Refactor code to be more clean and modularized. Bump package version

2025-10-05 22:22:59 +00:00 · 2025-09-29 18:27:29 +02:00 · 2025-09-29 18:27:29 +02:00 · 70a077ca51
commit 70a077ca51
parent ef1526c76a
11 changed files with 701 additions and 251 deletions
--- a/package.json
+++ b/package.json
@ -4,7 +4,7 @@
  "name": "copilot-bridge",
  "displayName": "Copilot Bridge",
  "description": "Local OpenAI-compatible chat endpoint (inference) bridging to GitHub Copilot via the VS Code Language Model API.",
-  "version": "0.2.2",
+  "version": "1.0.0",
  "publisher": "thinkability",
  "repository": {
    "type": "git",
--- a/src/http/routes/chat.ts
+++ b/src/http/routes/chat.ts
@ -1,54 +1,20 @@
 import * as vscode from 'vscode';
 import type { IncomingMessage, ServerResponse } from 'http';
 import { state } from '../../state';
-import { getBridgeConfig } from '../../config';
-import { isChatCompletionRequest, normalizeMessagesLM, convertOpenAIToolsToLM, convertFunctionsToTools } from '../../messages';
-import { getModel, hasLMApi } from '../../models';
-import { readJson, writeErrorResponse, writeJson } from '../utils';
+import { isChatCompletionRequest, type ChatCompletionRequest } from '../../messages';
+import { readJson, writeErrorResponse } from '../utils';
 import { verbose } from '../../log';
+import { ModelService } from '../../services/model-service';
+import { StreamingResponseHandler } from '../../services/streaming-handler';
+import { processLanguageModelResponse, sendCompletionResponse } from '../../services/response-formatter';
+import type { ChatCompletionContext } from '../../types/openai-types';

-// OpenAI response interfaces for better typing
-interface OpenAIToolCall {
-  id: string;
-  type: 'function';
-  function: {
-    name: string;
-    arguments: string;
-  };
-}
-
-interface OpenAIMessage {
-  role: 'assistant';
-  content: string | null;
-  tool_calls?: OpenAIToolCall[];
-  function_call?: {
-    name: string;
-    arguments: string;
-  };
-}
-
-interface OpenAIChoice {
-  index: number;
-  message?: OpenAIMessage;
-  delta?: Partial<OpenAIMessage>;
-  finish_reason: 'stop' | 'length' | 'tool_calls' | 'content_filter' | 'function_call' | null;
-}
-
-interface OpenAIResponse {
-  id: string;
-  object: 'chat.completion' | 'chat.completion.chunk';
-  created: number;
-  model: string;
-  choices: OpenAIChoice[];
-  usage?: {
-    prompt_tokens: number;
-    completion_tokens: number;
-    total_tokens: number;
-  };
-}
-
-export const handleChatCompletion = async (req: IncomingMessage, res: ServerResponse): Promise<void> => {
-  const config = getBridgeConfig();
+/**
+ * Handles OpenAI-compatible chat completion requests with support for streaming and tool calling
+ * @param req - HTTP request object
+ * @param res - HTTP response object
+ */
+export async function handleChatCompletion(req: IncomingMessage, res: ServerResponse): Promise<void> {
  state.activeRequests++;
  verbose(`Request started (active=${state.activeRequests})`);

@ -58,202 +24,75 @@ export const handleChatCompletion = async (req: IncomingMessage, res: ServerResp
      return writeErrorResponse(res, 400, 'invalid request', 'invalid_request_error', 'invalid_payload');
    }

-    const requestedModel = body.model;
-    const stream = body.stream !== false; // default true
+    const modelService = new ModelService();
    
-    // Handle tools and deprecated functions
-    let tools = body.tools || [];
-    if (body.functions) {
-      // Convert deprecated functions to tools format
-      tools = [...tools, ...convertFunctionsToTools(body.functions)];
-    }
-    
-    const model = await getModel(false, requestedModel);
-
-    if (!model) {
-      const hasLM = hasLMApi();
-      if (requestedModel && hasLM) {
-        state.lastReason = 'not_found';
-        return writeErrorResponse(res, 404, 'model not found', 'invalid_request_error', 'model_not_found', 'not_found');
-      }
-      const reason = !hasLM ? 'missing_language_model_api' : (state.lastReason || 'copilot_model_unavailable');
-      return writeErrorResponse(res, 503, 'Copilot unavailable', 'server_error', 'copilot_unavailable', reason);
+    // Validate model availability
+    const modelValidation = await modelService.validateModel(body.model);
+    if (!modelValidation.isValid) {
+      const errorMessage = body.model ? 'model not found' : 'Copilot unavailable';
+      return writeErrorResponse(
+        res, 
+        modelValidation.statusCode!, 
+        errorMessage, 
+        modelValidation.errorType!, 
+        modelValidation.errorCode!, 
+        modelValidation.reason || 'unknown_error'
+      );
    }

