 400;
+  return new ChatCompletionError(
+    `${messagePrefix}${castError.message}`,
+    requestData,
+    status,
+    shouldRetry4,
+    error instanceof Error ? error : void 0
+  );
+};
 
 // src/lib/openai/OpenAI.tsx
 var defaultClient = null;
-var OpenAIClientContext = createContext(() => {
+var OpenAIClientContext = createContext(async () => {
   if (defaultClient) {
     return defaultClient;
   }
@@ -2021,9 +2057,16 @@ function OpenAIChatCompletion(props, ctx) {
 async function* OpenAIChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(OpenAIClientContext)();
+  const getClientFn = getContext(OpenAIClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error("[OpenAI] must supply OpenAI model via context");
   }
@@ -2068,38 +2111,32 @@ async function* OpenAIChatCompletionInner(props, ctx) {
   try {
     chatResponse = await client.chat.completions.create(chatCompletionRequest);
   } catch (ex) {
-    const retry = shouldRetryOpenAI(ex);
-    if (ex instanceof OpenAIClient2.APIError) {
-      throw new ChatCompletionError(
-        `OpenAIClient.APIError: ${ex.message}`,
-        logRequestData,
-        ex.status,
-        retry
-      );
-    } else if (ex instanceof Error) {
-      throw new ChatCompletionError(
-        ex.message,
-        logRequestData,
-        void 0,
-        retry
-      );
-    }
-    throw ex;
+    throw errorToChatCompletionError(ex, logRequestData);
   }
   let finishReason = void 0;
   let content = "";
-  for await (const message of chatResponse) {
-    if (!message.choices || !message.choices[0]) {
-      continue;
-    }
-    const delta = message.choices[0].delta;
-    if (message.choices[0].finish_reason) {
-      finishReason = message.choices[0].finish_reason;
-    }
-    if (delta.content) {
-      content += delta.content;
-      yield delta.content;
+  try {
+    for await (const message of chatResponse) {
+      if (!message.choices || !message.choices[0]) {
+        continue;
+      }
+      const delta = message.choices[0].delta;
+      if (message.choices[0].finish_reason) {
+        finishReason = message.choices[0].finish_reason;
+        span.setAttributes({
+          finishReason
+        });
+      }
+      if (delta.content) {
+        content += delta.content;
+        yield delta.content;
+      }
     }
+  } catch (e) {
+    span.setAttributes({
+      output: content
+    });
+    throw errorToChatCompletionError(e, logRequestData);
   }
   const outputMessage = {
     role: "assistant",
@@ -2193,7 +2230,7 @@ var anthropicTokenizer = (message) => {
 
 // src/lib/anthropic/Anthropic.tsx
 var defaultClient2 = null;
-var AnthropicClientContext = createContext(() => {
+var AnthropicClientContext = createContext(async () => {
   if (defaultClient2) {
     return defaultClient2;
   }
@@ -2272,7 +2309,7 @@ var shouldRetry2 = (error) => {
 var shouldRetryFromStatus = (status) => Boolean(status && [424, 429, 500].includes(status));
 var RE_INTERNAL_SERVER_MESSASGE = /The system encountered an unexpected error during processing/i;
 var RE_RATE_LIMIT_MESSAGE = /Too many requests, please wait before trying again/;
-var extractStatusFromError = (error) => {
+var extractStatusFromError2 = (error) => {
   if (typeof error !== "object" || !(error instanceof Error)) {
     return;
   }
@@ -2311,11 +2348,16 @@ function AnthropicChatCompletion(props, ctx) {
 async function* AnthropicChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(
-    AnthropicClientContext
-  )();
+  const getClientFn = getContext(AnthropicClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error(
       "[AnthropicChatCompletion] must supply AnthropicClient via context"
@@ -2366,7 +2408,7 @@ async function* AnthropicChatCompletionInner(props, ctx) {
     response = client.messages.stream(anthropicCompletionRequest);
   } catch (err) {
     if (err instanceof AnthropicClient.APIError) {
-      const status = extractStatusFromError(err);
+      const status = extractStatusFromError2(err);
       const retry = shouldRetryFromStatus(status);
       throw new ChatCompletionError(
         `AnthropicClient.APIError: ${err.message}`,
@@ -2375,7 +2417,7 @@ async function* AnthropicChatCompletionInner(props, ctx) {
         retry
       );
     } else if (err instanceof Error) {
-      const status = extractStatusFromError(err);
+      const status = extractStatusFromError2(err);
       const retry = shouldRetryFromStatus(status);
       throw new ChatCompletionError(err.message, logRequestData, status, retry);
     }
@@ -2398,10 +2440,16 @@ async function* AnthropicChatCompletionInner(props, ctx) {
       if (event.type === "message_delta") {
         finishReason = event.delta.stop_reason;
         outputUsage = event.usage?.output_tokens;
+        span.setAttributes({
+          finishReason
+        });
       }
     }
   } catch (e) {
-    const status = extractStatusFromError(e);
+    span.setAttributes({
+      output: content
+    });
+    const status = extractStatusFromError2(e);
     const retry = shouldRetryFromStatus(status);
     throw new ChatCompletionError(e.message, logRequestData, status, retry);
   }
@@ -2492,14 +2540,15 @@ var extractStatusFromMessage = (message) => {
   }
   return 500;
 };
-var errorToChatCompletionError = (error, requestData) => {
+var errorToChatCompletionError2 = (error, requestData) => {
   const status = extractStatusFromMessage(error.message);
   const shouldRetry4 = status !== 400;
   return new ChatCompletionError(
     error.message,
     requestData,
     status,
-    shouldRetry4
+    shouldRetry4,
+    error
   );
 };
 
