wogiflow 1.0.0

package/scripts/flow-orchestrate-llm.js

@@ -0,0 +1,459 @@
+#!/usr/bin/env node
+
+/**
+ * Wogi Flow - Orchestrator LLM Clients
+ *
+ * LLM client implementations for the hybrid mode orchestrator.
+ * Supports both local LLMs (Ollama, LM Studio) and cloud providers.
+ *
+ * Extracted from flow-orchestrate.js for modularity.
+ */
+
+const http = require('http');
+const https = require('https');
+
+const {
+  createExecutorFromConfig,
+  MODEL_CAPABILITIES
+} = require('./flow-providers');
+
+// ============================================================
+// Logging Helper
+// ============================================================
+
+const colors = {
+  reset: '\x1b[0m',
+  cyan: '\x1b[36m',
+  dim: '\x1b[2m'
+};
+
+function log(color, ...args) {
+  console.log(colors[color] + args.join(' ') + colors.reset);
+}
+
+// ============================================================
+// Model Defaults
+// ============================================================
+
+// Model-specific context window defaults for popular models
+const MODEL_DEFAULTS = {
+  'qwen/qwen3-coder-30b': { contextWindow: 32768 },
+  'qwen/qwen3-coder': { contextWindow: 32768 },
+  'qwen3-coder': { contextWindow: 32768 },
+  'nvidia/nemotron-3-nano': { contextWindow: 8192 },
+  'nemotron': { contextWindow: 8192 },
+  'meta/llama-3.3-70b': { contextWindow: 131072 },
+  'llama-3.3': { contextWindow: 131072 },
+  'llama-3.1': { contextWindow: 131072 },
+  'deepseek-coder': { contextWindow: 16384 },
+  'codellama': { contextWindow: 16384 },
+  'mistral': { contextWindow: 32768 },
+  'mixtral': { contextWindow: 32768 },
+};
+
+/**
+ * Gets default settings for a model by name
+ * @param {string} modelName - The model name from config
+ * @returns {Object} - Default settings including contextWindow
+ */
+function getModelDefaults(modelName) {
+  if (!modelName) return { contextWindow: 4096 };
+
+  const lowerName = modelName.toLowerCase();
+
+  // Try exact match first
+  if (MODEL_DEFAULTS[modelName]) {
+    return MODEL_DEFAULTS[modelName];
+  }
+
+  // Try partial match
+  for (const [key, defaults] of Object.entries(MODEL_DEFAULTS)) {
+    if (lowerName.includes(key.toLowerCase())) {
+      return defaults;
+    }
+  }
+
+  return { contextWindow: 4096 }; // Conservative fallback
+}
+
+// ============================================================
+// Local LLM Client
+// ============================================================
+
+class LocalLLM {
+  constructor(config) {
+    this.config = config;
+    this.contextWindow = config.contextWindow || null; // Will be auto-detected or use defaults
+    this.modelInfoFetched = false;
+  }
+
+  /**
+   * Fetches model info including context window from the provider.
+   * Called once on first generate() call.
+   *
+   * Priority order:
+   * 1. Config override (hybrid.settings.contextWindow)
+   * 2. Auto-detection from provider API
+   * 3. Model-specific defaults
+   * 4. Conservative fallback (4096)
+   */
+  async fetchModelInfo() {
+    if (this.modelInfoFetched) return;
+    this.modelInfoFetched = true;
+
+    // Priority 1: Config override
+    if (this.config.contextWindow) {
+      this.contextWindow = this.config.contextWindow;
+      log('dim', `   📊 Using configured context window: ${this.contextWindow.toLocaleString()} tokens`);
+      return;
+    }
+
+    // Get model defaults for fallback
+    const modelDefaults = getModelDefaults(this.config.model);
+
+    try {
+      // Priority 2: Auto-detection from provider
+      if (this.config.provider === 'ollama') {
+        const info = await this.ollamaShowModel();
+        if (info.contextLength) {
+          this.contextWindow = info.contextLength;
+          log('dim', `   📊 Model context window (detected): ${this.contextWindow.toLocaleString()} tokens`);
+          return;
+        }
+      } else {
+        // LM Studio / OpenAI-compatible
+        const info = await this.lmStudioGetModelInfo();
+        if (info.contextLength) {
+          this.contextWindow = info.contextLength;
+          log('dim', `   📊 Model context window (detected): ${this.contextWindow.toLocaleString()} tokens`);
+          return;
+        }
+      }
+
+      // Priority 3: Model-specific defaults
+      this.contextWindow = modelDefaults.contextWindow;
+      log('dim', `   📊 Using model default context window: ${this.contextWindow.toLocaleString()} tokens`);
+    } catch (err) {
+      log('dim', `   ⚠️ Could not fetch model info: ${err.message}`);
+      // Priority 3/4: Model-specific defaults or conservative fallback
+      this.contextWindow = modelDefaults.contextWindow;
+      log('dim', `   📊 Using model default context window: ${this.contextWindow.toLocaleString()} tokens`);
+    }
+  }
+
+  /**
+   * Ollama: GET /api/show to get model parameters
+   */
+  async ollamaShowModel() {
+    return new Promise((resolve, reject) => {
+      const url = new URL('/api/show', this.config.endpoint);
+      const postData = JSON.stringify({ name: this.config.model });
+
+      const req = http.request(url, {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          'Content-Length': Buffer.byteLength(postData)
+        },
+        timeout: 10000
+      }, (res) => {
+        let data = '';