-    const lmMessages = normalizeMessagesLM(body.messages, config.historyWindow) as vscode.LanguageModelChatMessage[];
-    const lmTools = convertOpenAIToolsToLM(tools);
+    // Create processing context
+    const context = await modelService.createProcessingContext(body);
+    const chatContext = modelService.createChatCompletionContext(body, context.lmTools.length > 0);
    
-    // Prepare request options for Language Model API
-    const requestOptions: any = {};
-    if (lmTools.length > 0) {
-      requestOptions.tools = lmTools;
-    }
-    
-    verbose(`LM request via API model=${model.family || model.id || model.name || 'unknown'} tools=${lmTools.length}`);
+    verbose(`LM request via API model=${context.model.family || context.model.id || context.model.name || 'unknown'} tools=${context.lmTools.length}`);

-    const cts = new vscode.CancellationTokenSource();
-    const response = await model.sendRequest(lmMessages, requestOptions, cts.token);
-    await sendResponse(res, response, stream, body, tools);
-  } catch (e) {
-    const msg = e instanceof Error ? e.message : String(e);
-    writeErrorResponse(res, 500, msg || 'internal_error', 'server_error', 'internal_error');
+    // Execute the Language Model request
+    const cancellationToken = new vscode.CancellationTokenSource();
+    const response = await context.model.sendRequest(
+      context.lmMessages, 
+      context.requestOptions, 
+      cancellationToken.token
+    );
+
+    // Handle response based on streaming preference
+    if (chatContext.isStreaming) {
+      await handleStreamingResponse(res, response, chatContext, body);
+    } else {
+      await handleNonStreamingResponse(res, response, chatContext, body);
+    }
+
+  } catch (error) {
+    const errorMessage = error instanceof Error ? error.message : String(error);
+    writeErrorResponse(res, 500, errorMessage || 'internal_error', 'server_error', 'internal_error');
  } finally {
    state.activeRequests--;
    verbose(`Request complete (active=${state.activeRequests})`);
  }
-};
+}

-const sendResponse = async (
-  res: ServerResponse, 
-  response: vscode.LanguageModelChatResponse, 
-  stream: boolean,
-  requestBody?: any,
-  tools?: any[]
-): Promise<void> => {
-  const modelName = requestBody?.model || 'copilot';
-  const responseId = `chatcmpl-${Math.random().toString(36).slice(2)}`;
-  const created = Math.floor(Date.now() / 1000);
+/**
+ * Handles streaming response using Server-Sent Events
+ */
+async function handleStreamingResponse(
+  res: ServerResponse,
+  response: vscode.LanguageModelChatResponse,
+  chatContext: ChatCompletionContext,
+  requestBody: ChatCompletionRequest
+): Promise<void> {
+  const streamHandler = new StreamingResponseHandler(res, chatContext, requestBody);
+  streamHandler.initializeStream();
+  await streamHandler.processAndStreamResponse(response);
+}

