Tools/Notes
1
0
Fork 0
mirror of https://github.com/TriliumNext/Notes.git synced 2025-07-29 19:12:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

tool calling is close to working

Browse Source 
getting closer to calling tools...

we definitely need this

closer to tool execution...

agentic tool calling is...kind of working?
This commit is contained in:
perf3ct 2025-04-06 20:50:08 +00:00
parent eb353df010
commit 26b1b08129
No known key found for this signature in database
GPG Key ID: 569C4EEC436F5232
23 changed files with 1826 additions and 133 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/services/llm/agent_tools/query_decomposition_tool.ts
View File

@ -10,11 +10,16 @@
* - Extract multiple intents from a single question
* - Create a multi-stage research plan
* - Track progress through complex information gathering
*
* Integration with pipeline architecture:
* - Can use pipeline stages when available
* - Falls back to direct methods when needed
*/
import log from '../../log.js';
import { AGENT_TOOL_PROMPTS } from '../constants/llm_prompt_constants.js';
import { QUERY_DECOMPOSITION_STRINGS } from '../constants/query_decomposition_constants.js';
import aiServiceManager from '../ai_service_manager.js';
export interface SubQuery {
id: string;

102
src/services/llm/agent_tools/vector_search_tool.ts
View File

@ -13,14 +13,9 @@
*/
import log from '../../log.js';
import { VectorSearchStage } from '../pipeline/stages/vector_search_stage.js';
import type { ContextService } from '../context/modules/context_service.js';
// Define interface for context service to avoid circular imports
interface IContextService {
findRelevantNotesMultiQuery(queries: string[], contextNoteId: string | null, limit: number): Promise<any[]>;
processQuery(userQuestion: string, llmService: any, contextNoteId: string | null, showThinking: boolean): Promise<any>;
}
export interface VectorSearchResult {
noteId: string;
title: string;
@ -56,17 +51,20 @@ export interface VectorSearchOptions {
}
export class VectorSearchTool {
private contextService: IContextService | null = null;
private contextService: any = null;
private maxResults: number = 5;
private vectorSearchStage: VectorSearchStage;
constructor() {
// Initialization is done by setting context service
// Initialize the vector search stage
this.vectorSearchStage = new VectorSearchStage();
log.info('VectorSearchTool initialized with VectorSearchStage pipeline component');
}
/**
* Set the context service for performing vector searches
*/
setContextService(contextService: IContextService): void {
setContextService(contextService: any): void {
this.contextService = contextService;
log.info('Context service set in VectorSearchTool');
}
@ -79,49 +77,42 @@ export class VectorSearchTool {
contextNoteId?: string,
searchOptions: VectorSearchOptions = {}
): Promise<VectorSearchResult[]> {
if (!this.contextService) {
throw new Error("Context service not set, call setContextService() first");
}
try {
// Set more aggressive defaults to return more content
const options = {
limit: searchOptions.limit || 15, // Increased from default (likely 5 or 10)
threshold: searchOptions.threshold || 0.5, // Lower threshold to include more results (likely 0.65 or 0.7 before)
maxResults: searchOptions.limit || 15, // Increased from default
threshold: searchOptions.threshold || 0.5, // Lower threshold to include more results
useEnhancedQueries: true, // Enable query enhancement by default
includeContent: searchOptions.includeContent !== undefined ? searchOptions.includeContent : true,
...searchOptions
};
log.info(`Vector search: "${query.substring(0, 50)}..." with limit=${options.limit}, threshold=${options.threshold}`);
log.info(`Vector search: "${query.substring(0, 50)}..." with limit=${options.maxResults}, threshold=${options.threshold}`);
// Check if contextService is set again to satisfy TypeScript
if (!this.contextService) {
throw new Error("Context service not set, call setContextService() first");
}
// Use the pipeline stage for vector search
const result = await this.vectorSearchStage.execute({
query,
noteId: contextNoteId || null,
options: {
maxResults: options.maxResults,
threshold: options.threshold,
useEnhancedQueries: options.useEnhancedQueries
}
});
// Use contextService methods instead of direct imports
const results = await this.contextService.findRelevantNotesMultiQuery(
[query],
contextNoteId || null,
options.limit
);
const searchResults = result.searchResults;
log.info(`Vector search found ${searchResults.length} relevant notes via pipeline`);
// Log the number of results
log.info(`Vector search found ${results.length} relevant notes`);
// Include more content from each note to provide richer context
// If includeContent is true but we're missing content for some notes, fetch it
if (options.includeContent) {
// IMPORTANT: Get content directly without recursive processQuery calls
// This prevents infinite loops where one search triggers another
for (let i = 0; i < results.length; i++) {
const result = results[i];
for (let i = 0; i < searchResults.length; i++) {
const result = searchResults[i];
try {
// Get content directly from note content service
// Get content if missing
if (!result.content) {
const noteContent = await import('../context/note_content.js');
const content = await noteContent.getNoteContent(result.noteId);
if (content) {
// Add content directly without recursive calls
result.content = content.substring(0, 2000); // Limit to 2000 chars
log.info(`Added direct content for note ${result.noteId}, length: ${result.content.length} chars`);
}
@ -132,7 +123,18 @@ export class VectorSearchTool {
}
}
return results;
// Format results to match the expected VectorSearchResult interface
return searchResults.map(note => ({
noteId: note.noteId,
title: note.title,
contentPreview: note.content
? note.content.length > 200
? note.content.substring(0, 200) + '...'
: note.content
: 'No content available',
similarity: note.similarity,
parentId: note.parentId
}));
} catch (error) {
log.error(`Vector search error: ${error}`);
return [];
@ -148,26 +150,24 @@ export class VectorSearchTool {
similarityThreshold?: number
} = {}): Promise<VectorSearchResult[]> {
try {
// Validate contextService is set
if (!this.contextService) {
log.error('Context service not set in VectorSearchTool');
return [];
}
// Set defaults
const maxResults = options.maxResults || this.maxResults;
const threshold = options.similarityThreshold || 0.6;
const parentNoteId = options.parentNoteId || null;
// Use multi-query approach for more robust results
const queries = [query];
const results = await this.contextService.findRelevantNotesMultiQuery(
queries,
parentNoteId,
maxResults
);
// Use the pipeline for consistent search behavior
const result = await this.vectorSearchStage.execute({
query,
noteId: parentNoteId,
options: {
maxResults,
threshold,
useEnhancedQueries: true
}
});
// Format results to match the expected interface
return results.map(result => ({
return result.searchResults.map(result => ({
noteId: result.noteId,
title: result.title,
contentPreview: result.content ?

