refactor: create llm response (#5499)

* feat: add LLM response processing functions, including the creation of stream-based and complete responses

* feat: add volta configuration for node and pnpm versions

* refactor: update LLM response handling and event structure in tool choice logic

* feat: update LLM response structure and integrate with tool choice logic

* refactor: clean up imports and remove unused streamResponse function in chat and toolChoice modules

* refactor: rename answer variable to answerBuffer for clarity in LLM response handling

* feat: enhance LLM response handling with tool options and integrate tools into chat and tool choice logic

* refactor: remove volta configuration from package.json

* refactor: reorganize LLM response types and ensure default values for token counts

* refactor: streamline LLM response handling by consolidating response structure and removing redundant checks

* refactor: enhance LLM response handling by consolidating tool options and streamlining event callbacks

* fix: build error

* refactor: update tool type definitions for consistency in tool handling

Author: francismiko

.husky/pre-commit

@@ -1,6 +1,8 @@
 #!/usr/bin/env sh
 . "$(dirname -- "$0")/_/husky.sh"
 
-if command -v npx >/dev/null 2>&1; then
+if command -v pnpm >/dev/null 2>&1; then
+  pnpm lint-staged
+elif command -v npx >/dev/null 2>&1; then
   npx lint-staged
 fi

packages/global/core/ai/request.ts

@@ -0,0 +1,319 @@
+import type {
+  ChatCompletion,
+  ChatCompletionCreateParamsNonStreaming,
+  ChatCompletionCreateParamsStreaming,
+  ChatCompletionMessageToolCall,
+  ChatCompletionToolType,
+  CompletionFinishReason,
+  CompletionUsage,
+  StreamChatType
+} from './type';
+import {
+  parseLLMStreamResponse,
+  parseReasoningContent,
+  removeDatasetCiteText
+} from '../../../service/core/ai/utils';
+import { createChatCompletion } from '../../../service/core/ai/config';
+import type { OpenaiAccountType } from '../../support/user/team/type';
+import { getNanoid } from '../../common/string/tools';
+
+type BasicResponseParams = {
+  reasoning?: boolean;
+  toolMode?: 'toolChoice' | 'prompt';
+  abortSignal?: () => boolean | undefined;
+  retainDatasetCite?: boolean;
+};
+
+type CreateStreamResponseParams = BasicResponseParams & {
+  stream: StreamChatType;
+};
+
+type CreateCompleteResopnseParams = Omit<BasicResponseParams, 'abortSignal'> & {
+  completion: ChatCompletion;
+};
+
+type ResponseEvents = {
+  onStreaming?: ({ responseContent }: { responseContent: string }) => void;
+  onReasoning?: ({ reasoningContent }: { reasoningContent: string }) => void;
+  onToolCalling?: ({
+    callingTool,
+    toolId
+  }: {
+    callingTool: { name: string; arguments: string };
+    toolId: string;
+  }) => void;
+  onToolParaming?: ({
+    currentTool,
+    params
+  }: {
+    currentTool: ChatCompletionMessageToolCall;
+    params: string;
+  }) => void;
+  onReasoned?: ({ reasoningContent }: { reasoningContent: string }) => void;
+  onToolCalled?: ({ calls }: { calls: ChatCompletionMessageToolCall[] }) => void;
+  onCompleted?: ({ responseContent }: { responseContent: string }) => void;
+};
+
+type CreateStreamResponseProps = {
+  params: CreateStreamResponseParams;
+} & ResponseEvents;
+
+type CreateCompleteResopnseProps = {
+  params: CreateCompleteResopnseParams;
+} & ResponseEvents;
+
+type CreateLLMResponseProps = {
+  llmOptions: Omit<
+    ChatCompletionCreateParamsNonStreaming | ChatCompletionCreateParamsStreaming,
+    'tools'
+  >;
+  params: BasicResponseParams;
+  tools: ChatCompletionToolType[];
+  userKey?: OpenaiAccountType;
+} & ResponseEvents;
+
+type LLMResponse = {
+  answerText: string;
+  reasoningText: string;
