@vfarcic/dot-ai 0.124.0 → 0.125.0

package/dist/core/providers/vercel-provider.js

@@ -14,6 +14,7 @@ const anthropic_1 = require("@ai-sdk/anthropic");
 const xai_1 = require("@ai-sdk/xai");
 const mistral_1 = require("@ai-sdk/mistral");
 const deepseek_1 = require("@ai-sdk/deepseek");
+const ai_sdk_provider_1 = require("@openrouter/ai-sdk-provider");
 const provider_debug_utils_1 = require("./provider-debug-utils");
 const model_config_1 = require("../model-config");
 // Get all supported provider keys dynamically from CURRENT_MODELS
@@ -23,12 +24,14 @@ class VercelProvider {
     model;
     apiKey;
     debugMode;
+    baseURL; // PRD #194: Custom endpoint URL for OpenAI-compatible APIs
     modelInstance; // Vercel AI SDK model instance
     constructor(config) {
         this.apiKey = config.apiKey;
         this.providerType = config.provider;
         this.model = config.model || this.getDefaultModel();
         this.debugMode = config.debugMode ?? (process.env.DEBUG_DOT_AI === 'true');
+        this.baseURL = config.baseURL; // PRD #194: Store custom endpoint URL
         this.validateConfiguration();
         this.initializeModel();
     }
@@ -46,7 +49,9 @@ class VercelProvider {
             switch (this.providerType) {
                 case 'openai':
                 case 'openai_pro':
-                    provider = (0, openai_1.createOpenAI)({ apiKey: this.apiKey });
+                    provider = (0, openai_1.createOpenAI)({
+                        apiKey: this.apiKey
+                    });
                     break;
                 case 'google':
                 case 'google_fast':
@@ -71,7 +76,27 @@ class VercelProvider {
                     provider = (0, mistral_1.createMistral)({ apiKey: this.apiKey });
                     break;
                 case 'deepseek':
-                    provider = (0, deepseek_1.createDeepSeek)({ apiKey: this.apiKey });
+                    provider = (0, deepseek_1.createDeepSeek)({
+                        apiKey: this.apiKey
+                    });
+                    break;
+                case 'openrouter':
+                    // PRD #194: OpenRouter custom endpoint support
+                    // Use dedicated OpenRouter provider for proper format conversion
+                    provider = (0, ai_sdk_provider_1.createOpenRouter)({
+                        apiKey: this.apiKey
+                    });
+                    break;
+                case 'custom':
+                    // PRD #194: Generic custom endpoint support for OpenAI-compatible APIs
+                    // For non-OpenRouter custom endpoints (Ollama, vLLM, LiteLLM, etc.)
+                    if (!this.baseURL) {
+                        throw new Error('Custom endpoint requires CUSTOM_LLM_BASE_URL to be set');
+                    }
+                    provider = (0, openai_1.createOpenAI)({
+                        apiKey: this.apiKey,
+                        baseURL: this.baseURL
+                    });
                     break;
                 default:
                     throw new Error(`Cannot initialize model for provider: ${this.providerType}`);
@@ -115,11 +140,10 @@ class VercelProvider {
         const startTime = Date.now();
         try {
             // Use Vercel AI SDK generateText
-            // Set maxOutputTokens to 8192 for better support of comprehensive responses
+            // Note: maxOutputTokens not specified - provider will use model's natural maximum
             const result = await (0, ai_1.generateText)({
                 model: this.modelInstance,
                 prompt: message,
-                maxOutputTokens: 8192, // Increased from default 4096 to support longer responses
             });
             const response = {
                 content: result.text,
@@ -165,6 +189,11 @@ class VercelProvider {
             return response;
         }
         catch (error) {
+            // Log the prompt that caused the error for debugging
+            if (this.debugMode) {
+                const debugId = (0, provider_debug_utils_1.generateDebugId)(operation);
+                (0, provider_debug_utils_1.debugLogPromptOnly)(debugId, message, operation, this.getProviderType(), this.model, this.debugMode);
+            }
             // Generate dataset for failed AI interaction
             if (this.debugMode && evaluationContext) {
                 const failureMetrics = {
@@ -284,12 +313,12 @@ class VercelProvider {
         try {
             // Use Vercel AI SDK's generateText with stopWhen for automatic loop
             // Default is stepCountIs(1) - we need to increase for multi-step investigation
+            // Note: maxOutputTokens not specified - provider will use model's natural maximum
             const generateConfig = {
                 model: this.modelInstance,
                 messages,
                 tools,
-                stopWhen: (0, ai_1.stepCountIs)(maxIterations),
-                maxOutputTokens: 8192 // Increased from default 4096 to support longer responses
+                stopWhen: (0, ai_1.stepCountIs)(maxIterations)
             };
             // Add system parameter for non-Anthropic providers
             if (systemParam) {

package/dist/core/unified-creation-session.js

@@ -833,7 +833,7 @@ Please try again or modify your policy description.`,
         const capabilityService = new capability_vector_service_1.CapabilityVectorService(collection);
         // Use existing searchCapabilities function - no fallback, let it throw if it fails
         const searchResults = await capabilityService.searchCapabilities(searchQuery, {
-            limit: 50 // Higher limit to get more relevant resources - aligns with recommendation tool
+            limit: 30 // Reduced from 50 to stay within 200K token context limit (testing shows 35 still exceeds)
         });
         if (searchResults.length === 0) {
             throw new Error(`No relevant capabilities found for policy description: "${policyDescription}"`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vfarcic/dot-ai",
-  "version": "0.124.0",
+  "version": "0.125.0",
   "description": "AI-powered development productivity platform that enhances software development workflows through intelligent automation and AI-driven assistance",
   "mcpName": "io.github.vfarcic/dot-ai",
   "main": "dist/index.js",
@@ -28,6 +28,7 @@
     "test:integration:grok-fast": "AI_PROVIDER=xai_fast AI_PROVIDER_SDK=vercel DEBUG_DOT_AI=true ./tests/integration/infrastructure/run-integration-tests.sh",
     "test:integration:mistral": "AI_PROVIDER=mistral AI_PROVIDER_SDK=vercel DEBUG_DOT_AI=true ./tests/integration/infrastructure/run-integration-tests.sh",
     "test:integration:deepseek": "AI_PROVIDER=deepseek AI_PROVIDER_SDK=vercel DEBUG_DOT_AI=true ./tests/integration/infrastructure/run-integration-tests.sh",
+    "test:integration:custom-endpoint": "AI_PROVIDER=openai AI_PROVIDER_SDK=vercel DEBUG_DOT_AI=true ./tests/integration/infrastructure/run-integration-tests.sh",
     "eval:comparative": "DEBUG_DOT_AI=true npx tsx src/evaluation/eval-runner.ts",
     "eval:platform-synthesis": "DEBUG_DOT_AI=true npx tsx src/evaluation/run-platform-synthesis.ts",
     "clean": "rm -rf dist",
@@ -104,6 +105,7 @@
     "@anthropic-ai/sdk": "^0.65.0",
     "@kubernetes/client-node": "^1.3.0",
     "@modelcontextprotocol/sdk": "^1.13.2",
+    "@openrouter/ai-sdk-provider": "^1.2.0",
     "@qdrant/js-client-rest": "^1.15.0",
     "ai": "^5.0.60",
     "glob": "^11.0.3",