opc-agent 4.1.0 → 4.1.2

package/src/providers/index.ts

@@ -1,632 +1,632 @@
-import type { Message } from '../core/types';
-import type { MCPToolDefinition } from '../tools/mcp';
-import * as https from 'https';
-import * as http from 'http';
-
-export interface ChatOptions {
-  tools?: MCPToolDefinition[];
-}
-
-export interface LLMProvider {
-  name: string;
-  chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string>;
-  chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string>;
-}
-
-interface OpenAIMessage {
-  role: 'system' | 'user' | 'assistant';
-  content: string;
-}
-
-function getApiKey(): string {
-  return process.env.OPC_LLM_API_KEY || process.env.OPENAI_API_KEY || '';
-}
-
-function getBaseUrl(): string {
-  return process.env.OPC_LLM_BASE_URL || 'https://api.openai.com/v1';
-}
-
-function buildToolPrompt(tools: MCPToolDefinition[]): string {
-  const toolsDesc = tools.map(t =>
-    `- ${t.name}: ${t.description}\n  Input schema: ${JSON.stringify(t.inputSchema)}`
-  ).join('\n');
-  return `\n\nYou have access to the following tools. To use a tool, respond with ONLY a JSON object in this format:\n<tool_call>{"name": "tool_name", "arguments": {...}}</tool_call>\n\nAvailable tools:\n${toolsDesc}\n\nIf you don't need a tool, respond normally with text.`;
-}
-
-class OpenAICompatibleProvider implements LLMProvider {
-  name: string;
-  private model: string;
-  private baseUrl: string;
-  private apiKey: string;
-
-  constructor(name: string, model: string, baseUrl?: string, apiKey?: string) {
-    this.name = name;
-    this.model = model;
-    this.baseUrl = baseUrl || getBaseUrl();
-    this.apiKey = apiKey || getApiKey();
-  }
-
-  private formatMessages(messages: Message[], systemPrompt?: string): OpenAIMessage[] {
-    const formatted: OpenAIMessage[] = [];
-    if (systemPrompt) {
-      formatted.push({ role: 'system', content: systemPrompt });
-    }
-    for (const m of messages) {
-      formatted.push({ role: m.role as 'user' | 'assistant', content: m.content });
-    }
-    return formatted;
-  }
-
-  private async request(body: any): Promise<any> {
-    if (!this.apiKey) {
-      throw new Error('No API key configured. Set OPC_LLM_API_KEY or OPENAI_API_KEY environment variable.');
-    }
-
-    const url = new URL(`${this.baseUrl}/chat/completions`);
-    const isGemini = url.hostname.includes('googleapis.com');
-    if (isGemini) {
-      url.searchParams.set('key', this.apiKey);
-    }
-    const isHttps = url.protocol === 'https:';
-    const lib = isHttps ? https : http;
-
-    const postData = JSON.stringify(body);
-
-    const headers: Record<string, string> = {
-      'Content-Type': 'application/json',
-      'Content-Length': String(Buffer.byteLength(postData)),
-    };
-    if (!isGemini) {
-      headers['Authorization'] = `Bearer ${this.apiKey}`;
-    }
-
-    return new Promise((resolve, reject) => {
-      const req = lib.request(
-        {
-          hostname: url.hostname,
-          port: url.port || (isHttps ? 443 : 80),
-          path: url.pathname + url.search,
-          method: 'POST',
-          headers,
-        },
-        (res) => {
-          let data = '';
-          res.on('data', (chunk: Buffer) => (data += chunk.toString()));
-          res.on('end', () => {
-            if (res.statusCode && res.statusCode >= 400) {
-              reject(new Error(`LLM API error ${res.statusCode}: ${data}`));
-              return;
-            }
-            try {
-              resolve(JSON.parse(data));
-            } catch {
-              reject(new Error(`Invalid JSON response: ${data.slice(0, 200)}`));
-            }
-          });
-        },
-      );
-      req.on('error', reject);
-      req.write(postData);
-      req.end();
-    });
-  }
-
-  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
-    if (!this.apiKey) {
-      const last = messages[messages.length - 1];
-      return `[${this.name}/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
-    }
-    let effectivePrompt = systemPrompt;
-    if (options?.tools && options.tools.length > 0) {
-      effectivePrompt = (systemPrompt || '') + buildToolPrompt(options.tools);
-    }
-    const formatted = this.formatMessages(messages, effectivePrompt);
-    const result = await this.request({
-      model: this.model,
-      messages: formatted,
-      temperature: 0.7,
-      max_tokens: 2048,
-    });
-    return result.choices?.[0]?.message?.content ?? '';
-  }
-
-  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
-    if (!this.apiKey) {
-      const last = messages[messages.length - 1];
-      yield `[${this.name}/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
-      return;
-    }
-
-    const formatted = this.formatMessages(messages, systemPrompt);
-    const url = new URL(`${this.baseUrl}/chat/completions`);
-    const isGemini = url.hostname.includes('googleapis.com');
-    if (isGemini) {
-      url.searchParams.set('key', this.apiKey);
-    }
-    const isHttps = url.protocol === 'https:';
-    const lib = isHttps ? https : http;
-    const postData = JSON.stringify({
-      model: this.model,
-      messages: formatted,
-      temperature: 0.7,
-      max_tokens: 2048,
-      stream: true,
-    });
-
-    const streamHeaders: Record<string, string> = {
-      'Content-Type': 'application/json',
-      'Content-Length': String(Buffer.byteLength(postData)),
-    };
-    if (!isGemini) {
-      streamHeaders['Authorization'] = `Bearer ${this.apiKey}`;
-    }
-
-    const response = await new Promise<http.IncomingMessage>((resolve, reject) => {
-      const req = lib.request(
-        {
-          hostname: url.hostname,
-          port: url.port || (isHttps ? 443 : 80),