11
src/services/llm/ai_interface.ts
View File

@ -1,6 +1,11 @@
import type { ToolCall } from './tools/tool_interfaces.js';
export interface Message {
role: 'user' | 'assistant' | 'system';
role: 'user' | 'assistant' | 'system' | 'tool';
content: string;
name?: string;
tool_call_id?: string;
tool_calls?: ToolCall[] | any[];
}
// Interface for streaming response chunks
@ -27,6 +32,8 @@ export interface ChatCompletionOptions {
bypassFormatter?: boolean; // Whether to bypass the message formatter entirely
expectsJsonResponse?: boolean; // Whether this request expects a JSON response
stream?: boolean; // Whether to stream the response
enableTools?: boolean; // Whether to enable tool calling
tools?: any[]; // Tools to provide to the LLM
}
export interface ChatResponse {
@ -40,6 +47,8 @@ export interface ChatResponse {
};
// Stream handler - only present when streaming is enabled
stream?: (callback: (chunk: StreamChunk) => Promise<void> | void) => Promise<string>;
// Tool calls from the LLM
tool_calls?: ToolCall[] | any[];
}
export interface AIService {

5
src/services/llm/ai_service_manager.ts
View File

@ -393,6 +393,11 @@ export class AIServiceManager implements IAIServiceManager {
// Initialize agent tools with this service manager instance
await agentTools.initialize(this);
// Initialize LLM tools - this is the single place where tools are initialized
const toolInitializer = await import('./tools/tool_initializer.js');
await toolInitializer.default.initializeTools();
log.info("LLM tools initialized successfully");
this.initialized = true;
log.info("AI service initialized successfully");
} catch (error: any) {

63
src/services/llm/context/index.ts
View File

@ -561,6 +561,69 @@ export class ContextExtractor {
return ContextExtractor.getFullContext(noteId);
}
/**
* Get note hierarchy information in a formatted string
* @param noteId - The ID of the note to get hierarchy information for
* @returns Formatted string with note hierarchy information
*/
static async getNoteHierarchyInfo(noteId: string): Promise<string> {
const note = becca.getNote(noteId);
if (!note) return 'Note not found';
let info = `**Title**: ${note.title}\n`;
// Add attributes if any
const attributes = note.getAttributes();
if (attributes && attributes.length > 0) {
const relevantAttrs = attributes.filter(attr => !attr.name.startsWith('_'));
if (relevantAttrs.length > 0) {
info += `**Attributes**: ${relevantAttrs.map(attr => `${attr.name}=${attr.value}`).join(', ')}\n`;
}
}
// Add parent path
const parents = await ContextExtractor.getParentNotes(noteId);
if (parents && parents.length > 0) {
const path = parents.map(p => p.title).join(' > ');
info += `**Path**: ${path}\n`;
}
// Add child count
const childNotes = note.getChildNotes();
if (childNotes && childNotes.length > 0) {
info += `**Child notes**: ${childNotes.length}\n`;
// List first few child notes
const childList = childNotes.slice(0, 5).map(child => child.title).join(', ');
if (childList) {
info += `**Examples**: ${childList}${childNotes.length > 5 ? '...' : ''}\n`;
}
}
// Add note type
if (note.type) {
info += `**Type**: ${note.type}\n`;
}
// Add creation/modification dates
if (note.utcDateCreated) {
info += `**Created**: ${new Date(note.utcDateCreated).toLocaleString()}\n`;
}
if (note.utcDateModified) {
info += `**Modified**: ${new Date(note.utcDateModified).toLocaleString()}\n`;
}
return info;
}
/**
* Get note hierarchy information - instance method
*/
async getNoteHierarchyInfo(noteId: string): Promise<string> {
return ContextExtractor.getNoteHierarchyInfo(noteId);
}
/**
* Get note summary - for backward compatibility
*/

56
src/services/llm/context/modules/context_service.ts
View File

@ -97,28 +97,60 @@ export class ContextService {
}
try {
// Step 1: Generate search queries
// Step 1: Generate search queries (skip if tool calling might be enabled)
let searchQueries: string[];
try {
searchQueries = await queryEnhancer.generateSearchQueries(userQuestion, llmService);
} catch (error) {
log.error(`Error generating search queries, using fallback: ${error}`);
searchQueries = [userQuestion]; // Fallback to using the original question
// Check if llmService has tool calling enabled
const isToolsEnabled = llmService &&
typeof llmService === 'object' &&
'constructor' in llmService &&
llmService.constructor.name === 'OllamaService';
if (isToolsEnabled) {
// Skip query generation if tools might be used to avoid race conditions
log.info(`Skipping query enhancement for potential tool-enabled service: ${llmService.constructor.name}`);
searchQueries = [userQuestion]; // Use simple fallback
} else {
try {
searchQueries = await queryEnhancer.generateSearchQueries(userQuestion, llmService);
} catch (error) {
log.error(`Error generating search queries, using fallback: ${error}`);
searchQueries = [userQuestion]; // Fallback to using the original question
}
}
log.info(`Generated search queries: ${JSON.stringify(searchQueries)}`);
// Step 2: Find relevant notes using multi-query approach
// Step 2: Find relevant notes using the pipeline's VectorSearchStage
let relevantNotes: NoteSearchResult[] = [];
try {
// Find notes for each query and combine results
log.info(`Using VectorSearchStage pipeline component to find relevant notes`);
// Create or import the vector search stage
const VectorSearchStage = (await import('../../pipeline/stages/vector_search_stage.js')).VectorSearchStage;
const vectorSearchStage = new VectorSearchStage();
// Use multi-query approach through the pipeline
const allResults: Map<string, NoteSearchResult> = new Map();
// Process searches using the pipeline stage
for (const query of searchQueries) {
const results = await semanticSearch.findRelevantNotes(
log.info(`Executing pipeline vector search for query: "${query.substring(0, 50)}..."`);
// Use the pipeline stage directly
const result = await vectorSearchStage.execute({
query,
contextNoteId,
5 // Limit per query
);
noteId: contextNoteId,
options: {
maxResults: 5, // Limit per query
useEnhancedQueries: false, // Don't enhance these - we already have enhanced queries
threshold: 0.6,
llmService // Pass the LLM service for potential use
}
});
const results = result.searchResults;
log.info(`Pipeline vector search found ${results.length} results for query "${query.substring(0, 50)}..."`);
// Combine results, avoiding duplicates
for (const result of results) {