+        res.on('data', chunk => data += chunk);
+        res.on('end', () => {
+          try {
+            const parsed = JSON.parse(data);
+            // Ollama returns model_info with context_length or parameters.num_ctx
+            const contextLength =
+              parsed.model_info?.['context_length'] ||
+              parsed.model_info?.context_length ||
+              parsed.parameters?.num_ctx ||
+              parsed.details?.parameter_size && 4096; // fallback
+            resolve({ contextLength: contextLength || 4096 });
+          } catch (err) {
+            reject(new Error('Invalid response from Ollama /api/show'));
+          }
+        });
+      });
+
+      req.on('error', reject);
+      req.on('timeout', () => {
+        req.destroy();
+        reject(new Error('Timeout fetching model info'));
+      });
+
+      req.write(postData);
+      req.end();
+    });
+  }
+
+  /**
+   * LM Studio: GET /v1/models to get model info
+   */
+  async lmStudioGetModelInfo() {
+    return new Promise((resolve, reject) => {
+      const url = new URL('/v1/models', this.config.endpoint);
+      const client = url.protocol === 'https:' ? https : http;
+
+      const req = client.request(url, {
+        method: 'GET',
+        headers: { 'Content-Type': 'application/json' },
+        timeout: 10000
+      }, (res) => {
+        let data = '';
+        res.on('data', chunk => data += chunk);
+        res.on('end', () => {
+          try {
+            const parsed = JSON.parse(data);
+            // Find our model in the list
+            const model = parsed.data?.find(m =>
+              m.id === this.config.model ||
+              m.id?.includes(this.config.model)
+            );
+            // LM Studio may include context_length in model object
+            const contextLength = model?.context_length || model?.max_tokens || 4096;
+            resolve({ contextLength });
+          } catch (err) {
+            reject(new Error('Invalid response from /v1/models'));
+          }
+        });
+      });
+
+      req.on('error', reject);
+      req.on('timeout', () => {
+        req.destroy();
+        reject(new Error('Timeout fetching model info'));
+      });
+
+      req.end();
+    });
+  }
+
+  async generate(prompt) {
+    // Fetch model info on first call
+    await this.fetchModelInfo();
+
+    if (this.config.provider === 'ollama') {
+      return this.ollamaGenerate(prompt);
+    } else {
+      return this.openaiCompatibleGenerate(prompt);
+    }
+  }
+
+  async ollamaGenerate(prompt) {
+    return new Promise((resolve, reject) => {
+      const url = new URL('/api/generate', this.config.endpoint);
+      const postData = JSON.stringify({
+        model: this.config.model,
+        prompt: prompt,
+        stream: false,
+        options: {
+          temperature: this.config.temperature,
+          num_predict: this.config.maxTokens
+        }
+      });
+
+      const req = http.request(url, {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          'Content-Length': Buffer.byteLength(postData)
+        },
+        timeout: this.config.timeout
+      }, (res) => {
+        let data = '';
+        res.on('data', chunk => data += chunk);
+        res.on('end', () => {
+          try {
+            const parsed = JSON.parse(data);
+            resolve(parsed.response || '');
+          } catch (err) {
+            reject(new Error('Invalid response from Ollama'));
+          }
+        });
+      });
+
+      req.on('error', reject);
+      req.on('timeout', () => {
+        req.destroy();
+        reject(new Error('Request timeout'));
+      });
+
+      req.write(postData);
+      req.end();
+    });
+  }
+
+  async openaiCompatibleGenerate(prompt) {
+    return new Promise((resolve, reject) => {
+      const url = new URL('/v1/chat/completions', this.config.endpoint);
+      const postData = JSON.stringify({
+        model: this.config.model,
+        messages: [{ role: 'user', content: prompt }],
+        temperature: this.config.temperature,
+        max_tokens: this.config.maxTokens
+      });
+
+      const client = url.protocol === 'https:' ? https : http;
+
+      const req = client.request(url, {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          'Content-Length': Buffer.byteLength(postData)
+        },
+        timeout: this.config.timeout
+      }, (res) => {
+        let data = '';
+        res.on('data', chunk => data += chunk);
+        res.on('end', () => {
+          try {
+            const parsed = JSON.parse(data);
+            resolve(parsed.choices?.[0]?.message?.content || '');
+          } catch (err) {
+            reject(new Error('Invalid response from LLM'));
+          }
+        });
+      });
+
+      req.on('error', reject);
+      req.on('timeout', () => {
+        req.destroy();
+        reject(new Error('Request timeout'));
+      });
+
+      req.write(postData);
+      req.end();
+    });
+  }
+}
+
+// ============================================================
+// Cloud Executor Client
+// ============================================================
+
+/**
+ * CloudExecutor wraps cloud providers from flow-providers.js
+ * and exposes the same interface as LocalLLM (generate, contextWindow)
+ * for seamless integration with the Orchestrator.