-          path: url.pathname + url.search,
-          method: 'POST',
-          headers: streamHeaders,
-        },
-        resolve,
-      );
-      req.on('error', reject);
-      req.write(postData);
-      req.end();
-    });
-
-    if (response.statusCode && response.statusCode >= 400) {
-      let data = '';
-      for await (const chunk of response) data += chunk.toString();
-      throw new Error(`LLM API error ${response.statusCode}: ${data}`);
-    }
-
-    let buffer = '';
-    for await (const chunk of response) {
-      buffer += chunk.toString();
-      const lines = buffer.split('\n');
-      buffer = lines.pop() ?? '';
-
-      for (const line of lines) {
-        const trimmed = line.trim();
-        if (!trimmed || !trimmed.startsWith('data: ')) continue;
-        const data = trimmed.slice(6);
-        if (data === '[DONE]') return;
-        try {
-          const parsed = JSON.parse(data);
-          const content = parsed.choices?.[0]?.delta?.content;
-          if (content) yield content;
-        } catch {
-          // skip malformed lines
-        }
-      }
-    }
-  }
-}
-
-class GeminiNativeProvider implements LLMProvider {
-  name = 'gemini';
-  private model: string;
-  private apiKey: string;
-
-  constructor(model: string, apiKey?: string) {
-    this.model = model;
-    this.apiKey = apiKey || getApiKey();
-  }
-
-  private buildUrl(stream: boolean): string {
-    const action = stream ? 'streamGenerateContent?alt=sse&' : 'generateContent?';
-    return `https://generativelanguage.googleapis.com/v1beta/models/${this.model}:${action}key=${this.apiKey}`;
-  }
-
-  private formatContents(messages: Message[], systemPrompt?: string): { contents: any[]; systemInstruction?: any } {
-    const contents: any[] = [];
-    for (const m of messages) {
-      contents.push({ role: m.role === 'assistant' ? 'model' : 'user', parts: [{ text: m.content }] });
-    }
-    const result: any = { contents };
-    if (systemPrompt) {
-      result.systemInstruction = { parts: [{ text: systemPrompt }] };
-    }
-    return result;
-  }
-
-  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
-    if (!this.apiKey) {
-      const last = messages[messages.length - 1];
-      return `[gemini/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
-    }
-    let effectivePrompt = systemPrompt;
-    if (options?.tools && options.tools.length > 0) {
-      effectivePrompt = (systemPrompt || '') + buildToolPrompt(options.tools);
-    }
-    const body = this.formatContents(messages, effectivePrompt);
-    const url = this.buildUrl(false);
-    const postData = JSON.stringify(body);
-
-    return new Promise((resolve, reject) => {
-      const parsedUrl = new URL(url);
-      const req = https.request({
-        hostname: parsedUrl.hostname,
-        path: parsedUrl.pathname + parsedUrl.search,
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json', 'Content-Length': String(Buffer.byteLength(postData)) },
-      }, (res) => {
-        let data = '';
-        res.on('data', (chunk: Buffer) => (data += chunk.toString()));
-        res.on('end', () => {
-          if (res.statusCode && res.statusCode >= 400) { reject(new Error(`Gemini API error ${res.statusCode}: ${data}`)); return; }
-          try {
-            const parsed = JSON.parse(data);
-            resolve(parsed.candidates?.[0]?.content?.parts?.[0]?.text ?? '');
-          } catch { reject(new Error(`Invalid Gemini response: ${data.slice(0, 200)}`)); }
-        });
-      });
-      req.on('error', reject);
-      req.write(postData);
-      req.end();
-    });
-  }
-
-  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
-    if (!this.apiKey) {
-      const last = messages[messages.length - 1];
-      yield `[gemini/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
-      return;
-    }
-    const body = this.formatContents(messages, systemPrompt);
-    const url = this.buildUrl(true);
-    const postData = JSON.stringify(body);
-    const parsedUrl = new URL(url);
-
-    const response = await new Promise<http.IncomingMessage>((resolve, reject) => {
-      const req = https.request({
-        hostname: parsedUrl.hostname,
-        path: parsedUrl.pathname + parsedUrl.search,
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json', 'Content-Length': String(Buffer.byteLength(postData)) },
-      }, resolve);
-      req.on('error', reject);
-      req.write(postData);
-      req.end();
-    });
-
-    if (response.statusCode && response.statusCode >= 400) {
-      let data = '';
-      for await (const chunk of response) data += chunk.toString();
-      throw new Error(`Gemini API error ${response.statusCode}: ${data}`);
-    }
-
-    let buffer = '';
-    for await (const chunk of response) {
-      buffer += chunk.toString();
-      const lines = buffer.split('\n');
-      buffer = lines.pop() ?? '';
-      for (const line of lines) {
-        const trimmed = line.trim();
-        if (!trimmed.startsWith('data: ')) continue;
-        const data = trimmed.slice(6);
-        if (data === '[DONE]') return;
-        try {
-          const parsed = JSON.parse(data);
-          const text = parsed.candidates?.[0]?.content?.parts?.[0]?.text;
-          if (text) yield text;
-        } catch {}
-      }
-    }
-  }
-}
-
-function isGeminiNative(): boolean {
-  const baseUrl = process.env.OPC_LLM_BASE_URL || '';
-  const key = getApiKey();
-  return key.startsWith('AQ.') || (baseUrl.includes('googleapis.com') && !baseUrl.includes('/openai'));
-}
-
-class ClaudeCLIProvider implements LLMProvider {
-  name = 'claude-cli';
-  private model: string;
-
-  constructor(model?: string) {
-    // Claude CLI uses short model names; don't pass API-style model names