-  if (stream) {
-    res.writeHead(200, {
-      'Content-Type': 'text/event-stream',
-      'Cache-Control': 'no-cache',
-      'Connection': 'keep-alive',
-    });
-    
-    verbose(`SSE start id=${responseId}`);
-    
-    let toolCalls: OpenAIToolCall[] = [];
-    
-    for await (const part of response.stream) {
-      // Check if this part is a LanguageModelToolCallPart
-      if (part && typeof part === 'object' && 'callId' in part && 'name' in part && 'input' in part) {
-        const toolCallPart = part as vscode.LanguageModelToolCallPart;
-        const toolCall: OpenAIToolCall = {
-          id: toolCallPart.callId,
-          type: 'function',
-          function: {
-            name: toolCallPart.name,
-            arguments: JSON.stringify(toolCallPart.input)
-          }
-        };
-        toolCalls.push(toolCall);
-        
-        // Send tool call in streaming format
-        const chunkResponse: OpenAIResponse = {
-          id: responseId,
-          object: 'chat.completion.chunk',
-          created,
-          model: modelName,
-          choices: [{
-            index: 0,
-            delta: {
-              tool_calls: [toolCall]
-            },
-            finish_reason: null
-          }]
-        };
-        res.write(`data: ${JSON.stringify(chunkResponse)}\n\n`);
-      } else if (typeof part === 'string' || (part && typeof part === 'object' && 'value' in part)) {
-        // Handle text content
-        const content = typeof part === 'string' ? part : (part as any).value || '';
-        if (content) {
-          const chunkResponse: OpenAIResponse = {
-            id: responseId,
-            object: 'chat.completion.chunk',
-            created,
-            model: modelName,
-            choices: [{
-              index: 0,
-              delta: { content },
-              finish_reason: null
-            }]
-          };
-          res.write(`data: ${JSON.stringify(chunkResponse)}\n\n`);
-        }
-      }
-    }
-    
-    // Send final chunk
-    const finishReason: OpenAIChoice['finish_reason'] = toolCalls.length > 0 ? 'tool_calls' : 'stop';
-    const finalChunkResponse: OpenAIResponse = {
-      id: responseId,
-      object: 'chat.completion.chunk',
-      created,
-      model: modelName,
-      choices: [{
-        index: 0,
-        delta: {},
-        finish_reason: finishReason
-      }]
-    };
-    res.write(`data: ${JSON.stringify(finalChunkResponse)}\n\n`);
-    
-    verbose(`SSE end id=${responseId}`);
-    res.write('data: [DONE]\n\n');
-    res.end();
-    return;
-  }
-
-  // Non-streaming response
-  let content = '';
-  let toolCalls: OpenAIToolCall[] = [];
-  
-  for await (const part of response.stream) {
-    if (part && typeof part === 'object' && 'callId' in part && 'name' in part && 'input' in part) {
-      // Handle VS Code LanguageModelToolCallPart
-      const toolCallPart = part as vscode.LanguageModelToolCallPart;
-      const toolCall: OpenAIToolCall = {
-        id: toolCallPart.callId,
-        type: 'function',
-        function: {
-          name: toolCallPart.name,
-          arguments: JSON.stringify(toolCallPart.input)
-        }
-      };
-      toolCalls.push(toolCall);
-    } else if (typeof part === 'string' || (part && typeof part === 'object' && 'value' in part)) {
-      // Handle text content
-      content += typeof part === 'string' ? part : (part as any).value || '';
-    }
-  }
-  
-  verbose(`Non-stream complete len=${content.length} tool_calls=${toolCalls.length}`);
-  
-  const message: OpenAIMessage = {
-    role: 'assistant',
-    content: toolCalls.length > 0 ? null : content,
-  };
-  
-  // Add tool_calls if present
-  if (toolCalls.length > 0) {
-    message.tool_calls = toolCalls;
-    
-    // For backward compatibility, also add function_call if there's exactly one tool call
-    if (toolCalls.length === 1 && requestBody?.function_call !== undefined) {
-      message.function_call = {
-        name: toolCalls[0].function.name,
-        arguments: toolCalls[0].function.arguments
-      };
-    }
-  }
-  
-  const responseObj: OpenAIResponse = {
-    id: responseId,
-    object: 'chat.completion',
-    created,
-    model: modelName,
-    choices: [{
-      index: 0,
-      message,
-      finish_reason: toolCalls.length > 0 ? 'tool_calls' : 'stop',
-    }],
-    usage: {
-      prompt_tokens: 0, // VS Code API doesn't provide token counts
-      completion_tokens: 0,
-      total_tokens: 0
-    }
-  };
-  
-  writeJson(res, 200, responseObj);
-};
+/**
+ * Handles non-streaming response with complete data
+ */
+async function handleNonStreamingResponse(
+  res: ServerResponse,
+  response: vscode.LanguageModelChatResponse,
+  chatContext: ChatCompletionContext,
+  requestBody: ChatCompletionRequest
+): Promise<void> {
+  const processedData = await processLanguageModelResponse(response);
+  sendCompletionResponse(res, chatContext, processedData, requestBody);
+}
--- a/src/http/routes/health.ts
+++ b/src/http/routes/health.ts
@ -5,17 +5,42 @@ import { hasLMApi, getModel } from '../../models';
 import { state } from '../../state';
 import { verbose } from '../../log';