+ */
+class CloudExecutor {
+  constructor(config) {
+    this.config = config;
+    this.provider = createExecutorFromConfig({ executor: config });
+    this.contextWindow = null;
+    this.modelInfoFetched = false;
+
+    if (!this.provider) {
+      throw new Error(`Failed to create cloud executor for provider: ${config.provider}`);
+    }
+
+    log('cyan', `   ☁️  Cloud executor: ${config.provider} / ${config.model}`);
+  }
+
+  /**
+   * Fetches model info including context window from MODEL_CAPABILITIES.
+   * Called once on first generate() call.
+   */
+  async fetchModelInfo() {
+    if (this.modelInfoFetched) return;
+    this.modelInfoFetched = true;
+
+    // Priority 1: Config override
+    if (this.config.contextWindow) {
+      this.contextWindow = this.config.contextWindow;
+      log('dim', `   📊 Using configured context window: ${this.contextWindow.toLocaleString()} tokens`);
+      return;
+    }
+
+    // Priority 2: Look up in MODEL_CAPABILITIES
+    const modelName = this.config.model || '';
+    const lowerModel = modelName.toLowerCase();
+
+    // Try exact match first
+    if (MODEL_CAPABILITIES[modelName]) {
+      this.contextWindow = MODEL_CAPABILITIES[modelName].contextWindow;
+      log('dim', `   📊 Model context window: ${this.contextWindow.toLocaleString()} tokens`);
+      return;
+    }
+
+    // Try partial match
+    for (const [key, caps] of Object.entries(MODEL_CAPABILITIES)) {
+      if (lowerModel.includes(key.toLowerCase()) || key.toLowerCase().includes(lowerModel)) {
+        this.contextWindow = caps.contextWindow;
+        log('dim', `   📊 Model context window (matched ${key}): ${this.contextWindow.toLocaleString()} tokens`);
+        return;
+      }
+    }
+
+    // Priority 3: Provider-specific defaults
+    const providerDefaults = {
+      'openai': 128000,    // GPT-4o-mini
+      'anthropic': 200000, // Claude Haiku
+      'google': 1000000    // Gemini Flash
+    };
+
+    this.contextWindow = providerDefaults[this.config.provider] || 128000;
+    log('dim', `   📊 Using provider default context window: ${this.contextWindow.toLocaleString()} tokens`);
+  }
+
+  /**
+   * Generate a response from the cloud LLM.
+   * Matches the LocalLLM interface.
+   */
+  async generate(prompt) {
+    // Fetch model info on first call
+    await this.fetchModelInfo();
+
+    try {
+      const response = await this.provider.complete(prompt, {
+        model: this.config.model,
+        temperature: this.config.temperature,
+        maxTokens: this.config.maxTokens
+      });
+      return response;
+    } catch (error) {
+      // Enhance error message with cloud-specific context
+      const enhancedError = new Error(
+        `Cloud executor error (${this.config.provider}/${this.config.model}): ${error.message}`
+      );
+      enhancedError.originalError = error;
+      throw enhancedError;
+    }
+  }
+}
+
+// ============================================================
+// Factory Function
+// ============================================================
+
+/**
+ * Factory function to create the appropriate executor based on config.
+ * Returns either LocalLLM or CloudExecutor.
+ */
+function createExecutor(config) {
+  const executorType = config.executorType || 'local';
+
+  if (executorType === 'cloud') {
+    // Validate cloud config
+    const cloudProviders = ['openai', 'anthropic', 'google'];
+    if (!cloudProviders.includes(config.provider)) {
+      throw new Error(
+        `Invalid cloud provider: ${config.provider}. ` +
+        `Supported: ${cloudProviders.join(', ')}`
+      );
+    }
+
+    return new CloudExecutor(config);
+  }
+
+  // Default to local LLM (ollama, lm-studio)
+  return new LocalLLM(config);
+}
+
+module.exports = {
+  LocalLLM,
+  CloudExecutor,
+  createExecutor,
+  getModelDefaults,
+  MODEL_DEFAULTS
+};