-    // Let Claude CLI use its default model unless explicitly set to a known CLI model
-    const cliModels = ['sonnet', 'opus', 'haiku', 'claude-sonnet-4-20250514', 'claude-opus-4-20250514'];
-    if (model && !cliModels.includes(model)) {
-      // Map common patterns
-      if (model.includes('opus')) this.model = 'opus';
-      else if (model.includes('haiku')) this.model = 'haiku';
-      else this.model = ''; // let CLI choose default
-    } else {
-      this.model = model || '';
-    }
-  }
-
-  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
-    const { writeFileSync, unlinkSync, mkdtempSync } = await import('fs');
-    const { join } = await import('path');
-    const { tmpdir } = await import('os');
-
-    // Build the prompt from messages
-    const lastMessage = messages[messages.length - 1];
-    if (!lastMessage) return '';
-
-    let prompt = lastMessage.content;
-
-    // Add tool prompt if tools provided
-    if (options?.tools && options.tools.length > 0) {
-      prompt += buildToolPrompt(options.tools);
-    }
-
-    const args = ['-p'];
-    // Write system prompt to temp file to avoid shell escaping issues
-    let tmpFile: string | undefined;
-    if (systemPrompt) {
-      const tmpDir = mkdtempSync(join(tmpdir(), 'opc-'));
-      tmpFile = join(tmpDir, 'system.txt');
-      writeFileSync(tmpFile, systemPrompt, 'utf8');
-      args.push('--system-prompt-file', tmpFile);
-    }
-    if (this.model) {
-      args.push('--model', this.model);
-    }
-    args.push(prompt);
-
-    try {
-      const { spawn } = await import('child_process');
-      const result = await new Promise<string>((resolve, reject) => {
-        const proc = spawn('claude', args, {
-          env: { ...process.env },
-          stdio: ['pipe', 'pipe', 'pipe'],
-        });
-        // Close stdin immediately to avoid "no stdin data" warning
-        proc.stdin.end();
-        let stdout = '';
-        let stderr = '';
-        proc.stdout.on('data', (d: Buffer) => { stdout += d.toString(); });
-        proc.stderr.on('data', (d: Buffer) => { stderr += d.toString(); });
-        proc.on('close', (code) => {
-          if (code === 0 || stdout.trim()) {
-            resolve(stdout.trim());
-          } else {
-            reject(new Error(`Claude CLI exited ${code}: ${stderr}`));
-          }
-        });
-        proc.on('error', (err: any) => {
-          if (err.code === 'ENOENT') {
-            reject(new Error(
-              'Claude CLI not found. Install it: npm install -g @anthropic-ai/claude-code\n' +
-              'Then authenticate: claude login'
-            ));
-          } else {
-            reject(err);
-          }
-        });
-        // Timeout
-        setTimeout(() => {
-          proc.kill();
-          reject(new Error('Claude CLI timed out after 120s'));
-        }, 120_000);
-      });
-      return result;
-    } catch (err: any) {
-      throw new Error(`Claude CLI error: ${err.message}`);
-    } finally {
-      if (tmpFile) {
-        try { unlinkSync(tmpFile); } catch {}
-      }
-    }
-  }
-
-  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
-    const args = ['-p', '--verbose', '--output-format', 'stream-json'];
-    if (this.model) {
-      args.push('--model', this.model);
-    }
-
-    // Write system prompt to temp file if needed
-    let tmpFile: string | undefined;
-    if (systemPrompt) {
-      const { writeFileSync } = await import('fs');
-      const { join } = await import('path');
-      const { tmpdir } = await import('os');
-      tmpFile = join(tmpdir(), `opc-claude-stream-${Date.now()}.txt`);
-      writeFileSync(tmpFile, systemPrompt);
-      args.push('--system-prompt-file', tmpFile);
-    }
-
-    const lastMsg = messages[messages.length - 1];
-    args.push(lastMsg?.content ?? '');
-
-    const { spawn } = await import('child_process');
-
-    try {
-      const proc = spawn('claude', args, {
-        env: { ...process.env },
-        stdio: ['pipe', 'pipe', 'pipe'],
-      });
-      proc.stdin.end();
-
-      let buffer = '';
-      let lastContent = '';
-
-      for await (const chunk of proc.stdout) {
-        buffer += (chunk as Buffer).toString();
-        const lines = buffer.split('\n');
-        buffer = lines.pop() ?? '';
-
-        for (const line of lines) {
-          const trimmed = line.trim();
-          if (!trimmed) continue;
-          try {
-            const event = JSON.parse(trimmed);
-            // Claude CLI stream-json format:
-            // {"type":"assistant","message":{"content":[{"type":"text","text":"..."}]}}
-            if (event.type === 'assistant' && event.message?.content) {
-              for (const block of event.message.content) {
-                if (block.type === 'text' && block.text) {
-                  const newContent = block.text;
-                  if (newContent.length > lastContent.length) {
-                    yield newContent.slice(lastContent.length);
-                    lastContent = newContent;
-                  }
-                }
-              }
-            }
-            // Also handle result type for final content
-            if (event.type === 'result' && event.result) {
-              const remaining = event.result.slice(lastContent.length);
-              if (remaining) yield remaining;
-            }
-            // Handle assistant message with content array
-            if (event.type === 'assistant' && event.message?.content) {
-              for (const block of event.message.content) {
-                if (block.type === 'text' && block.text) {
-                  const newText = block.text;
-                  if (newText.length > lastContent.length) {
-                    yield newText.slice(lastContent.length);