+interface HealthResponse {
+  readonly ok: boolean;
+  readonly status: string;
+  readonly copilot: string;
+  readonly reason?: string;
+  readonly version: string;
+  readonly features: {
+    readonly chat_completions: boolean;
+    readonly streaming: boolean;
+    readonly tool_calling: boolean;
+    readonly function_calling: boolean;
+    readonly models_list: boolean;
+  };
+  readonly active_requests: number;
+  readonly model_attempted?: boolean;
+}
+
 export const handleHealthCheck = async (res: ServerResponse, v: boolean): Promise<void> => {
  const hasLM = hasLMApi();
+  
+  // Attempt model resolution if cache is empty and verbose logging is enabled
  if (!state.modelCache && v) {
    verbose(`Healthz: model=${state.modelCache ? 'present' : 'missing'} lmApi=${hasLM ? 'ok' : 'missing'}`);
-    await getModel();
+    try {
+      await getModel();
+    } catch (e) {
+      const msg = e instanceof Error ? e.message : String(e);
+      verbose(`Health check model resolution failed: ${msg}`);
+    }
  }
+  
  const unavailableReason = state.modelCache
    ? undefined
    : (!hasLM ? 'missing_language_model_api' : (state.lastReason || 'copilot_model_unavailable'));
  
-  writeJson(res, 200, {
+  const response: HealthResponse = {
    ok: true,
    status: 'operational',
    copilot: state.modelCache ? 'ok' : 'unavailable',
@ -30,5 +55,7 @@ export const handleHealthCheck = async (res: ServerResponse, v: boolean): Promis
    },
    active_requests: state.activeRequests,
    model_attempted: state.modelAttempted
-  });
+  };
+  
+  writeJson(res, 200, response);
 };
--- a/src/http/routes/models.ts
+++ b/src/http/routes/models.ts
@ -1,13 +1,31 @@
-import { writeJson } from '../utils';
+import { writeJson, writeErrorResponse } from '../utils';
 import { listCopilotModels } from '../../models';
+import { verbose } from '../../log';
 import type { ServerResponse } from 'http';

+interface ModelObject {
+  readonly id: string;
+  readonly object: 'model';
+  readonly created: number;
+  readonly owned_by: string;
+  readonly permission: readonly unknown[];
+  readonly root: string;
+  readonly parent: null;
+}
+
+interface ModelsListResponse {
+  readonly object: 'list';
+  readonly data: readonly ModelObject[];
+}
+
 export const handleModelsRequest = async (res: ServerResponse): Promise<void> => {
  try {
    const modelIds = await listCopilotModels();
-    const models = modelIds.map((id: string) => ({
+    verbose(`Models listed: ${modelIds.length} available`);
+    
+    const models: ModelObject[] = modelIds.map((id: string) => ({
      id,
-      object: 'model',
+      object: 'model' as const,
      created: Math.floor(Date.now() / 1000),
      owned_by: 'copilot',
      permission: [],
@ -15,18 +33,15 @@ export const handleModelsRequest = async (res: ServerResponse): Promise<void> =>
      parent: null,
    }));

-    writeJson(res, 200, {
+    const response: ModelsListResponse = {
      object: 'list',
      data: models,
-    });
+    };
+    
+    writeJson(res, 200, response);
  } catch (e) {
    const msg = e instanceof Error ? e.message : String(e);
-    writeJson(res, 500, {
-      error: {
-        message: msg || 'Failed to list models',
-        type: 'server_error',
-        code: 'internal_error'
-      }
-    });
+    verbose(`Models request failed: ${msg}`);
+    writeErrorResponse(res, 500, msg || 'Failed to list models', 'server_error', 'internal_error');
  }
 };
--- a/src/models.ts
+++ b/src/models.ts
@ -65,7 +65,8 @@ export const handleModelSelectionError = (error: unknown, family?: string): void

