From 0d7cfe8061c7add1d97b3acc386e85c3be6a6c16 Mon Sep 17 00:00:00 2001
From: perf3ct <jonfuller2012@gmail.com>
Date: Tue, 15 Apr 2025 17:44:29 +0000
Subject: [PATCH] fix hardcoded values part 2

---
 src/services/llm/pipeline/chat_pipeline.ts    |  7 +++---
 .../pipeline/stages/model_selection_stage.ts  | 24 ++++++++++++-------
 .../semantic_context_extraction_stage.ts      |  3 ++-
 .../pipeline/stages/vector_search_stage.ts    |  5 ++--
 4 files changed, 24 insertions(+), 15 deletions(-)

diff --git a/src/services/llm/pipeline/chat_pipeline.ts b/src/services/llm/pipeline/chat_pipeline.ts
index 12be376a0..0136be93c 100644
--- a/src/services/llm/pipeline/chat_pipeline.ts
+++ b/src/services/llm/pipeline/chat_pipeline.ts
@@ -13,6 +13,7 @@ import toolRegistry from '../tools/tool_registry.js';
 import toolInitializer from '../tools/tool_initializer.js';
 import log from '../../log.js';
 import type { LLMServiceInterface } from '../interfaces/agent_tool_interfaces.js';
+import { SEARCH_CONSTANTS } from '../constants/search_constants.js';
 
 /**
  * Pipeline for managing the entire chat flow
@@ -56,7 +57,7 @@ export class ChatPipeline {
         this.config = {
             enableStreaming: true,
             enableMetrics: true,
-            maxToolCallIterations: 5,
+            maxToolCallIterations: SEARCH_CONSTANTS.TOOL_EXECUTION.MAX_TOOL_CALL_ITERATIONS,
             ...config
         };
 
@@ -194,9 +195,9 @@ export class ChatPipeline {
                 query: userQuery,
                 noteId: input.noteId || 'global',
                 options: {
-                    maxResults: 5, // Can be adjusted
+                    maxResults: SEARCH_CONSTANTS.CONTEXT.MAX_SIMILAR_NOTES,
                     useEnhancedQueries: true,
-                    threshold: 0.6,
+                    threshold: SEARCH_CONSTANTS.VECTOR_SEARCH.DEFAULT_THRESHOLD,
                     llmService: llmService || undefined
                 }
             });
diff --git a/src/services/llm/pipeline/stages/model_selection_stage.ts b/src/services/llm/pipeline/stages/model_selection_stage.ts
index 5245e3a4c..35634c210 100644
--- a/src/services/llm/pipeline/stages/model_selection_stage.ts
+++ b/src/services/llm/pipeline/stages/model_selection_stage.ts
@@ -5,6 +5,7 @@ import type { ModelMetadata } from '../../providers/provider_options.js';
 import log from '../../../log.js';
 import options from '../../../options.js';
 import aiServiceManager from '../../ai_service_manager.js';
+import { SEARCH_CONSTANTS, MODEL_CAPABILITIES } from "../../constants/search_constants.js";
 /**
  * Pipeline stage for selecting the appropriate LLM model
  */
@@ -28,7 +29,7 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
 
         // Start with provided options or create a new object
         const updatedOptions: ChatCompletionOptions = { ...(inputOptions || {}) };
-        
+
         // Preserve the stream option exactly as it was provided, including undefined state
         // This is critical for ensuring the stream option propagates correctly down the pipeline
         log.info(`[ModelSelectionStage] After copy, stream: ${updatedOptions.stream}, type: ${typeof updatedOptions.stream}`);
@@ -154,9 +155,9 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
         }
 
         // Check content length if provided
-        if (contentLength && contentLength > 5000) {
+        if (contentLength && contentLength > SEARCH_CONSTANTS.CONTEXT.CONTENT_LENGTH.MEDIUM_THRESHOLD) {
             // For large content, favor more powerful models
-            queryComplexity = contentLength > 10000 ? 'high' : 'medium';
+            queryComplexity = contentLength > SEARCH_CONSTANTS.CONTEXT.CONTENT_LENGTH.HIGH_THRESHOLD ? 'high' : 'medium';
         }
 