-                    lastContent = newText;
-                  }
-                }
-              }
-            }
-            // Handle result message
-            if (event.type === 'result' && event.result) {
-              const resultText = typeof event.result === 'string' ? event.result : '';
-              if (resultText && resultText.length > lastContent.length) {
-                yield resultText.slice(lastContent.length);
-              }
-            }
-          } catch {
-            // Not JSON, might be raw text
-          }
-        }
-      }
-
-      // Process remaining buffer
-      if (buffer.trim()) {
-        try {
-          const event = JSON.parse(buffer.trim());
-          if (event.type === 'result' && event.result) {
-            const resultText = typeof event.result === 'string' ? event.result : '';
-            if (resultText && resultText.length > lastContent.length) {
-              yield resultText.slice(lastContent.length);
-            }
-          }
-        } catch {
-          // If not JSON, yield as raw text if we haven't yielded anything
-          if (!lastContent && buffer.trim()) {
-            yield buffer.trim();
-          }
-        }
-      }
-
-      await new Promise<void>((resolve) => {
-        proc.on('close', () => resolve());
-      });
-    } finally {
-      if (tmpFile) {
-        try { const { unlinkSync } = await import('fs'); unlinkSync(tmpFile); } catch {}
-      }
-    }
-  }
-}
-
-import { execSync } from 'child_process';
-
-function detectClaudeCLI(): boolean {
-  try {
-    execSync('claude --version', { stdio: 'pipe', timeout: 3000 });
-    return true;
-  } catch { return false; }
-}
-
-function detectOllama(): boolean {
-  try {
-    // Use node http instead of curl for Windows compatibility
-    const { execSync: es } = require('child_process');
-    // Quick check: try to connect to Ollama API via node
-    const result = es(
-      `node -e "const h=require('http');const r=h.get('http://localhost:11434/api/tags',{timeout:2000},s=>{let d='';s.on('data',c=>d+=c);s.on('end',()=>{process.stdout.write(d.includes('models')?'1':'0')})});r.on('error',()=>process.stdout.write('0'));r.on('timeout',()=>{r.destroy();process.stdout.write('0')})"`,
-      { stdio: 'pipe', timeout: 5000 }
-    );
-    return result.toString().trim() === '1';
-  } catch { return false; }
-}
-
-function detectApiKeys(): { provider: string; key: string; baseUrl?: string } | null {
-  if (process.env.ANTHROPIC_API_KEY) return { provider: 'anthropic', key: process.env.ANTHROPIC_API_KEY };
-  if (process.env.OPENAI_API_KEY && process.env.OPENAI_API_KEY !== 'your-api-key-here') return { provider: 'openai', key: process.env.OPENAI_API_KEY };
-  if (process.env.DEEPSEEK_API_KEY) return { provider: 'deepseek', key: process.env.DEEPSEEK_API_KEY, baseUrl: 'https://api.deepseek.com/v1' };
-  if (process.env.DASHSCOPE_API_KEY) return { provider: 'qwen', key: process.env.DASHSCOPE_API_KEY, baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1' };
-  if (process.env.GEMINI_API_KEY) return { provider: 'gemini', key: process.env.GEMINI_API_KEY };
-  return null;
-}
-
-export function autoDetectProvider(): { name: string; model?: string; baseUrl?: string; apiKey?: string } {
-  // 1. Claude CLI (zero config, Claude Max/Pro subscription)
-  if (detectClaudeCLI()) {
-    return { name: 'claude-cli' };
-  }
-
-  // 2. API keys from environment
-  const apiKey = detectApiKeys();
-  if (apiKey) {
-    return { name: apiKey.provider, apiKey: apiKey.key, baseUrl: apiKey.baseUrl };
-  }
-
-  // 3. Ollama (local, free)
-  if (detectOllama()) {
-    return { name: 'ollama', model: 'qwen2.5:7b' };
-  }
-
-  // 4. Nothing found
-  return { name: 'none' };
-}
-
-export function createProvider(name: string = 'auto', model?: string, baseUrl?: string, apiKey?: string): LLMProvider {
-  // Auto-detect if name is 'auto' or default openai with no real key
-  const needsAutoDetect = name === 'auto' || (name === 'openai' && !apiKey && (!process.env.OPENAI_API_KEY || process.env.OPENAI_API_KEY === 'your-api-key-here'));
-  if (needsAutoDetect) {
-    const detected = autoDetectProvider();
-    if (detected.name !== 'none') {
-      console.log(`[provider] Auto-detected: ${detected.name}`);
-      name = detected.name;
-      model = model || detected.model;
-      baseUrl = baseUrl || detected.baseUrl;
-      apiKey = apiKey || detected.apiKey;
-    }
-  }
-
-  const finalModel = model || process.env.OPC_LLM_MODEL || 'gpt-4o-mini';
-
-  // Claude CLI mode: use local claude command (Claude Max/Pro subscription)
-  if (name === 'claude-cli' || process.env.OPC_LLM_PROVIDER === 'claude-cli') {
-    return new ClaudeCLIProvider(finalModel !== 'gpt-4o-mini' ? finalModel : undefined);
-  }
-
-  if (name === 'ollama') {
-    const ollamaBase = baseUrl || process.env.OPC_LLM_BASE_URL || 'http://localhost:11434/v1';
-    const ollamaKey = apiKey || process.env.OPC_LLM_API_KEY || 'ollama';
-    return new OpenAICompatibleProvider('ollama', finalModel, ollamaBase, ollamaKey);
-  }
-
-  const finalKey = apiKey || getApiKey();
-  const finalBaseUrl = baseUrl || getBaseUrl();
-
-  if (finalKey.startsWith('AQ.') || isGeminiNative()) {
-    return new GeminiNativeProvider(finalModel, finalKey);
-  }
-
-  let resolvedName = name;
-  if (finalBaseUrl.includes('deepseek.com')) {
-    resolvedName = 'deepseek';
-  } else if (finalBaseUrl.includes('dashscope.aliyuncs.com')) {
-    resolvedName = 'qwen';
-  }
-
-  return new OpenAICompatibleProvider(resolvedName, finalModel, finalBaseUrl, finalKey);
-}
-
-export const SUPPORTED_PROVIDERS = ['openai', 'ollama', 'claude-cli', 'deepseek', 'qwen', 'gemini', 'dashscope', 'zhipu', 'moonshot'] as const;