67
src/services/llm/context_service.ts
View File

@ -107,6 +107,73 @@ class TriliumContextService {
contextNoteId: string | null = null,
limit = 10
): Promise<any[]> {
try {
// Use the VectorSearchStage for all searches to ensure consistency
const VectorSearchStage = (await import('./pipeline/stages/vector_search_stage.js')).VectorSearchStage;
const vectorSearchStage = new VectorSearchStage();
const allResults: Map<string, any> = new Map();
log.info(`Finding relevant notes for ${queries.length} queries in context ${contextNoteId || 'global'}`);
// Process each query in parallel using Promise.all for better performance
const searchPromises = queries.map(query =>
vectorSearchStage.execute({
query,
noteId: contextNoteId,
options: {
maxResults: Math.ceil(limit / queries.length), // Distribute limit among queries
useEnhancedQueries: false, // Don't enhance the queries here, as they're already enhanced
threshold: 0.5 // Lower threshold to get more diverse results
}
})
);
const searchResults = await Promise.all(searchPromises);
// Combine all results
for (let i = 0; i < searchResults.length; i++) {
const results = searchResults[i].searchResults;
log.info(`Query "${queries[i].substring(0, 30)}..." returned ${results.length} results`);
// Combine results, avoiding duplicates
for (const result of results) {
if (!allResults.has(result.noteId)) {
allResults.set(result.noteId, result);
} else {
// If note already exists, update similarity to max of both values
const existing = allResults.get(result.noteId);
if (result.similarity > existing.similarity) {
existing.similarity = result.similarity;
allResults.set(result.noteId, existing);
}
}
}
}
// Convert map to array and limit to top results
const finalResults = Array.from(allResults.values())
.sort((a, b) => b.similarity - a.similarity)
.slice(0, limit);
log.info(`Combined ${queries.length} queries into ${finalResults.length} final results`);
return finalResults;
} catch (error) {
log.error(`Error in findRelevantNotesMultiQuery: ${error}`);
// Fall back to legacy approach if the new approach fails
return this.findRelevantNotesMultiQueryLegacy(queries, contextNoteId, limit);
}
}
/**
* Legacy implementation of multi-query search (for fallback)
* @private
*/
private async findRelevantNotesMultiQueryLegacy(
queries: string[],
contextNoteId: string | null = null,
limit = 10
): Promise<any[]> {
log.info(`Using legacy findRelevantNotesMultiQuery implementation for ${queries.length} queries`);
const allResults: Map<string, any> = new Map();
for (const query of queries) {

87
src/services/llm/formatters/ollama_formatter.ts
View File

@ -9,6 +9,7 @@ import {
OLLAMA_CLEANING,
FORMATTER_LOGS
} from '../constants/formatter_constants.js';
import log from '../../log.js';
/**
* Ollama-specific message formatter
@ -31,14 +32,33 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
formatMessages(messages: Message[], systemPrompt?: string, context?: string, preserveSystemPrompt?: boolean): Message[] {
const formattedMessages: Message[] = [];
// First identify user and system messages
// Log the input messages with all their properties
log.info(`Ollama formatter received ${messages.length} messages`);
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`);
}
if (msg.tool_call_id) {
log.info(`Message ${index} has tool_call_id: ${msg.tool_call_id}`);
}
if (msg.name) {
log.info(`Message ${index} has name: ${msg.name}`);
}
});
// First identify user, system, and tool messages
const systemMessages = messages.filter(msg => msg.role === 'system');
const userMessages = messages.filter(msg => msg.role === 'user' || msg.role === 'assistant');
const nonSystemMessages = messages.filter(msg => msg.role !== 'system');
// Determine if we should preserve the existing system message
if (preserveSystemPrompt && systemMessages.length > 0) {
// Preserve the existing system message
formattedMessages.push(systemMessages[0]);
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;
@ -46,49 +66,78 @@ export class OllamaMessageFormatter extends BaseMessageFormatter {
role: 'system',
content: basePrompt
});
log.info(`Using new system message: ${basePrompt.substring(0, 50)}...`);
}
// If we have context, inject it into the first user message
if (context && userMessages.length > 0) {
if (context && nonSystemMessages.length > 0) {
let injectedContext = false;
for (let i = 0; i < userMessages.length; i++) {
const msg = userMessages[i];
for (let i = 0; i < nonSystemMessages.length; i++) {
const msg = nonSystemMessages[i];
if (msg.role === 'user' && !injectedContext) {
// Simple context injection directly in the user's message
const cleanedContext = this.cleanContextContent(context);
// DEBUG: Log the context before and after cleaning
console.log(`[OllamaFormatter] Context (first 500 chars): ${context.substring(0, 500).replace(/\n/g, '\\n')}...`);
console.log(`[OllamaFormatter] Cleaned context (first 500 chars): ${cleanedContext.substring(0, 500).replace(/\n/g, '\\n')}...`);
log.info(`Injecting context (${cleanedContext.length} chars) into user message`);
const formattedContext = PROVIDER_PROMPTS.OLLAMA.CONTEXT_INJECTION(
cleanedContext,
msg.content
);
// DEBUG: Log the final formatted context
console.log(`[OllamaFormatter] Formatted context (first 500 chars): ${formattedContext.substring(0, 500).replace(/\n/g, '\\n')}...`);
// Log what properties we're preserving
const msgKeys = Object.keys(msg);
const preservedKeys = msgKeys.filter(key => key !== 'role' && key !== 'content');
log.info(`Preserving additional properties in user message: ${preservedKeys.join(', ')}`);
formattedMessages.push({
role: 'user',
content: formattedContext
});
// Create a new message with all original properties, but updated content
const newMessage = {
...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(', ')}`);
injectedContext = true;
} else {
formattedMessages.push(msg);
// 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
});
}
}
} else {
// No context, just add all messages as-is
for (const msg of userMessages) {
formattedMessages.push(msg);
// Make sure to preserve all properties including tool_calls, tool_call_id, etc.
for (const msg of nonSystemMessages) {
log.info(`Adding message with role ${msg.role} without context injection, keys: ${Object.keys(msg).join(', ')}`);
formattedMessages.push({
...msg // Copy all properties
});
}
}
console.log(FORMATTER_LOGS.OLLAMA.PROCESSED(messages.length, formattedMessages.length));
// Log the final formatted messages
log.info(`Ollama formatter produced ${formattedMessages.length} formatted messages`);
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`);
}
if (msg.tool_call_id) {
log.info(`Formatted message ${index} has tool_call_id: ${msg.tool_call_id}`);
}
if (msg.name) {
log.info(`Formatted message ${index} has name: ${msg.name}`);
}
});
return formattedMessages;
}

