feat(llm): really try to coax ollama to run tools

apps/server/src/services/llm/constants/llm_prompt_constants.ts

@@ -213,7 +213,23 @@ Be concise and informative in your responses.
 
 ${context}
 
-Based on this information, please answer: <query>${query}</query>`
+Based on this information, please answer: <query>${query}</query>`,
+
+        // Tool instructions for Ollama
+        TOOL_INSTRUCTIONS: `
+CRITICAL INSTRUCTIONS FOR TOOL USAGE:
+1. YOU MUST TRY MULTIPLE TOOLS AND SEARCH VARIATIONS before concluding information isn't available
+2. ALWAYS PERFORM AT LEAST 3 DIFFERENT SEARCHES with different parameters before giving up on finding information
+3. If a search returns no results, IMMEDIATELY TRY ANOTHER SEARCH with different parameters:
+   - Use broader terms: If "Kubernetes deployment" fails, try just "Kubernetes" or "container orchestration"
+   - Try synonyms: If "meeting notes" fails, try "conference", "discussion", or "conversation"
+   - Remove specific qualifiers: If "quarterly financial report 2024" fails, try just "financial report"
+   - Try semantic variations: If keyword_search fails, use vector_search which finds conceptually related content
+4. CHAIN TOOLS TOGETHER: Use the results of one tool to inform parameters for the next tool
+5. NEVER respond with "there are no notes about X" until you've tried at least 3 different search variations
+6. DO NOT ask the user what to do next when searches fail - AUTOMATICALLY try different approaches
+7. ALWAYS EXPLAIN what you're doing: "I didn't find results for X, so I'm now searching for Y instead"
+8. If all reasonable search variations fail (minimum 3 attempts), THEN you may inform the user that the information might not be in their notes`
     },
 
     // Common prompts across providers

apps/server/src/services/llm/formatters/ollama_formatter.ts

@@ -1,7 +1,7 @@
 import type { Message } from '../ai_interface.js';
 import { BaseMessageFormatter } from './base_formatter.js';
 import sanitizeHtml from 'sanitize-html';
-import { PROVIDER_PROMPTS, FORMATTING_PROMPTS } from '../constants/llm_prompt_constants.js';
+import { PROVIDER_PROMPTS } from '../constants/llm_prompt_constants.js';
 import { LLM_CONSTANTS } from '../constants/provider_constants.js';
 import {
     HTML_ALLOWED_TAGS,
@@ -29,7 +29,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
      * @param context Optional context to include
      * @param preserveSystemPrompt When true, preserves existing system messages rather than replacing them
      */
-    formatMessages(messages: Message[], systemPrompt?: string, context?: string, preserveSystemPrompt?: boolean): Message[] {
+    formatMessages(messages: Message[], systemPrompt?: string, context?: string, preserveSystemPrompt?: boolean, useTools?: boolean): Message[] {
         const formattedMessages: Message[] = [];
 
         // Log the input messages with all their properties
@@ -37,7 +37,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
         messages.forEach((msg, index) => {
             const msgKeys = Object.keys(msg);
             log.info(`Message ${index} - role: ${msg.role}, keys: ${msgKeys.join(', ')}, content length: ${msg.content.length}`);
-            
+
             // Log special properties if present
             if (msg.tool_calls) {
                 log.info(`Message ${index} has ${msg.tool_calls.length} tool_calls`);
@@ -61,7 +61,19 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
             log.info(`Preserving existing system message: ${systemMessages[0].content.substring(0, 50)}...`);
         } else {
             // Use provided systemPrompt or default
-            const basePrompt = systemPrompt || PROVIDER_PROMPTS.COMMON.DEFAULT_ASSISTANT_INTRO;
+            let basePrompt = systemPrompt || PROVIDER_PROMPTS.COMMON.DEFAULT_ASSISTANT_INTRO;
+
+            // Check if any message has tool_calls or if useTools flag is set, indicating this is a tool-using conversation
+            const hasPreviousToolCalls = messages.some(msg => msg.tool_calls && msg.tool_calls.length > 0);
+            const hasToolResults = messages.some(msg => msg.role === 'tool');
+            const isToolUsingConversation = useTools || hasPreviousToolCalls || hasToolResults;
+
+            // Add tool instructions for Ollama when tools are being used
+            if (isToolUsingConversation && PROVIDER_PROMPTS.OLLAMA.TOOL_INSTRUCTIONS) {
+                log.info('Adding tool instructions to system prompt for Ollama');
+                basePrompt = `${basePrompt}\n\n${PROVIDER_PROMPTS.OLLAMA.TOOL_INSTRUCTIONS}`;
+            }
+
             formattedMessages.push({
                 role: 'system',
                 content: basePrompt
@@ -96,7 +108,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
                         ...msg, // Copy all properties
                         content: formattedContext // Override content with injected context
                     };
-                    
+
                     formattedMessages.push(newMessage);
                     log.info(`Created user message with context, final keys: ${Object.keys(newMessage).join(', ')}`);
 