+import type { Message } from '../core/types';
+import type { MCPToolDefinition } from '../tools/mcp';
+import * as https from 'https';
+import * as http from 'http';
+
+export interface ChatOptions {
+  tools?: MCPToolDefinition[];
+}
+
+export interface LLMProvider {
+  name: string;
+  chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string>;
+  chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string>;
+}
+
+interface OpenAIMessage {
+  role: 'system' | 'user' | 'assistant';
+  content: string;
+}
+
+function getApiKey(): string {
+  return process.env.OPC_LLM_API_KEY || process.env.OPENAI_API_KEY || '';
+}
+
+function getBaseUrl(): string {
+  return process.env.OPC_LLM_BASE_URL || 'https://api.openai.com/v1';
+}
+
+function buildToolPrompt(tools: MCPToolDefinition[]): string {
+  const toolsDesc = tools.map(t =>
+    `- ${t.name}: ${t.description}\n  Input schema: ${JSON.stringify(t.inputSchema)}`
+  ).join('\n');
+  return `\n\nYou have access to the following tools. To use a tool, respond with ONLY a JSON object in this format:\n<tool_call>{"name": "tool_name", "arguments": {...}}</tool_call>\n\nAvailable tools:\n${toolsDesc}\n\nIf you don't need a tool, respond normally with text.`;
+}
+
+class OpenAICompatibleProvider implements LLMProvider {
+  name: string;
+  private model: string;
+  private baseUrl: string;
+  private apiKey: string;
+
+  constructor(name: string, model: string, baseUrl?: string, apiKey?: string) {
+    this.name = name;
+    this.model = model;
+    this.baseUrl = baseUrl || getBaseUrl();
+    this.apiKey = apiKey || getApiKey();
+  }
+
+  private formatMessages(messages: Message[], systemPrompt?: string): OpenAIMessage[] {
+    const formatted: OpenAIMessage[] = [];
+    if (systemPrompt) {
+      formatted.push({ role: 'system', content: systemPrompt });
+    }
+    for (const m of messages) {
+      formatted.push({ role: m.role as 'user' | 'assistant', content: m.content });
+    }
+    return formatted;
+  }
+
+  private async request(body: any): Promise<any> {
+    if (!this.apiKey) {
+      throw new Error('No API key configured. Set OPC_LLM_API_KEY or OPENAI_API_KEY environment variable.');
+    }
+
+    const url = new URL(`${this.baseUrl}/chat/completions`);
+    const isGemini = url.hostname.includes('googleapis.com');
+    if (isGemini) {
+      url.searchParams.set('key', this.apiKey);
+    }
+    const isHttps = url.protocol === 'https:';
+    const lib = isHttps ? https : http;
+
+    const postData = JSON.stringify(body);
+
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+      'Content-Length': String(Buffer.byteLength(postData)),
+    };
+    if (!isGemini) {
+      headers['Authorization'] = `Bearer ${this.apiKey}`;
+    }
+
+    return new Promise((resolve, reject) => {
+      const req = lib.request(
+        {
+          hostname: url.hostname,
+          port: url.port || (isHttps ? 443 : 80),
+          path: url.pathname + url.search,
+          method: 'POST',
+          headers,
+        },
+        (res) => {
+          let data = '';
+          res.on('data', (chunk: Buffer) => (data += chunk.toString()));
+          res.on('end', () => {
+            if (res.statusCode && res.statusCode >= 400) {
+              reject(new Error(`LLM API error ${res.statusCode}: ${data}`));
+              return;
+            }
+            try {
+              resolve(JSON.parse(data));
+            } catch {
+              reject(new Error(`Invalid JSON response: ${data.slice(0, 200)}`));
+            }
+          });
+        },
+      );
+      req.on('error', reject);
+      req.write(postData);
+      req.end();
+    });
+  }
+
+  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
+    if (!this.apiKey) {
+      const last = messages[messages.length - 1];
+      return `[${this.name}/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
+    }
+    let effectivePrompt = systemPrompt;
+    if (options?.tools && options.tools.length > 0) {
+      effectivePrompt = (systemPrompt || '') + buildToolPrompt(options.tools);
+    }
+    const formatted = this.formatMessages(messages, effectivePrompt);
+    const result = await this.request({
+      model: this.model,
+      messages: formatted,
+      temperature: 0.7,
+      max_tokens: 2048,
+    });
+    return result.choices?.[0]?.message?.content ?? '';
+  }
+
+  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
+    if (!this.apiKey) {
+      const last = messages[messages.length - 1];
+      yield `[${this.name}/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
+      return;
+    }
+
+    const formatted = this.formatMessages(messages, systemPrompt);
+    const url = new URL(`${this.baseUrl}/chat/completions`);
+    const isGemini = url.hostname.includes('googleapis.com');
+    if (isGemini) {
+      url.searchParams.set('key', this.apiKey);
+    }
+    const isHttps = url.protocol === 'https:';
+    const lib = isHttps ? https : http;
+    const postData = JSON.stringify({
+      model: this.model,
+      messages: formatted,
+      temperature: 0.7,
+      max_tokens: 2048,
+      stream: true,
+    });
+
+    const streamHeaders: Record<string, string> = {
+      'Content-Type': 'application/json',
+      'Content-Length': String(Buffer.byteLength(postData)),
+    };
+    if (!isGemini) {
+      streamHeaders['Authorization'] = `Bearer ${this.apiKey}`;
+    }
+
+    const response = await new Promise<http.IncomingMessage>((resolve, reject) => {
+      const req = lib.request(
+        {
+          hostname: url.hostname,
+          port: url.port || (isHttps ? 443 : 80),
+          path: url.pathname + url.search,
+          method: 'POST',
+          headers: streamHeaders,