191
src/services/llm/pipeline/chat_pipeline.ts
View File

@ -7,6 +7,10 @@ import { MessagePreparationStage } from './stages/message_preparation_stage.js';
import { ModelSelectionStage } from './stages/model_selection_stage.js';
import { LLMCompletionStage } from './stages/llm_completion_stage.js';
import { ResponseProcessingStage } from './stages/response_processing_stage.js';
import { ToolCallingStage } from './stages/tool_calling_stage.js';
import { VectorSearchStage } from './stages/vector_search_stage.js';
import toolRegistry from '../tools/tool_registry.js';
import toolInitializer from '../tools/tool_initializer.js';
import log from '../../log.js';
/**
@ -22,6 +26,8 @@ export class ChatPipeline {
modelSelection: ModelSelectionStage;
llmCompletion: LLMCompletionStage;
responseProcessing: ResponseProcessingStage;
toolCalling: ToolCallingStage;
vectorSearch: VectorSearchStage;
};
config: ChatPipelineConfig;
@ -40,7 +46,9 @@ export class ChatPipeline {
messagePreparation: new MessagePreparationStage(),
modelSelection: new ModelSelectionStage(),
llmCompletion: new LLMCompletionStage(),
responseProcessing: new ResponseProcessingStage()
responseProcessing: new ResponseProcessingStage(),
toolCalling: new ToolCallingStage(),
vectorSearch: new VectorSearchStage()
};
// Set default configuration values
@ -87,6 +95,34 @@ export class ChatPipeline {
contentLength += message.content.length;
}
// Initialize tools if needed
try {
const toolCount = toolRegistry.getAllTools().length;
// If there are no tools registered, initialize them
if (toolCount === 0) {
log.info('No tools found in registry, initializing tools...');
await toolInitializer.initializeTools();
log.info(`Tools initialized, now have ${toolRegistry.getAllTools().length} tools`);
} else {
log.info(`Found ${toolCount} tools already registered`);
}
} catch (error: any) {
log.error(`Error checking/initializing tools: ${error.message || String(error)}`);
}
// First, select the appropriate model based on query complexity and content length
const modelSelectionStartTime = Date.now();
const modelSelection = await this.stages.modelSelection.execute({
options: input.options,
query: input.query,
contentLength
});
this.updateStageMetrics('modelSelection', modelSelectionStartTime);
// Determine if we should use tools or semantic context
const useTools = modelSelection.options.enableTools === true;
// Determine which pipeline flow to use
let context: string | undefined;
@ -102,27 +138,63 @@ export class ChatPipeline {
});
context = agentContext.context;
this.updateStageMetrics('agentToolsContext', contextStartTime);
} else {
// Get semantic context for regular queries
} else if (!useTools) {
// Only get semantic context if tools are NOT enabled
// When tools are enabled, we'll let the LLM request context via tools instead
log.info('Getting semantic context for note using pipeline stages');
// First use the vector search stage to find relevant notes
const vectorSearchStartTime = Date.now();
log.info(`Executing vector search stage for query: "${input.query?.substring(0, 50)}..."`);
const vectorSearchResult = await this.stages.vectorSearch.execute({
query: input.query || '',
noteId: input.noteId,
options: {
maxResults: 10,
useEnhancedQueries: true,
threshold: 0.6
}
});
this.updateStageMetrics('vectorSearch', vectorSearchStartTime);
log.info(`Vector search found ${vectorSearchResult.searchResults.length} relevant notes`);
// Then pass to the semantic context stage to build the formatted context
const semanticContext = await this.stages.semanticContextExtraction.execute({
noteId: input.noteId,
query: input.query,
messages: input.messages
});
context = semanticContext.context;
this.updateStageMetrics('semanticContextExtraction', contextStartTime);
} else {
log.info('Tools are enabled - using minimal direct context to avoid race conditions');
// Get context from current note directly without semantic search
if (input.noteId) {
try {
const contextExtractor = new (await import('../../llm/context/index.js')).ContextExtractor();
// Just get the direct content of the current note
context = await contextExtractor.extractContext(input.noteId, {
includeContent: true,
includeParents: true,
includeChildren: true,
includeLinks: true,
includeSimilar: false // Skip semantic search to avoid race conditions
});
log.info(`Direct context extracted (${context.length} chars) without semantic search`);
} catch (error: any) {
log.error(`Error extracting direct context: ${error.message}`);
context = ""; // Fallback to empty context if extraction fails
}
} else {
context = ""; // No note ID, so no context
}
}
}
// Select the appropriate model based on query complexity and content length
const modelSelectionStartTime = Date.now();
const modelSelection = await this.stages.modelSelection.execute({
options: input.options,
query: input.query,
contentLength
});
this.updateStageMetrics('modelSelection', modelSelectionStartTime);
// Prepare messages with context and system prompt
const messagePreparationStartTime = Date.now();
const preparedMessages = await this.stages.messagePreparation.execute({
@ -167,17 +239,106 @@ export class ChatPipeline {
});
}
// For non-streaming responses, process the full response
// Process any tool calls in the response
let currentMessages = preparedMessages.messages;
let currentResponse = completion.response;
let needsFollowUp = false;
let toolCallIterations = 0;
const maxToolCallIterations = this.config.maxToolCallIterations;
// Check if tools were enabled in the options
const toolsEnabled = modelSelection.options.enableTools !== false;
log.info(`========== TOOL CALL PROCESSING ==========`);
log.info(`Tools enabled: ${toolsEnabled}`);
log.info(`Tool calls in response: ${currentResponse.tool_calls ? currentResponse.tool_calls.length : 0}`);
log.info(`Current response format: ${typeof currentResponse}`);
log.info(`Response keys: ${Object.keys(currentResponse).join(', ')}`);
// Detailed tool call inspection
if (currentResponse.tool_calls) {
currentResponse.tool_calls.forEach((tool, idx) => {
log.info(`Tool call ${idx+1}: ${JSON.stringify(tool)}`);
});
}
// Process tool calls if present and tools are enabled
if (toolsEnabled && currentResponse.tool_calls && currentResponse.tool_calls.length > 0) {
log.info(`Response contains ${currentResponse.tool_calls.length} tool calls, processing...`);
// Start tool calling loop
log.info(`Starting tool calling loop with max ${maxToolCallIterations} iterations`);
do {
log.info(`Tool calling iteration ${toolCallIterations + 1}`);
// Execute tool calling stage
const toolCallingStartTime = Date.now();
const toolCallingResult = await this.stages.toolCalling.execute({
response: currentResponse,
messages: currentMessages,
options: modelSelection.options
});
this.updateStageMetrics('toolCalling', toolCallingStartTime);
// Update state for next iteration
currentMessages = toolCallingResult.messages;
needsFollowUp = toolCallingResult.needsFollowUp;
// Make another call to the LLM if needed
if (needsFollowUp) {
log.info(`Tool execution completed, making follow-up LLM call (iteration ${toolCallIterations + 1})...`);
// Generate a new LLM response with the updated messages
const followUpStartTime = Date.now();
log.info(`Sending follow-up request to LLM with ${currentMessages.length} messages (including tool results)`);
const followUpCompletion = await this.stages.llmCompletion.execute({
messages: currentMessages,
options: modelSelection.options
});
this.updateStageMetrics('llmCompletion', followUpStartTime);
// Update current response for next iteration
currentResponse = followUpCompletion.response;
// Check for more tool calls
const hasMoreToolCalls = !!(currentResponse.tool_calls && currentResponse.tool_calls.length > 0);
if (hasMoreToolCalls) {
log.info(`Follow-up response contains ${currentResponse.tool_calls?.length || 0} more tool calls`);
} else {
log.info(`Follow-up response contains no more tool calls - completing tool loop`);
}
// Continue loop if there are more tool calls
needsFollowUp = hasMoreToolCalls;
}
// Increment iteration counter
toolCallIterations++;
} while (needsFollowUp && toolCallIterations < maxToolCallIterations);
// If we hit max iterations but still have tool calls, log a warning
if (toolCallIterations >= maxToolCallIterations && needsFollowUp) {
log.error(`Reached maximum tool call iterations (${maxToolCallIterations}), stopping`);
}
log.info(`Completed ${toolCallIterations} tool call iterations`);
}
// For non-streaming responses, process the final response
const processStartTime = Date.now();
const processed = await this.stages.responseProcessing.execute({
response: completion.response,
response: currentResponse,
options: input.options
});
this.updateStageMetrics('responseProcessing', processStartTime);
// Combine response with processed text, using accumulated text if streamed
const finalResponse: ChatResponse = {
...completion.response,
...currentResponse,
text: accumulatedText || processed.text
};