 export const listCopilotModels = async (): Promise<string[]> => {
  try {
-    const models = await selectChatModels();
+    // Filter for Copilot models only, consistent with getModel behavior
+    const models = await vscode.lm.selectChatModels({ vendor: 'copilot' });
    const ids = models.map((m: vscode.LanguageModelChat) => {
      const normalized = m.family || m.id || m.name || 'copilot';
      return `${normalized}`;
--- a/src/services/model-service.ts
+++ b/src/services/model-service.ts
@ -0,0 +1,99 @@
+import type * as vscode from 'vscode';
+import type { ChatCompletionRequest } from '../messages';
+import type { 
+  ModelValidationResult, 
+  RequestProcessingContext,
+  ChatCompletionContext
+} from '../types/openai-types';
+import { 
+  extractAndMergeTools, 
+  createLanguageModelRequestOptions 
+} from './request-processor';
+import { getModel, hasLMApi } from '../models';
+import { normalizeMessagesLM, convertOpenAIToolsToLM } from '../messages';
+import { getBridgeConfig } from '../config';
+
+/**
+ * Service for validating models and creating request processing context
+ */
+export class ModelService {
+  
+  /**
+   * Validates the requested model and returns appropriate error details if invalid
+   * @param requestedModel - The model identifier from the request
+   * @returns Validation result with error details if model is unavailable
+   */
+  public async validateModel(requestedModel?: string): Promise<ModelValidationResult> {
+    const model = await getModel(false, requestedModel);
+    
+    if (!model) {
+      const hasLM = hasLMApi();
+      
+      if (requestedModel && hasLM) {
+        return {
+          isValid: false,
+          statusCode: 404,
+          errorType: 'invalid_request_error',
+          errorCode: 'model_not_found',
+          reason: 'not_found'
+        };
+      }
+      
+      const reason = !hasLM ? 'missing_language_model_api' : 'copilot_model_unavailable';
+      return {
+        isValid: false,
+        statusCode: 503,
+        errorType: 'server_error',
+        errorCode: 'copilot_unavailable',
+        reason
+      };
+    }
+    
+    return { isValid: true };
+  }
+
+  /**
+   * Creates a complete request processing context from validated inputs
+   * @param body - The validated chat completion request
+   * @returns Processing context with all required elements for the Language Model API
+   */
+  public async createProcessingContext(body: ChatCompletionRequest): Promise<RequestProcessingContext> {
+    const model = await getModel(false, body.model);
+    if (!model) {
+      throw new Error('Model validation should be performed before creating processing context');
+    }
+
+    const config = getBridgeConfig();
+    const mergedTools = extractAndMergeTools(body);
+    const lmMessages = normalizeMessagesLM(body.messages, config.historyWindow);
+    const lmTools = convertOpenAIToolsToLM(mergedTools);
+    const requestOptions = createLanguageModelRequestOptions(lmTools);
+
+    return {
+      model,
+      lmMessages: lmMessages as vscode.LanguageModelChatMessage[],
+      lmTools,
+      requestOptions,
+      mergedTools
+    };
+  }
+
+  /**
+   * Creates chat completion context for response formatting
+   * @param body - The chat completion request
+   * @param hasTools - Whether tools are present in the request
+   * @returns Context object for response handling
+   */
+  public createChatCompletionContext(
+    body: ChatCompletionRequest, 
+    hasTools: boolean
+  ): ChatCompletionContext {
+    return {
+      requestId: `chatcmpl-${Math.random().toString(36).slice(2)}`,
+      modelName: body.model || 'copilot',
+      created: Math.floor(Date.now() / 1000),
+      hasTools,
+      isStreaming: body.stream !== false
+    };
+  }
+}
--- a/src/services/request-processor.ts
+++ b/src/services/request-processor.ts
@ -0,0 +1,39 @@
+import type { ChatCompletionRequest, Tool } from '../messages';
+import type * as vscode from 'vscode';
+
+/**
+ * Validates and extracts tool configurations from request body
+ * @param body - The parsed request body
+ * @returns Combined tools array including converted deprecated functions
+ */
+export function extractAndMergeTools(body: ChatCompletionRequest): Tool[] {
+  const tools = body.tools || [];
+  
+  if (body.functions) {
+    // Convert deprecated functions to tools format