+        },
+        resolve,
+      );
+      req.on('error', reject);
+      req.write(postData);
+      req.end();
+    });
+
+    if (response.statusCode && response.statusCode >= 400) {
+      let data = '';
+      for await (const chunk of response) data += chunk.toString();
+      throw new Error(`LLM API error ${response.statusCode}: ${data}`);
+    }
+
+    let buffer = '';
+    for await (const chunk of response) {
+      buffer += chunk.toString();
+      const lines = buffer.split('\n');
+      buffer = lines.pop() ?? '';
+
+      for (const line of lines) {
+        const trimmed = line.trim();
+        if (!trimmed || !trimmed.startsWith('data: ')) continue;
+        const data = trimmed.slice(6);
+        if (data === '[DONE]') return;
+        try {
+          const parsed = JSON.parse(data);
+          const content = parsed.choices?.[0]?.delta?.content;
+          if (content) yield content;
+        } catch {
+          // skip malformed lines
+        }
+      }
+    }
+  }
+}
+
+class GeminiNativeProvider implements LLMProvider {
+  name = 'gemini';
+  private model: string;
+  private apiKey: string;
+
+  constructor(model: string, apiKey?: string) {
+    this.model = model;
+    this.apiKey = apiKey || getApiKey();
+  }
+
+  private buildUrl(stream: boolean): string {
+    const action = stream ? 'streamGenerateContent?alt=sse&' : 'generateContent?';
+    return `https://generativelanguage.googleapis.com/v1beta/models/${this.model}:${action}key=${this.apiKey}`;
+  }
+
+  private formatContents(messages: Message[], systemPrompt?: string): { contents: any[]; systemInstruction?: any } {
+    const contents: any[] = [];
+    for (const m of messages) {
+      contents.push({ role: m.role === 'assistant' ? 'model' : 'user', parts: [{ text: m.content }] });
+    }
+    const result: any = { contents };
+    if (systemPrompt) {
+      result.systemInstruction = { parts: [{ text: systemPrompt }] };
+    }
+    return result;
+  }
+
+  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
+    if (!this.apiKey) {
+      const last = messages[messages.length - 1];
+      return `[gemini/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
+    }
+    let effectivePrompt = systemPrompt;
+    if (options?.tools && options.tools.length > 0) {
+      effectivePrompt = (systemPrompt || '') + buildToolPrompt(options.tools);
+    }
+    const body = this.formatContents(messages, effectivePrompt);
+    const url = this.buildUrl(false);
+    const postData = JSON.stringify(body);
+
+    return new Promise((resolve, reject) => {
+      const parsedUrl = new URL(url);
+      const req = https.request({
+        hostname: parsedUrl.hostname,
+        path: parsedUrl.pathname + parsedUrl.search,
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json', 'Content-Length': String(Buffer.byteLength(postData)) },
+      }, (res) => {
+        let data = '';
+        res.on('data', (chunk: Buffer) => (data += chunk.toString()));
+        res.on('end', () => {
+          if (res.statusCode && res.statusCode >= 400) { reject(new Error(`Gemini API error ${res.statusCode}: ${data}`)); return; }
+          try {
+            const parsed = JSON.parse(data);
+            resolve(parsed.candidates?.[0]?.content?.parts?.[0]?.text ?? '');
+          } catch { reject(new Error(`Invalid Gemini response: ${data.slice(0, 200)}`)); }
+        });
+      });
+      req.on('error', reject);
+      req.write(postData);
+      req.end();
+    });
+  }
+
+  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
+    if (!this.apiKey) {
+      const last = messages[messages.length - 1];
+      yield `[gemini/${this.model} - no API key] Echo: ${last?.content ?? ''}`;
+      return;
+    }
+    const body = this.formatContents(messages, systemPrompt);
+    const url = this.buildUrl(true);
+    const postData = JSON.stringify(body);
+    const parsedUrl = new URL(url);
+
+    const response = await new Promise<http.IncomingMessage>((resolve, reject) => {
+      const req = https.request({
+        hostname: parsedUrl.hostname,
+        path: parsedUrl.pathname + parsedUrl.search,
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json', 'Content-Length': String(Buffer.byteLength(postData)) },
+      }, resolve);
+      req.on('error', reject);
+      req.write(postData);
+      req.end();
+    });
+
+    if (response.statusCode && response.statusCode >= 400) {
+      let data = '';
+      for await (const chunk of response) data += chunk.toString();
+      throw new Error(`Gemini API error ${response.statusCode}: ${data}`);
+    }
+
+    let buffer = '';
+    for await (const chunk of response) {
+      buffer += chunk.toString();
+      const lines = buffer.split('\n');
+      buffer = lines.pop() ?? '';
+      for (const line of lines) {
+        const trimmed = line.trim();
+        if (!trimmed.startsWith('data: ')) continue;
+        const data = trimmed.slice(6);
+        if (data === '[DONE]') return;
+        try {
+          const parsed = JSON.parse(data);
+          const text = parsed.candidates?.[0]?.content?.parts?.[0]?.text;
+          if (text) yield text;
+        } catch {}
+      }
+    }
+  }
+}
+
+function isGeminiNative(): boolean {
+  const baseUrl = process.env.OPC_LLM_BASE_URL || '';
+  const key = getApiKey();
+  return key.startsWith('AQ.') || (baseUrl.includes('googleapis.com') && !baseUrl.includes('/openai'));
+}
+
+class ClaudeCLIProvider implements LLMProvider {
+  name = 'claude-cli';
+  private model: string;
+
+  constructor(model?: string) {
+    // Claude CLI uses short model names; don't pass API-style model names
+    // Let Claude CLI use its default model unless explicitly set to a known CLI model