21
src/services/llm/pipeline/interfaces.ts
View File

@ -1,4 +1,5 @@
import type { Message, ChatCompletionOptions, ChatResponse, StreamChunk } from '../ai_interface.js';
import type { LLMServiceInterface } from '../interfaces/agent_tool_interfaces.js';
/**
* Base interface for pipeline input
@ -61,6 +62,25 @@ export interface ChatPipelineInput extends PipelineInput {
streamCallback?: StreamCallback;
}
/**
* Options for vector search operations
*/
export interface VectorSearchOptions {
maxResults?: number;
useEnhancedQueries?: boolean;
threshold?: number;
llmService?: LLMServiceInterface;
}
/**
* Input for vector search pipeline stage
*/
export interface VectorSearchInput extends PipelineInput {
query: string;
noteId?: string | null;
options?: VectorSearchOptions;
}
/**
* Base interface for pipeline stage output
*/
@ -130,6 +150,7 @@ export interface ToolExecutionInput extends PipelineInput {
response: ChatResponse;
messages: Message[];
options?: ChatCompletionOptions;
maxIterations?: number;
}
/**

23
src/services/llm/pipeline/stages/llm_completion_stage.ts
View File

@ -2,6 +2,7 @@ import { BasePipelineStage } from '../pipeline_stage.js';
import type { LLMCompletionInput } from '../interfaces.js';
import type { ChatResponse } from '../../ai_interface.js';
import aiServiceManager from '../../ai_service_manager.js';
import toolRegistry from '../../tools/tool_registry.js';
import log from '../../../log.js';
/**
@ -18,17 +19,33 @@ export class LLMCompletionStage extends BasePipelineStage<LLMCompletionInput, {
protected async process(input: LLMCompletionInput): Promise<{ response: ChatResponse }> {
const { messages, options, provider } = input;
log.info(`Generating LLM completion, provider: ${provider || 'auto'}, model: ${options?.model || 'default'}`);
// Create a copy of options to avoid modifying the original
const updatedOptions = { ...options };
// Check if tools should be enabled
if (updatedOptions.enableTools !== false) {
// Get all available tools from the registry
const toolDefinitions = toolRegistry.getAllToolDefinitions();
if (toolDefinitions.length > 0) {
// Enable tools and add them to the options
updatedOptions.enableTools = true;
updatedOptions.tools = toolDefinitions;
log.info(`Adding ${toolDefinitions.length} tools to LLM request`);
}
}
log.info(`Generating LLM completion, provider: ${provider || 'auto'}, model: ${updatedOptions?.model || 'default'}`);
// If provider is specified, use that specific provider
if (provider && aiServiceManager.isProviderAvailable(provider)) {
const service = aiServiceManager.getService(provider);
const response = await service.generateChatCompletion(messages, options);
const response = await service.generateChatCompletion(messages, updatedOptions);
return { response };
}
// Otherwise use the service manager to select an available provider
const response = await aiServiceManager.generateChatCompletion(messages, options);
const response = await aiServiceManager.generateChatCompletion(messages, updatedOptions);
return { response };
}
}

21
src/services/llm/pipeline/stages/message_preparation_stage.ts
View File

@ -3,6 +3,7 @@ import type { MessagePreparationInput } from '../interfaces.js';
import type { Message } from '../../ai_interface.js';
import { SYSTEM_PROMPTS } from '../../constants/llm_prompt_constants.js';
import { MessageFormatterFactory } from '../interfaces/message_formatter.js';
import toolRegistry from '../../tools/tool_registry.js';
import log from '../../../log.js';
/**
@ -27,15 +28,31 @@ export class MessagePreparationStage extends BasePipelineStage<MessagePreparatio
provider = providerName;
}
log.info(`Preparing messages for provider: ${provider}, context: ${!!context}, system prompt: ${!!systemPrompt}`);
// Check if tools are enabled
const toolsEnabled = options?.enableTools === true;
log.info(`Preparing messages for provider: ${provider}, context: ${!!context}, system prompt: ${!!systemPrompt}, tools: ${toolsEnabled}`);
// Get appropriate formatter for this provider
const formatter = MessageFormatterFactory.getFormatter(provider);
// Determine the system prompt to use
let finalSystemPrompt = systemPrompt || SYSTEM_PROMPTS.DEFAULT_SYSTEM_PROMPT;
// If tools are enabled, enhance system prompt with tools guidance
if (toolsEnabled) {
const toolCount = toolRegistry.getAllTools().length;
const toolsPrompt = `You have access to ${toolCount} tools to help you respond. When you need information that might be in the user's notes, use the search_notes tool to find relevant content or the read_note tool to read a specific note by ID. Use tools when specific information is required rather than making assumptions.`;
// Add tools guidance to system prompt
finalSystemPrompt = finalSystemPrompt + '\n\n' + toolsPrompt;
log.info(`Enhanced system prompt with tools guidance: ${toolCount} tools available`);
}
// Format messages using provider-specific approach
const formattedMessages = formatter.formatMessages(
messages,
systemPrompt || SYSTEM_PROMPTS.DEFAULT_SYSTEM_PROMPT,
finalSystemPrompt,
context
);