+    const convertedTools: Tool[] = body.functions.map(func => ({
+      type: 'function' as const,
+      function: func
+    }));
+    return [...tools, ...convertedTools];
+  }
+  
+  return tools;
+}
+
+/**
+ * Creates VS Code Language Model request options from processed context
+ * @param lmTools - Array of Language Model compatible tools
+ * @returns Request options object for the Language Model API
+ */
+export function createLanguageModelRequestOptions(
+  lmTools: vscode.LanguageModelChatTool[]
+): vscode.LanguageModelChatRequestOptions {
+  const options: vscode.LanguageModelChatRequestOptions = {};
+  
+  if (lmTools.length > 0) {
+    options.tools = lmTools;
+  }
+  
+  return options;
+}
--- a/src/services/response-formatter.ts
+++ b/src/services/response-formatter.ts
@ -0,0 +1,158 @@
+import type * as vscode from 'vscode';
+import type { ServerResponse } from 'http';
+import type { 
+  OpenAIResponse, 
+  OpenAIChoice, 
+  OpenAIMessage, 
+  OpenAIToolCall, 
+  ChatCompletionContext,
+  ProcessedResponseData 
+} from '../types/openai-types';
+import type { ChatCompletionRequest } from '../messages';
+import { writeJson } from '../http/utils';
+import { verbose } from '../log';
+
+/**
+ * Processes VS Code Language Model stream parts into structured data
+ * @param response - The VS Code Language Model chat response
+ * @returns Promise resolving to processed content and tool calls
+ */
+export async function processLanguageModelResponse(
+  response: vscode.LanguageModelChatResponse
+): Promise<ProcessedResponseData> {
+  let content = '';
+  const toolCalls: OpenAIToolCall[] = [];
+  
+  for await (const part of response.stream) {
+    if (isToolCallPart(part)) {
+      const toolCall: OpenAIToolCall = {
+        id: part.callId,
+        type: 'function',
+        function: {
+          name: part.name,
+          arguments: JSON.stringify(part.input)
+        }
+      };
+      toolCalls.push(toolCall);
+    } else if (isTextPart(part)) {
+      content += extractTextContent(part);
+    }
+  }
+  
+  const finishReason: OpenAIChoice['finish_reason'] = toolCalls.length > 0 ? 'tool_calls' : 'stop';
+  
+  return {
+    content,
+    toolCalls,
+    finishReason
+  };
+}
+
+/**
+ * Creates an OpenAI-compatible response message
+ * @param data - The processed response data
+ * @param requestBody - Original request body for backward compatibility
+ * @returns OpenAI message object
+ */
+export function createOpenAIMessage(
+  data: ProcessedResponseData,
+  requestBody?: ChatCompletionRequest
+): OpenAIMessage {
+  const baseMessage = {
+    role: 'assistant' as const,
+    content: data.toolCalls.length > 0 ? null : data.content,
+  };
+  
+  // Add tool_calls if present
+  if (data.toolCalls.length > 0) {
+    const messageWithTools = {
+      ...baseMessage,
+      tool_calls: data.toolCalls,
+    };
+    
+    // For backward compatibility, also add function_call if there's exactly one tool call
+    if (data.toolCalls.length === 1 && requestBody?.function_call !== undefined) {
+      return {
+        ...messageWithTools,
+        function_call: {
+          name: data.toolCalls[0].function.name,
+          arguments: data.toolCalls[0].function.arguments
+        }
+      };
+    }
+    
+    return messageWithTools;
+  }
+  
+  return baseMessage;
+}
+
+/**
+ * Sends a complete (non-streaming) OpenAI-compatible response
+ * @param res - HTTP response object
+ * @param context - Chat completion context
+ * @param data - Processed response data
+ * @param requestBody - Original request body
+ */
+export function sendCompletionResponse(
+  res: ServerResponse,
+  context: ChatCompletionContext,
+  data: ProcessedResponseData,
+  requestBody?: ChatCompletionRequest
+): void {
+  const message = createOpenAIMessage(data, requestBody);
+  
+  const responseObj: OpenAIResponse = {
+    id: context.requestId,
+    object: 'chat.completion',
+    created: context.created,
+    model: context.modelName,
+    choices: [{
+      index: 0,
+      message,
+      finish_reason: data.finishReason,
+    }],
+    usage: {
+      prompt_tokens: 0, // VS Code API doesn't provide token counts
+      completion_tokens: 0,
+      total_tokens: 0
+    }
+  };
+  
+  verbose(`Non-stream complete len=${data.content.length} tool_calls=${data.toolCalls.length}`);