+  toolCalls: ChatCompletionMessageToolCall[];
+  finish_reason: CompletionFinishReason;
+  isStreamResponse: boolean;
+  getEmptyResponseTip: () => string;
+  inputTokens: number;
+  outputTokens: number;
+};
+
+type StreamResponse = Pick<
+  LLMResponse,
+  'answerText' | 'reasoningText' | 'toolCalls' | 'finish_reason'
+> & {
+  usage?: CompletionUsage;
+};
+
+type CompleteResopnse = Pick<
+  LLMResponse,
+  'answerText' | 'reasoningText' | 'toolCalls' | 'finish_reason'
+> & {
+  usage?: CompletionUsage;
+};
+
+export const createLLMResponse = async (args: CreateLLMResponseProps): Promise<LLMResponse> => {
+  const { llmOptions, tools, userKey, params, ...events } = args;
+
+  const body: ChatCompletionCreateParamsNonStreaming | ChatCompletionCreateParamsStreaming = {
+    tools,
+    ...llmOptions
+  };
+
+  const { response, isStreamResponse, getEmptyResponseTip } = await createChatCompletion({
+    body,
+    userKey,
+    options: {
+      headers: {
+        Accept: 'application/json, text/plain, */*'
+      }
+    }
+  });
+
+  if (isStreamResponse) {
+    const { usage, ...streamResults } = await createStreamResponse({
+      params: { stream: response, ...params },
+      ...events
+    });
+
+    return {
+      inputTokens: usage?.prompt_tokens ?? 0,
+      outputTokens: usage?.completion_tokens ?? 0,
+      isStreamResponse,
+      getEmptyResponseTip,
+      ...streamResults
+    };
+  } else {
+    const { usage, ...completeResults } = await createCompleteResponse({
+      params: { completion: response, ...params },
+      ...events
+    });
+
+    return {
+      inputTokens: usage?.prompt_tokens ?? 0,
+      outputTokens: usage?.completion_tokens ?? 0,
+      isStreamResponse,
+      getEmptyResponseTip,
+      ...completeResults
+    };
+  }
+};
+
+export const createStreamResponse = async (
+  args: CreateStreamResponseProps
+): Promise<StreamResponse> => {
+  const { params, ...events } = args;
+
+  const {
+    abortSignal,
+    stream,
+    reasoning,
+    retainDatasetCite = true,
+    toolMode = 'toolChoice'
+  } = params;
+
+  const { parsePart, getResponseData, updateFinishReason } = parseLLMStreamResponse();
+
+  let calls: ChatCompletionMessageToolCall[] = [];
+  let startResponseWrite = false;
+  let answer = '';
+
+  for await (const part of stream) {
+    if (abortSignal && abortSignal()) {
+      stream.controller?.abort();
+      updateFinishReason('close');
+      break;
+    }
+
+    const responseChoice = part.choices?.[0]?.delta;
+    const {
+      reasoningContent,
+      responseContent,
+      content: originContent
+    } = parsePart({
+      part,
+      parseThinkTag: true,
+      retainDatasetCite
+    });
+
+    if (reasoning && reasoningContent) {
+      events?.onReasoning?.({ reasoningContent });
+    }
+    if (responseContent && originContent) {
+      if (toolMode === 'prompt') {
+        answer += originContent;
+        if (startResponseWrite) {
+          events?.onStreaming?.({ responseContent });
+        } else if (answer.length >= 3) {
+          answer = answer.trimStart();
+          if (/0(:|：)/.test(answer)) {
+            startResponseWrite = true;
+
+            // find first : index
+            const firstIndex =
+              answer.indexOf('0:') !== -1 ? answer.indexOf('0:') : answer.indexOf('0：');
+            answer = answer.substring(firstIndex + 2).trim();
+
+            events?.onStreaming?.({ responseContent: answer });
+          }
+        }
+      } else {
+        events?.onStreaming?.({ responseContent });
+      }
+    }
+    if (responseChoice?.tool_calls?.length) {
+      let callingTool: { name: string; arguments: string } | null = null;