@@ -104,7 +116,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
                 } else {
                     // For other messages, preserve all properties including any tool-related ones
                     log.info(`Preserving message with role ${msg.role}, keys: ${Object.keys(msg).join(', ')}`);
-                    
+
                     formattedMessages.push({
                         ...msg // Copy all properties
                     });
@@ -126,7 +138,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
         formattedMessages.forEach((msg, index) => {
             const msgKeys = Object.keys(msg);
             log.info(`Formatted message ${index} - role: ${msg.role}, keys: ${msgKeys.join(', ')}, content length: ${msg.content.length}`);
-            
+
             // Log special properties if present
             if (msg.tool_calls) {
                 log.info(`Formatted message ${index} has ${msg.tool_calls.length} tool_calls`);
@@ -151,13 +163,11 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
         if (!content) return '';
 
         try {
-            // Store our XML tags so we can restore them after cleaning
+            // Define regexes for identifying and preserving tagged content
-            const noteTagsRegex = /<\/?note>/g;
             const notesTagsRegex = /<\/?notes>/g;
-            const queryTagsRegex = /<\/?query>[^<]*<\/query>/g;
+            // const queryTagsRegex = /<\/?query>/g; // Commenting out unused variable
 
             // Capture tags to restore later
-            const noteTags = content.match(noteTagsRegex) || [];
             const noteTagPositions: number[] = [];
             let match;
             const regex = /<\/?note>/g;
@@ -166,17 +176,15 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
             }
 
             // Remember the notes tags
-            const notesTagsMatch = content.match(notesTagsRegex) || [];
             const notesTagPositions: number[] = [];
             while ((match = notesTagsRegex.exec(content)) !== null) {
                 notesTagPositions.push(match.index);
             }
 
-            // Remember the query tags
+            // Remember the query tag
-            const queryTagsMatch = content.match(queryTagsRegex) || [];
 
             // Temporarily replace XML tags with markers that won't be affected by sanitization
-            let modified = content
+            const modified = content
                 .replace(/<note>/g, '[NOTE_START]')
                 .replace(/<\/note>/g, '[NOTE_END]')
                 .replace(/<notes>/g, '[NOTES_START]')
@@ -184,7 +192,7 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
                 .replace(/<query>(.*?)<\/query>/g, '[QUERY]$1[/QUERY]');
 
             // First use the parent class to do standard cleaning
-            let sanitized = super.cleanContextContent(modified);
+            const sanitized = super.cleanContextContent(modified);
 
             // Then apply Ollama-specific aggressive cleaning
             // Remove any remaining HTML using sanitizeHtml while keeping our markers

apps/server/src/services/llm/providers/ollama_service.ts

@@ -144,14 +144,19 @@ export class OllamaService extends BaseAIService {
                 messagesToSend = [...messages];
                 log.info(`Bypassing formatter for Ollama request with ${messages.length} messages`);
             } else {
+                // Determine if tools will be used in this request
+                const willUseTools = providerOptions.enableTools !== false;
+                
                 // Use the formatter to prepare messages
                 messagesToSend = this.formatter.formatMessages(
                     messages,
                     systemPrompt,
                     undefined, // context
-                    providerOptions.preserveSystemPrompt
+                    providerOptions.preserveSystemPrompt,
+                    willUseTools // Pass flag indicating if tools will be used
                 );
-                log.info(`Sending to Ollama with formatted messages: ${messagesToSend.length}`);
+                
+                log.info(`Sending to Ollama with formatted messages: ${messagesToSend.length}${willUseTools ? ' (with tool instructions)' : ''}`);
             }
 
             // Get tools if enabled