53
src/services/llm/pipeline/stages/model_selection_stage.ts
View File

@ -28,6 +28,39 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
// Get default model based on provider precedence
let defaultModel = 'openai:gpt-3.5-turbo'; // Fallback default
// Enable tools by default unless explicitly disabled
updatedOptions.enableTools = updatedOptions.enableTools !== false;
// Add tools if not already provided
if (updatedOptions.enableTools && (!updatedOptions.tools || updatedOptions.tools.length === 0)) {
try {
// Import tool registry and fetch tool definitions
const toolRegistry = (await import('../../tools/tool_registry.js')).default;
const toolDefinitions = toolRegistry.getAllToolDefinitions();
if (toolDefinitions.length > 0) {
updatedOptions.tools = toolDefinitions;
log.info(`Added ${toolDefinitions.length} tools to options`);
} else {
// Try to initialize tools
log.info('No tools found in registry, trying to initialize them');
try {
const toolInitializer = await import('../../tools/tool_initializer.js');
await toolInitializer.default.initializeTools();
// Try again after initialization
const reinitToolDefinitions = toolRegistry.getAllToolDefinitions();
updatedOptions.tools = reinitToolDefinitions;
log.info(`After initialization, added ${reinitToolDefinitions.length} tools to options`);
} catch (initError: any) {
log.error(`Failed to initialize tools: ${initError.message}`);
}
}
} catch (error: any) {
log.error(`Error loading tools: ${error.message}`);
}
}
try {
// Get provider precedence list
const providerPrecedence = await options.getOption('aiProviderPrecedence');
@ -55,7 +88,25 @@ export class ModelSelectionStage extends BasePipelineStage<ModelSelectionInput,
if (model) defaultModel = `anthropic:${model}`;
} else if (firstProvider === 'ollama') {
const model = await options.getOption('ollamaDefaultModel');
if (model) defaultModel = `ollama:${model}`;
if (model) {
defaultModel = `ollama:${model}`;
// Special configuration for Ollama
// Since Ollama models have different requirements for tool calling,
// configure based on the model being used
const modelLower = model.toLowerCase();
if (modelLower.includes('llama3') ||
modelLower.includes('mistral') ||
modelLower.includes('dolphin') ||
modelLower.includes('neural') ||
modelLower.includes('mist') ||
modelLower.includes('wizard')) {
// These models are known to support tool calling
log.info(`Using Ollama model ${model} with tool calling support`);
updatedOptions.enableTools = true;
}
}
}
}
}

47
src/services/llm/pipeline/stages/semantic_context_extraction_stage.ts
View File

@ -2,13 +2,20 @@ import { BasePipelineStage } from '../pipeline_stage.js';
import type { SemanticContextExtractionInput } from '../interfaces.js';
import aiServiceManager from '../../ai_service_manager.js';
import log from '../../../log.js';
import { VectorSearchStage } from './vector_search_stage.js';
import contextFormatter from '../../context/modules/context_formatter.js';
import providerManager from '../../context/modules/provider_manager.js';
/**
* Pipeline stage for extracting semantic context from notes
* This uses the new VectorSearchStage to find relevant content
*/
export class SemanticContextExtractionStage extends BasePipelineStage<SemanticContextExtractionInput, { context: string }> {
private vectorSearchStage: VectorSearchStage;
constructor() {
super('SemanticContextExtraction');
this.vectorSearchStage = new VectorSearchStage();
}
/**
@ -18,9 +25,43 @@ export class SemanticContextExtractionStage extends BasePipelineStage<SemanticCo
const { noteId, query, maxResults = 5, messages = [] } = input;
log.info(`Extracting semantic context from note ${noteId}, query: ${query?.substring(0, 50)}...`);
const contextService = aiServiceManager.getContextService();
const context = await contextService.getSemanticContext(noteId, query, maxResults, messages);
try {
// Step 1: Use vector search stage to find relevant notes
const vectorSearchResult = await this.vectorSearchStage.execute({
query,
noteId,
options: {
maxResults,
useEnhancedQueries: true,
threshold: 0.6,
llmService: null // Will use default service
}
});
return { context };
log.info(`Vector search found ${vectorSearchResult.searchResults.length} relevant notes`);
// If no results, return empty context
if (vectorSearchResult.searchResults.length === 0) {
log.info(`No relevant notes found for context extraction`);
return { context: "" };
}
// Step 2: Format search results into a context string
const provider = await providerManager.getPreferredEmbeddingProvider();
const providerId = provider?.name || 'default';
const context = await contextFormatter.buildContextFromNotes(
vectorSearchResult.searchResults,
query,
providerId,
messages
);
log.info(`Built context of ${context.length} chars from ${vectorSearchResult.searchResults.length} notes`);
return { context };
} catch (error) {
log.error(`Error extracting semantic context: ${error}`);
return { context: "" };
}
}
}