+      responseChoice.tool_calls.forEach((toolCall, i) => {
+        const index = toolCall.index ?? i;
+
+        // Call new tool
+        const hasNewTool = toolCall?.function?.name || callingTool;
+        if (hasNewTool) {
+          // 有 function name，代表新 call 工具
+          if (toolCall?.function?.name) {
+            callingTool = {
+              name: toolCall.function?.name || '',
+              arguments: toolCall.function?.arguments || ''
+            };
+          } else if (callingTool) {
+            // Continue call(Perhaps the name of the previous function was incomplete)
+            callingTool.name += toolCall.function?.name || '';
+            callingTool.arguments += toolCall.function?.arguments || '';
+          }
+
+          if (!callingTool) {
+            return;
+          }
+
+          const toolId = getNanoid();
+
+          events?.onToolCalling?.({ callingTool, toolId });
+
+          calls[index] = {
+            ...toolCall,
+            id: toolId,
+            type: 'function',
+            function: callingTool
+          };
+          callingTool = null;
+        } else {
+          /* 追加到当前工具的参数里 */
+          const arg: string = toolCall?.function?.arguments ?? '';
+          const currentTool = calls[index];
+          if (currentTool && arg) {
+            currentTool.function.arguments += arg;
+            events?.onToolParaming?.({ currentTool, params: arg });
+          }
+        }
+      });
+    }
+  }
+
+  const { reasoningContent, content, finish_reason, usage } = getResponseData();
+
+  return {
+    answerText: content,
+    reasoningText: reasoningContent,
+    toolCalls: calls.filter(Boolean),
+    finish_reason,
+    usage
+  };
+};
+
+export const createCompleteResponse = async (
+  args: CreateCompleteResopnseProps
+): Promise<CompleteResopnse> => {
+  const { params, ...events } = args;
+
+  const { completion, reasoning, retainDatasetCite = true } = params;
+
+  const finish_reason = completion.choices?.[0]?.finish_reason as CompletionFinishReason;
+  const calls = completion.choices?.[0]?.message?.tool_calls || [];
+  const usage = completion.usage;
+
+  const { content, reasoningContent } = (() => {
+    const content = completion.choices?.[0]?.message?.content || '';
+    // @ts-ignore
+    const reasoningContent: string = completion.choices?.[0]?.message?.reasoning_content || '';
+
+    // API already parse reasoning content
+    if (reasoningContent || !reasoning) {
+      return {
+        content,
+        reasoningContent
+      };
+    }
+
+    const [think, answer] = parseReasoningContent(content);
+    return {
+      content: answer,
+      reasoningContent: think
+    };
+  })();
+
+  const formatReasonContent = removeDatasetCiteText(reasoningContent, retainDatasetCite);
+  const formatContent = removeDatasetCiteText(content, retainDatasetCite);
+
+  if (reasoning && reasoningContent) {
+    events?.onReasoned?.({ reasoningContent: formatReasonContent });
+  }
+  if (calls.length !== 0) {
+    events?.onToolCalled?.({ calls });
+  }
+  if (content) {
+    events?.onCompleted?.({ responseContent: formatContent });
+  }
+
+  return {
+    reasoningText: formatReasonContent,
+    answerText: formatContent,
+    toolCalls: calls,
+    finish_reason,
+    usage
+  };
+};

packages/global/core/ai/type.d.ts

@@ -64,6 +64,11 @@ export type ChatCompletionMessageToolCall = ChatCompletionMessageToolCall & {
   toolName?: string;
   toolAvatar?: string;
 };
+export type ChatCompletionToolType = ChatCompletionTool & {
+  index?: number;
+  toolName?: string;
+  toolAvatar?: string;
+};
 export type ChatCompletionMessageFunctionCall =
   SdkChatCompletionAssistantMessageParam.FunctionCall & {
     id?: string;