         // Set the model and add provider metadata
@@ -256,7 +257,7 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
 
         // Get the first available provider and its default model
         const defaultProvider = availableProviders[0] as 'openai' | 'anthropic' | 'ollama' | 'local';
-        let defaultModel = 'gpt-3.5-turbo'; // Default fallback
+        let defaultModel = 'gpt-3.5-turbo'; // Use model from our constants
 
         // Set provider metadata
         if (!input.options.providerMetadata) {
@@ -274,17 +275,22 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
      * Get estimated context window for Ollama models
      */
     private getOllamaContextWindow(model: string): number {
+        // Try to find exact matches in MODEL_CAPABILITIES
+        if (model in MODEL_CAPABILITIES) {
+            return MODEL_CAPABILITIES[model as keyof typeof MODEL_CAPABILITIES].contextWindowTokens;
+        }
+
         // Estimate based on model family
         if (model.includes('llama3')) {
-            return 8192;
+            return MODEL_CAPABILITIES['gpt-4'].contextWindowTokens;
         } else if (model.includes('llama2')) {
-            return 4096;
+            return MODEL_CAPABILITIES['default'].contextWindowTokens;
         } else if (model.includes('mistral') || model.includes('mixtral')) {
-            return 8192;
+            return MODEL_CAPABILITIES['gpt-4'].contextWindowTokens;
         } else if (model.includes('gemma')) {
-            return 8192;
+            return MODEL_CAPABILITIES['gpt-4'].contextWindowTokens;
         } else {
-            return 4096; // Default fallback
+            return MODEL_CAPABILITIES['default'].contextWindowTokens;
         }
     }
 }
diff --git a/src/services/llm/pipeline/stages/semantic_context_extraction_stage.ts b/src/services/llm/pipeline/stages/semantic_context_extraction_stage.ts
index d386e38f2..bf2cc8fd7 100644
--- a/src/services/llm/pipeline/stages/semantic_context_extraction_stage.ts
+++ b/src/services/llm/pipeline/stages/semantic_context_extraction_stage.ts
@@ -7,6 +7,7 @@ import contextFormatter from '../../context/modules/context_formatter.js';
 import providerManager from '../../context/modules/provider_manager.js';
 import type { NoteSearchResult } from '../../interfaces/context_interfaces.js';
 import type { Message } from '../../ai_interface.js';
+import { SEARCH_CONSTANTS } from "../../constants/search_constants.js";
 
 /**
  * Pipeline stage for extracting semantic context from notes
@@ -35,7 +36,7 @@ export class SemanticContextExtractionStage extends BasePipelineStage<SemanticCo
                 options: {
                     maxResults,
                     useEnhancedQueries: true,
-                    threshold: 0.6,
+                    threshold: SEARCH_CONSTANTS.VECTOR_SEARCH.DEFAULT_THRESHOLD,
                     llmService: undefined // Let the vectorSearchStage use the default service
                 }
             });
diff --git a/src/services/llm/pipeline/stages/vector_search_stage.ts b/src/services/llm/pipeline/stages/vector_search_stage.ts
index f56fb74e1..9766d55e3 100644
--- a/src/services/llm/pipeline/stages/vector_search_stage.ts
+++ b/src/services/llm/pipeline/stages/vector_search_stage.ts
@@ -9,6 +9,7 @@ import log from '../../../log.js';
 import vectorSearchService from '../../context/services/vector_search_service.js';
 import type { NoteSearchResult } from '../../interfaces/context_interfaces.js';
 import type { LLMServiceInterface } from '../../interfaces/agent_tool_interfaces.js';
+import { SEARCH_CONSTANTS } from '../../constants/search_constants.js';
 
 export interface VectorSearchInput {
   query: string;
@@ -46,8 +47,8 @@ export class VectorSearchStage {
     } = input;
 
     const {
-      maxResults = 10,
-      threshold = 0.6,
+      maxResults = SEARCH_CONSTANTS.VECTOR_SEARCH.DEFAULT_MAX_RESULTS,
+      threshold = SEARCH_CONSTANTS.VECTOR_SEARCH.DEFAULT_THRESHOLD,
       useEnhancedQueries = false,
       llmService = undefined
     } = options;