216
src/services/llm/pipeline/stages/tool_calling_stage.ts Normal file
View File

@ -0,0 +1,216 @@
import { BasePipelineStage } from '../pipeline_stage.js';
import type { ToolExecutionInput } from '../interfaces.js';
import log from '../../../log.js';
import type { ChatResponse, Message } from '../../ai_interface.js';
import toolRegistry from '../../tools/tool_registry.js';
/**
* Pipeline stage for handling LLM tool calling
* This stage is responsible for:
* 1. Detecting tool calls in LLM responses
* 2. Executing the appropriate tools
* 3. Adding tool results back to the conversation
* 4. Determining if we need to make another call to the LLM
*/
export class ToolCallingStage extends BasePipelineStage<ToolExecutionInput, { response: ChatResponse, needsFollowUp: boolean, messages: Message[] }> {
constructor() {
super('ToolCalling');
}
/**
* Process the LLM response and execute any tool calls
*/
protected async process(input: ToolExecutionInput): Promise<{ response: ChatResponse, needsFollowUp: boolean, messages: Message[] }> {
const { response, messages, options } = input;
// Check if the response has tool calls
if (!response.tool_calls || response.tool_calls.length === 0) {
// No tool calls, return original response and messages
log.info(`No tool calls detected in response from provider: ${response.provider}`);
return { response, needsFollowUp: false, messages };
}
log.info(`LLM requested ${response.tool_calls.length} tool calls from provider: ${response.provider}`);
// Log response details for debugging
if (response.text) {
log.info(`Response text: "${response.text.substring(0, 200)}${response.text.length > 200 ? '...' : ''}"`);
}
// Check if the registry has any tools
const availableTools = toolRegistry.getAllTools();
log.info(`Available tools in registry: ${availableTools.length}`);
if (availableTools.length === 0) {
log.error(`No tools available in registry, cannot execute tool calls`);
// Try to initialize tools as a recovery step
try {
log.info('Attempting to initialize tools as recovery step');
const toolInitializer = await import('../../tools/tool_initializer.js');
await toolInitializer.default.initializeTools();
log.info(`After recovery initialization: ${toolRegistry.getAllTools().length} tools available`);
} catch (error: any) {
log.error(`Failed to initialize tools in recovery step: ${error.message}`);
}
}
// Create a copy of messages to add the assistant message with tool calls
const updatedMessages = [...messages];
// Add the assistant message with the tool calls
updatedMessages.push({
role: 'assistant',
content: response.text || "",
tool_calls: response.tool_calls
});
// Execute each tool call and add results to messages
const toolResults = await Promise.all(response.tool_calls.map(async (toolCall) => {
try {
log.info(`Tool call received - Name: ${toolCall.function.name}, ID: ${toolCall.id || 'unknown'}`);
// Log parameters
const argsStr = typeof toolCall.function.arguments === 'string'
? toolCall.function.arguments
: JSON.stringify(toolCall.function.arguments);
log.info(`Tool parameters: ${argsStr}`);
// Get the tool from registry
const tool = toolRegistry.getTool(toolCall.function.name);
if (!tool) {
throw new Error(`Tool not found: ${toolCall.function.name}`);
}
// Parse arguments (handle both string and object formats)
let args;
// At this stage, arguments should already be processed by the provider-specific service
// But we still need to handle different formats just in case
if (typeof toolCall.function.arguments === 'string') {
log.info(`Received string arguments in tool calling stage: ${toolCall.function.arguments.substring(0, 50)}...`);
try {
// Try to parse as JSON first
args = JSON.parse(toolCall.function.arguments);
log.info(`Parsed JSON arguments: ${Object.keys(args).join(', ')}`);
} catch (e) {
// If it's not valid JSON, try to check if it's a stringified object with quotes
log.info(`Failed to parse arguments as JSON, trying alternative parsing: ${e.message}`);
// Sometimes LLMs return stringified JSON with escaped quotes or incorrect quotes
// Try to clean it up
try {
const cleaned = toolCall.function.arguments
.replace(/^['"]|['"]$/g, '') // Remove surrounding quotes
.replace(/\\"/g, '"') // Replace escaped quotes
.replace(/([{,])\s*'([^']+)'\s*:/g, '$1"$2":') // Replace single quotes around property names
.replace(/([{,])\s*(\w+)\s*:/g, '$1"$2":'); // Add quotes around unquoted property names
log.info(`Cleaned argument string: ${cleaned}`);
args = JSON.parse(cleaned);
log.info(`Successfully parsed cleaned arguments: ${Object.keys(args).join(', ')}`);
} catch (cleanError) {
// If all parsing fails, treat it as a text argument
log.info(`Failed to parse cleaned arguments: ${cleanError.message}`);
args = { text: toolCall.function.arguments };
log.info(`Using text argument: ${args.text.substring(0, 50)}...`);
}
}
} else {
// Arguments are already an object
args = toolCall.function.arguments;
log.info(`Using object arguments with keys: ${Object.keys(args).join(', ')}`);
}
// Execute the tool
log.info(`================ EXECUTING TOOL: ${toolCall.function.name} ================`);
log.info(`Tool parameters: ${Object.keys(args).join(', ')}`);
log.info(`Parameters values: ${Object.entries(args).map(([k, v]) => `${k}=${typeof v === 'string' ? v : JSON.stringify(v)}`).join(', ')}`);
const executionStart = Date.now();
let result;
try {
log.info(`Starting tool execution for ${toolCall.function.name}...`);
result = await tool.execute(args);
const executionTime = Date.now() - executionStart;
log.info(`================ TOOL EXECUTION COMPLETED in ${executionTime}ms ================`);
} catch (execError: any) {
const executionTime = Date.now() - executionStart;
log.error(`================ TOOL EXECUTION FAILED in ${executionTime}ms: ${execError.message} ================`);
throw execError;
}
// Log execution result
const resultSummary = typeof result === 'string'
? `${result.substring(0, 100)}...`
: `Object with keys: ${Object.keys(result).join(', ')}`;
log.info(`Tool execution completed in ${executionTime}ms - Result: ${resultSummary}`);
// Return result with tool call ID
return {
toolCallId: toolCall.id,
name: toolCall.function.name,
result
};
} catch (error: any) {
log.error(`Error executing tool ${toolCall.function.name}: ${error.message || String(error)}`);
// Return error message as result
return {
toolCallId: toolCall.id,
name: toolCall.function.name,
result: `Error: ${error.message || String(error)}`
};
}
}));
// Add tool results as messages
toolResults.forEach(result => {
// Format the result content based on type
let content: string;
if (typeof result.result === 'string') {
content = result.result;
log.info(`Tool returned string result (${content.length} chars)`);
} else {
// For object results, format as JSON
try {
content = JSON.stringify(result.result, null, 2);
log.info(`Tool returned object result with keys: ${Object.keys(result.result).join(', ')}`);
} catch (error) {
content = String(result.result);
log.info(`Failed to stringify object result: ${error}`);
}
}
log.info(`Adding tool result message - Tool: ${result.name}, ID: ${result.toolCallId || 'unknown'}, Length: ${content.length}`);
// Create a properly formatted tool response message
updatedMessages.push({
role: 'tool',
content: content,
name: result.name,
tool_call_id: result.toolCallId
});
// Log a sample of the content for debugging
const contentPreview = content.substring(0, 100) + (content.length > 100 ? '...' : '');
log.info(`Tool result preview: ${contentPreview}`);
});
log.info(`Added ${toolResults.length} tool results to conversation`);
// If we have tool results, we need a follow-up call to the LLM
const needsFollowUp = toolResults.length > 0;
if (needsFollowUp) {
log.info(`Tool execution complete, LLM follow-up required with ${updatedMessages.length} messages`);
}
return {
response,
needsFollowUp,
messages: updatedMessages
};
}
}