+  writeJson(res, 200, responseObj);
+}
+
+/**
+ * Type guard for VS Code LanguageModelToolCallPart
+ */
+function isToolCallPart(part: unknown): part is vscode.LanguageModelToolCallPart {
+  return part !== null && 
+         typeof part === 'object' && 
+         'callId' in part && 
+         'name' in part && 
+         'input' in part;
+}
+
+/**
+ * Type guard for text content parts
+ */
+function isTextPart(part: unknown): boolean {
+  return typeof part === 'string' || 
+         (part !== null && typeof part === 'object' && 'value' in part);
+}
+
+/**
+ * Extracts text content from various part types
+ */
+function extractTextContent(part: unknown): string {
+  if (typeof part === 'string') {
+    return part;
+  }
+  
+  if (part !== null && typeof part === 'object' && 'value' in part) {
+    return String((part as { value: unknown }).value) || '';
+  }
+  
+  return '';
+}
--- a/src/services/streaming-handler.ts
+++ b/src/services/streaming-handler.ts
@ -0,0 +1,190 @@
+import type * as vscode from 'vscode';
+import type { ServerResponse } from 'http';
+import type { 
+  OpenAIResponse, 
+  OpenAIToolCall, 
+  ChatCompletionContext 
+} from '../types/openai-types';
+import type { ChatCompletionRequest } from '../messages';
+import { verbose } from '../log';
+
+/**
+ * Handles Server-Sent Events streaming for OpenAI-compatible chat completions
+ */
+export class StreamingResponseHandler {
+  private readonly response: ServerResponse;
+  private readonly context: ChatCompletionContext;
+  private readonly requestBody?: ChatCompletionRequest;
+  
+  constructor(
+    response: ServerResponse, 
+    context: ChatCompletionContext,
+    requestBody?: ChatCompletionRequest
+  ) {
+    this.response = response;
+    this.context = context;
+    this.requestBody = requestBody;
+  }
+
+  /**
+   * Initializes the SSE stream with proper headers
+   */
+  public initializeStream(): void {
+    this.response.writeHead(200, {
+      'Content-Type': 'text/event-stream',
+      'Cache-Control': 'no-cache',
+      'Connection': 'keep-alive',
+    });
+    
+    verbose(`SSE start id=${this.context.requestId}`);
+  }
+
+  /**
+   * Processes the Language Model response stream and sends SSE chunks
+   * @param languageModelResponse - VS Code Language Model response
+   */
+  public async processAndStreamResponse(
+    languageModelResponse: vscode.LanguageModelChatResponse
+  ): Promise<void> {
+    const toolCalls: OpenAIToolCall[] = [];
+    
+    for await (const part of languageModelResponse.stream) {
+      if (this.isToolCallPart(part)) {
+        const toolCall = this.createToolCallFromPart(part);
+        toolCalls.push(toolCall);
+        this.sendToolCallChunk(toolCall);
+      } else if (this.isTextPart(part)) {
+        const content = this.extractTextContent(part);
+        if (content) {
+          this.sendContentChunk(content);
+        }
+      }
+    }
+    
+    this.sendFinalChunk(toolCalls.length > 0 ? 'tool_calls' : 'stop');
+    this.endStream();
+  }
+
+  /**
+   * Sends a content delta chunk
+   */
+  private sendContentChunk(content: string): void {
+    const chunkResponse: OpenAIResponse = {
+      id: this.context.requestId,
+      object: 'chat.completion.chunk',
+      created: this.context.created,
+      model: this.context.modelName,
+      choices: [{
+        index: 0,
+        delta: { content },
+        finish_reason: null
+      }]
+    };
+    
+    this.writeSSEData(chunkResponse);
+  }
+
+  /**
+   * Sends a tool call chunk
+   */
+  private sendToolCallChunk(toolCall: OpenAIToolCall): void {
+    const chunkResponse: OpenAIResponse = {
+      id: this.context.requestId,
+      object: 'chat.completion.chunk',
+      created: this.context.created,
+      model: this.context.modelName,
+      choices: [{
+        index: 0,
+        delta: {
+          tool_calls: [toolCall]
+        },
+        finish_reason: null
+      }]
+    };
+    
+    this.writeSSEData(chunkResponse);
+  }
+
+  /**
+   * Sends the final completion chunk with finish reason
+   */
+  private sendFinalChunk(finishReason: 'stop' | 'tool_calls'): void {
+    const finalChunkResponse: OpenAIResponse = {
+      id: this.context.requestId,
+      object: 'chat.completion.chunk',
+      created: this.context.created,
+      model: this.context.modelName,
+      choices: [{
+        index: 0,