+    const cliModels = ['sonnet', 'opus', 'haiku', 'claude-sonnet-4-20250514', 'claude-opus-4-20250514'];
+    if (model && !cliModels.includes(model)) {
+      // Map common patterns
+      if (model.includes('opus')) this.model = 'opus';
+      else if (model.includes('haiku')) this.model = 'haiku';
+      else this.model = ''; // let CLI choose default
+    } else {
+      this.model = model || '';
+    }
+  }
+
+  async chat(messages: Message[], systemPrompt?: string, options?: ChatOptions): Promise<string> {
+    const { writeFileSync, unlinkSync, mkdtempSync } = await import('fs');
+    const { join } = await import('path');
+    const { tmpdir } = await import('os');
+
+    // Build the prompt from messages
+    const lastMessage = messages[messages.length - 1];
+    if (!lastMessage) return '';
+
+    let prompt = lastMessage.content;
+
+    // Add tool prompt if tools provided
+    if (options?.tools && options.tools.length > 0) {
+      prompt += buildToolPrompt(options.tools);
+    }
+
+    const args = ['-p'];
+    // Write system prompt to temp file to avoid shell escaping issues
+    let tmpFile: string | undefined;
+    if (systemPrompt) {
+      const tmpDir = mkdtempSync(join(tmpdir(), 'opc-'));
+      tmpFile = join(tmpDir, 'system.txt');
+      writeFileSync(tmpFile, systemPrompt, 'utf8');
+      args.push('--system-prompt-file', tmpFile);
+    }
+    if (this.model) {
+      args.push('--model', this.model);
+    }
+    args.push(prompt);
+
+    try {
+      const { spawn } = await import('child_process');
+      const result = await new Promise<string>((resolve, reject) => {
+        const proc = spawn('claude', args, {
+          env: { ...process.env },
+          stdio: ['pipe', 'pipe', 'pipe'],
+        });
+        // Close stdin immediately to avoid "no stdin data" warning
+        proc.stdin.end();
+        let stdout = '';
+        let stderr = '';
+        proc.stdout.on('data', (d: Buffer) => { stdout += d.toString(); });
+        proc.stderr.on('data', (d: Buffer) => { stderr += d.toString(); });
+        proc.on('close', (code) => {
+          if (code === 0 || stdout.trim()) {
+            resolve(stdout.trim());
+          } else {
+            reject(new Error(`Claude CLI exited ${code}: ${stderr}`));
+          }
+        });
+        proc.on('error', (err: any) => {
+          if (err.code === 'ENOENT') {
+            reject(new Error(
+              'Claude CLI not found. Install it: npm install -g @anthropic-ai/claude-code\n' +
+              'Then authenticate: claude login'
+            ));
+          } else {
+            reject(err);
+          }
+        });
+        // Timeout
+        setTimeout(() => {
+          proc.kill();
+          reject(new Error('Claude CLI timed out after 120s'));
+        }, 120_000);
+      });
+      return result;
+    } catch (err: any) {
+      throw new Error(`Claude CLI error: ${err.message}`);
+    } finally {
+      if (tmpFile) {
+        try { unlinkSync(tmpFile); } catch {}
+      }
+    }
+  }
+
+  async *chatStream(messages: Message[], systemPrompt?: string): AsyncIterable<string> {
+    const args = ['-p', '--verbose', '--output-format', 'stream-json'];
+    if (this.model) {
+      args.push('--model', this.model);
+    }
+
+    // Write system prompt to temp file if needed
+    let tmpFile: string | undefined;
+    if (systemPrompt) {
+      const { writeFileSync } = await import('fs');
+      const { join } = await import('path');
+      const { tmpdir } = await import('os');
+      tmpFile = join(tmpdir(), `opc-claude-stream-${Date.now()}.txt`);
+      writeFileSync(tmpFile, systemPrompt);
+      args.push('--system-prompt-file', tmpFile);
+    }
+
+    const lastMsg = messages[messages.length - 1];
+    args.push(lastMsg?.content ?? '');
+
+    const { spawn } = await import('child_process');
+
+    try {
+      const proc = spawn('claude', args, {
+        env: { ...process.env },
+        stdio: ['pipe', 'pipe', 'pipe'],
+      });
+      proc.stdin.end();
+
+      let buffer = '';
+      let lastContent = '';
+
+      for await (const chunk of proc.stdout) {
+        buffer += (chunk as Buffer).toString();
+        const lines = buffer.split('\n');
+        buffer = lines.pop() ?? '';
+
+        for (const line of lines) {
+          const trimmed = line.trim();
+          if (!trimmed) continue;
+          try {
+            const event = JSON.parse(trimmed);
+            // Claude CLI stream-json format:
+            // {"type":"assistant","message":{"content":[{"type":"text","text":"..."}]}}
+            if (event.type === 'assistant' && event.message?.content) {
+              for (const block of event.message.content) {
+                if (block.type === 'text' && block.text) {
+                  const newContent = block.text;
+                  if (newContent.length > lastContent.length) {
+                    yield newContent.slice(lastContent.length);
+                    lastContent = newContent;
+                  }
+                }
+              }
+            }
+            // Also handle result type for final content
+            if (event.type === 'result' && event.result) {
+              const remaining = event.result.slice(lastContent.length);
+              if (remaining) yield remaining;
+            }
+            // Handle assistant message with content array
+            if (event.type === 'assistant' && event.message?.content) {
+              for (const block of event.message.content) {
+                if (block.type === 'text' && block.text) {
+                  const newText = block.text;
+                  if (newText.length > lastContent.length) {
+                    yield newText.slice(lastContent.length);