@@ -2526,7 +2575,7 @@ var DEFAULT_SAFETY_SETTINGS = [
     threshold: HarmBlockThreshold.BLOCK_ONLY_HIGH
   }
 ];
-var GoogleClientContext = createContext(() => {
+var GoogleClientContext = createContext(async () => {
   if (defaultClient3) {
     return defaultClient3;
   }
@@ -2607,9 +2656,16 @@ function GoogleChatCompletion(props, ctx) {
 async function* GoogleChatCompletionInner(props, ctx) {
   const startTime = performance.now();
   const { logger, tracer, getContext } = ctx;
-  const retryCount = getContext(RetryCountContext);
+  const { retryCount, lastError } = getContext(RetryCountContext);
   const span = tracer.getActiveSpan();
-  const { client, provider, providerRegion, costFn } = getContext(GoogleClientContext)();
+  const getClientFn = getContext(GoogleClientContext);
+  const { client, provider, providerRegion, costFn } = await getClientFn(
+    props.model,
+    {
+      retryCount,
+      lastError
+    }
+  );
   if (!client) {
     throw new Error(
       "[GoogleChatCompletion] must supply GoogleClient via context"
@@ -2664,7 +2720,7 @@ async function* GoogleChatCompletionInner(props, ctx) {
   try {
     response = await model.generateContentStream(googleCompletionRequest);
   } catch (err) {
-    throw errorToChatCompletionError(err, logRequestData);
+    throw errorToChatCompletionError2(err, logRequestData);
   }
   let content = "";
   let outputUsage = 0;
@@ -2675,6 +2731,9 @@ async function* GoogleChatCompletionInner(props, ctx) {
       if (event.candidates) {
         if (event.candidates[0]?.finishReason) {
           finishReason = event.candidates[0].finishReason;
+          span.setAttributes({
+            finishReason
+          });
         }
         if (event.usageMetadata) {
           if (event.usageMetadata.promptTokenCount) {
@@ -2710,7 +2769,10 @@ async function* GoogleChatCompletionInner(props, ctx) {
       }
     }
   } catch (err) {
-    throw errorToChatCompletionError(err, logRequestData);
+    span.setAttributes({
+      output: content
+    });
+    throw errorToChatCompletionError2(err, logRequestData);
   }
   const outputMessage = {
     role: "assistant",
@@ -2734,8 +2796,7 @@ async function* GoogleChatCompletionInner(props, ctx) {
   span.setAttributes({
     tokensUsed,
     output: content,
-    cost,
-    finishReason
+    cost
   });
 }
 function cleanChatCompletionRequest3(chatCompletionRequest) {
@@ -2801,6 +2862,7 @@ export {
   PromptInvalidOutputError,
   Retry,
   RetryCountContext,
+  RetryLastErrorContext,
   SystemMessage,
   Trace,
   UserMessage,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gammatech/aijsx",
-  "version": "0.10.2-dev.2024-06-11",
+  "version": "0.11.1-dev.2024-06-23",
   "description": "Rewrite of aijsx",
   "author": "Jordan Garcia",
   "license": "MIT",