+        delta: {},
+        finish_reason: finishReason
+      }]
+    };
+    
+    this.writeSSEData(finalChunkResponse);
+  }
+
+  /**
+   * Ends the SSE stream
+   */
+  private endStream(): void {
+    verbose(`SSE end id=${this.context.requestId}`);
+    this.response.write('data: [DONE]\n\n');
+    this.response.end();
+  }
+
+  /**
+   * Writes data to the SSE stream
+   */
+  private writeSSEData(data: OpenAIResponse): void {
+    this.response.write(`data: ${JSON.stringify(data)}\n\n`);
+  }
+
+  /**
+   * Creates an OpenAI tool call from VS Code Language Model part
+   */
+  private createToolCallFromPart(part: vscode.LanguageModelToolCallPart): OpenAIToolCall {
+    return {
+      id: part.callId,
+      type: 'function',
+      function: {
+        name: part.name,
+        arguments: JSON.stringify(part.input)
+      }
+    };
+  }
+
+  /**
+   * Type guard for VS Code LanguageModelToolCallPart
+   */
+  private isToolCallPart(part: unknown): part is vscode.LanguageModelToolCallPart {
+    return part !== null && 
+           typeof part === 'object' && 
+           'callId' in part && 
+           'name' in part && 
+           'input' in part;
+  }
+
+  /**
+   * Type guard for text content parts
+   */
+  private isTextPart(part: unknown): boolean {
+    return typeof part === 'string' || 
+           (part !== null && typeof part === 'object' && 'value' in part);
+  }
+
+  /**
+   * Extracts text content from various part types
+   */
+  private extractTextContent(part: unknown): string {
+    if (typeof part === 'string') {
+      return part;
+    }
+    
+    if (part !== null && typeof part === 'object' && 'value' in part) {
+      return String((part as { value: unknown }).value) || '';
+    }
+    
+    return '';
+  }
+}
--- a/src/types/openai-types.ts
+++ b/src/types/openai-types.ts
@ -0,0 +1,81 @@
+import type * as vscode from 'vscode';
+import type { Tool } from '../messages';
+
+/**
+ * OpenAI API compatible types for request and response handling
+ */
+
+export interface OpenAIToolCall {
+  readonly id: string;
+  readonly type: 'function';
+  readonly function: {
+    readonly name: string;
+    readonly arguments: string;
+  };
+}
+
+export interface OpenAIMessage {
+  readonly role: 'assistant';
+  readonly content: string | null;
+  readonly tool_calls?: OpenAIToolCall[];
+  readonly function_call?: {
+    readonly name: string;
+    readonly arguments: string;
+  };
+}
+
+export interface OpenAIChoice {
+  readonly index: number;
+  readonly message?: OpenAIMessage;
+  readonly delta?: Partial<OpenAIMessage>;
+  readonly finish_reason: 'stop' | 'length' | 'tool_calls' | 'content_filter' | 'function_call' | null;
+}
+
+export interface OpenAIResponse {
+  readonly id: string;
+  readonly object: 'chat.completion' | 'chat.completion.chunk';
+  readonly created: number;
+  readonly model: string;
+  readonly choices: OpenAIChoice[];
+  readonly usage?: {
+    readonly prompt_tokens: number;
+    readonly completion_tokens: number;
+    readonly total_tokens: number;
+  };
+}
+
+export interface ChatCompletionContext {
+  readonly requestId: string;
+  readonly modelName: string;
+  readonly created: number;
+  readonly hasTools: boolean;
+  readonly isStreaming: boolean;
+}
+
+export interface ProcessedResponseData {
+  readonly content: string;
+  readonly toolCalls: OpenAIToolCall[];
+  readonly finishReason: OpenAIChoice['finish_reason'];
+}
+
+/**
+ * Validates that the request model is available and properly configured
+ */
+export interface ModelValidationResult {
+  readonly isValid: boolean;
+  readonly statusCode?: number;
+  readonly errorType?: string;
+  readonly errorCode?: string;
+  readonly reason?: string;
+}
+
+/**
+ * Consolidated request processing context for chat completions
+ */
+export interface RequestProcessingContext {
+  readonly model: vscode.LanguageModelChat;
+  readonly lmMessages: vscode.LanguageModelChatMessage[];
+  readonly lmTools: vscode.LanguageModelChatTool[];
+  readonly requestOptions: vscode.LanguageModelChatRequestOptions;
+  readonly mergedTools: Tool[];
+}
--- a/tsconfig.json
+++ b/tsconfig.json
@ -9,6 +9,7 @@
    "sourceMap": true,
    "esModuleInterop": true,
    "allowSyntheticDefaultImports": true,
+    "forceConsistentCasingInFileNames": true,
    "types": ["node", "vscode"]
  },
  "include": ["src/**/*.ts"]