+                    lastContent = newText;
+                  }
+                }
+              }
+            }
+            // Handle result message
+            if (event.type === 'result' && event.result) {
+              const resultText = typeof event.result === 'string' ? event.result : '';
+              if (resultText && resultText.length > lastContent.length) {
+                yield resultText.slice(lastContent.length);
+              }
+            }
+          } catch {
+            // Not JSON, might be raw text
+          }
+        }
+      }
+
+      // Process remaining buffer
+      if (buffer.trim()) {
+        try {
+          const event = JSON.parse(buffer.trim());
+          if (event.type === 'result' && event.result) {
+            const resultText = typeof event.result === 'string' ? event.result : '';
+            if (resultText && resultText.length > lastContent.length) {
+              yield resultText.slice(lastContent.length);
+            }
+          }
+        } catch {
+          // If not JSON, yield as raw text if we haven't yielded anything
+          if (!lastContent && buffer.trim()) {
+            yield buffer.trim();
+          }
+        }
+      }
+
+      await new Promise<void>((resolve) => {
+        proc.on('close', () => resolve());
+      });
+    } finally {
+      if (tmpFile) {
+        try { const { unlinkSync } = await import('fs'); unlinkSync(tmpFile); } catch {}
+      }
+    }
+  }
+}
+
+import { execSync } from 'child_process';
+
+function detectClaudeCLI(): boolean {
+  try {
+    execSync('claude --version', { stdio: 'pipe', timeout: 3000 });
+    return true;
+  } catch { return false; }
+}
+
+function detectOllama(): boolean {
+  try {
+    // Use node http instead of curl for Windows compatibility
+    const { execSync: es } = require('child_process');
+    // Quick check: try to connect to Ollama API via node
+    const result = es(
+      `node -e "const h=require('http');const r=h.get('http://localhost:11434/api/tags',{timeout:2000},s=>{let d='';s.on('data',c=>d+=c);s.on('end',()=>{process.stdout.write(d.includes('models')?'1':'0')})});r.on('error',()=>process.stdout.write('0'));r.on('timeout',()=>{r.destroy();process.stdout.write('0')})"`,
+      { stdio: 'pipe', timeout: 5000 }
+    );
+    return result.toString().trim() === '1';
+  } catch { return false; }
+}
+
+function detectApiKeys(): { provider: string; key: string; baseUrl?: string } | null {
+  if (process.env.ANTHROPIC_API_KEY) return { provider: 'anthropic', key: process.env.ANTHROPIC_API_KEY };
+  if (process.env.OPENAI_API_KEY && process.env.OPENAI_API_KEY !== 'your-api-key-here') return { provider: 'openai', key: process.env.OPENAI_API_KEY };
+  if (process.env.DEEPSEEK_API_KEY) return { provider: 'deepseek', key: process.env.DEEPSEEK_API_KEY, baseUrl: 'https://api.deepseek.com/v1' };
+  if (process.env.DASHSCOPE_API_KEY) return { provider: 'qwen', key: process.env.DASHSCOPE_API_KEY, baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1' };
+  if (process.env.GEMINI_API_KEY) return { provider: 'gemini', key: process.env.GEMINI_API_KEY };
+  return null;
+}
+
+export function autoDetectProvider(): { name: string; model?: string; baseUrl?: string; apiKey?: string } {
+  // 1. Claude CLI (zero config, Claude Max/Pro subscription)
+  if (detectClaudeCLI()) {
+    return { name: 'claude-cli' };
+  }
+
+  // 2. API keys from environment
+  const apiKey = detectApiKeys();
+  if (apiKey) {
+    return { name: apiKey.provider, apiKey: apiKey.key, baseUrl: apiKey.baseUrl };
+  }
+
+  // 3. Ollama (local, free)
+  if (detectOllama()) {
+    return { name: 'ollama', model: 'qwen2.5:7b' };
+  }
+
+  // 4. Nothing found
+  return { name: 'none' };
+}
+
+export function createProvider(name: string = 'auto', model?: string, baseUrl?: string, apiKey?: string): LLMProvider {
+  // Auto-detect if name is 'auto' or default openai with no real key
+  const needsAutoDetect = name === 'auto' || (name === 'openai' && !apiKey && (!process.env.OPENAI_API_KEY || process.env.OPENAI_API_KEY === 'your-api-key-here'));
+  if (needsAutoDetect) {
+    const detected = autoDetectProvider();
+    if (detected.name !== 'none') {
+      console.log(`[provider] Auto-detected: ${detected.name}`);
+      name = detected.name;
+      model = model || detected.model;
+      baseUrl = baseUrl || detected.baseUrl;
+      apiKey = apiKey || detected.apiKey;
+    }
+  }
+
+  const finalModel = model || process.env.OPC_LLM_MODEL || 'gpt-4o-mini';
+
+  // Claude CLI mode: use local claude command (Claude Max/Pro subscription)
+  if (name === 'claude-cli' || process.env.OPC_LLM_PROVIDER === 'claude-cli') {
+    return new ClaudeCLIProvider(finalModel !== 'gpt-4o-mini' ? finalModel : undefined);
+  }
+
+  if (name === 'ollama') {
+    const ollamaBase = baseUrl || process.env.OPC_LLM_BASE_URL || 'http://localhost:11434/v1';
+    const ollamaKey = apiKey || process.env.OPC_LLM_API_KEY || 'ollama';
+    return new OpenAICompatibleProvider('ollama', finalModel, ollamaBase, ollamaKey);
+  }
+
+  const finalKey = apiKey || getApiKey();
+  const finalBaseUrl = baseUrl || getBaseUrl();
+
+  if (finalKey.startsWith('AQ.') || isGeminiNative()) {
+    return new GeminiNativeProvider(finalModel, finalKey);
+  }
+
+  let resolvedName = name;
+  if (finalBaseUrl.includes('deepseek.com')) {
+    resolvedName = 'deepseek';
+  } else if (finalBaseUrl.includes('dashscope.aliyuncs.com')) {
+    resolvedName = 'qwen';
+  }
+
+  return new OpenAICompatibleProvider(resolvedName, finalModel, finalBaseUrl, finalKey);
+}
+
+export const SUPPORTED_PROVIDERS = ['openai', 'ollama', 'claude-cli', 'deepseek', 'qwen', 'gemini', 'dashscope', 'zhipu', 'moonshot'] as const;