206
src/services/llm/pipeline/stages/vector_search_stage.ts Normal file
View File

@ -0,0 +1,206 @@
import { BasePipelineStage } from '../pipeline_stage.js';
import type { VectorSearchInput } from '../interfaces.js';
import type { NoteSearchResult } from '../../interfaces/context_interfaces.js';
import log from '../../../log.js';
import queryEnhancer from '../../context/modules/query_enhancer.js';
import semanticSearch from '../../context/modules/semantic_search.js';
import aiServiceManager from '../../ai_service_manager.js';
/**
* Pipeline stage for handling semantic vector search with query enhancement
* This centralizes all semantic search operations into the pipeline
*/
export class VectorSearchStage extends BasePipelineStage<VectorSearchInput, {
searchResults: NoteSearchResult[],
enhancedQueries?: string[]
}> {
constructor() {
super('VectorSearch');
}
/**
* Execute semantic search with optional query enhancement
*/
protected async process(input: VectorSearchInput): Promise<{
searchResults: NoteSearchResult[],
enhancedQueries?: string[]
}> {
const { query, noteId, options = {} } = input;
const {
maxResults = 10,
useEnhancedQueries = true,
threshold = 0.6,
llmService = null
} = options;
log.info(`========== PIPELINE VECTOR SEARCH ==========`);
log.info(`Query: "${query.substring(0, 100)}${query.length > 100 ? '...' : ''}"`);
log.info(`Parameters: noteId=${noteId || 'global'}, maxResults=${maxResults}, useEnhancedQueries=${useEnhancedQueries}, threshold=${threshold}`);
log.info(`LLM Service provided: ${llmService ? 'yes' : 'no'}`);
log.info(`Start timestamp: ${new Date().toISOString()}`);
try {
// STEP 1: Generate enhanced search queries if requested
let searchQueries: string[] = [query];
if (useEnhancedQueries) {
log.info(`PIPELINE VECTOR SEARCH: Generating enhanced queries for: "${query.substring(0, 50)}..."`);
try {
// Get the LLM service to use for query enhancement
let enhancementService = llmService;
// If no service provided, use AI service manager to get the default service
if (!enhancementService) {
log.info(`No LLM service provided, using default from AI service manager`);
const manager = aiServiceManager.getInstance();
const provider = manager.getPreferredProvider();
enhancementService = manager.getService(provider);
log.info(`Using preferred provider "${provider}" with service type ${enhancementService.constructor.name}`);
}
// Create a special service wrapper that prevents recursion
const recursionPreventionService = {
generateChatCompletion: async (messages: any, options: any) => {
// Add flags to prevent recursive calls
const safeOptions = {
...options,
bypassFormatter: true,
_bypassContextProcessing: true,
bypassQueryEnhancement: true, // Critical flag
directToolExecution: true,
enableTools: false // Disable tools for query enhancement
};
// Use the actual service implementation but with safe options
return enhancementService.generateChatCompletion(messages, safeOptions);
}
};
// Call the query enhancer with the safe service
searchQueries = await queryEnhancer.generateSearchQueries(query, recursionPreventionService);
log.info(`PIPELINE VECTOR SEARCH: Generated ${searchQueries.length} enhanced queries`);
} catch (error) {
log.error(`PIPELINE VECTOR SEARCH: Error generating search queries, using original: ${error}`);
searchQueries = [query]; // Fall back to original query
}
} else {
log.info(`PIPELINE VECTOR SEARCH: Using direct query without enhancement: "${query}"`);
}
// STEP 2: Find relevant notes for each query
const allResults = new Map<string, NoteSearchResult>();
log.info(`PIPELINE VECTOR SEARCH: Searching for ${searchQueries.length} queries`);
for (const searchQuery of searchQueries) {
try {
log.info(`PIPELINE VECTOR SEARCH: Processing query: "${searchQuery.substring(0, 50)}..."`);
const results = await semanticSearch.findRelevantNotes(
searchQuery,
noteId || null,
maxResults
);
log.info(`PIPELINE VECTOR SEARCH: Found ${results.length} results for query "${searchQuery.substring(0, 50)}..."`);
// Combine results, avoiding duplicates and keeping the highest similarity score
for (const result of results) {
if (!allResults.has(result.noteId)) {
allResults.set(result.noteId, result);
} else {
// If note already exists, update similarity to max of both values
const existing = allResults.get(result.noteId);
if (existing && result.similarity > existing.similarity) {
existing.similarity = result.similarity;
allResults.set(result.noteId, existing);
}
}
}
} catch (error) {
log.error(`PIPELINE VECTOR SEARCH: Error searching for query "${searchQuery}": ${error}`);
}
}
// STEP 3: Convert to array, filter and sort
const filteredResults = Array.from(allResults.values())
.filter(note => {
// Filter out notes with no content or very minimal content
const hasContent = note.content && note.content.trim().length > 10;
// Apply similarity threshold
const meetsThreshold = note.similarity >= threshold;
if (!hasContent) {
log.info(`PIPELINE VECTOR SEARCH: Filtering out empty/minimal note: "${note.title}" (${note.noteId})`);
}
if (!meetsThreshold) {
log.info(`PIPELINE VECTOR SEARCH: Filtering out low similarity note: "${note.title}" - ${Math.round(note.similarity * 100)}% < ${Math.round(threshold * 100)}%`);
}
return hasContent && meetsThreshold;
})
.sort((a, b) => b.similarity - a.similarity)
.slice(0, maxResults);
log.info(`PIPELINE VECTOR SEARCH: Search complete, returning ${filteredResults.length} results after filtering`);
// Log top results in detail
if (filteredResults.length > 0) {
log.info(`========== VECTOR SEARCH RESULTS ==========`);
log.info(`Found ${filteredResults.length} relevant notes after filtering`);
const topResults = filteredResults.slice(0, 5); // Show top 5 for better diagnostics
topResults.forEach((result, idx) => {
log.info(`Result ${idx+1}:`);
log.info(` Title: "${result.title}"`);
log.info(` NoteID: ${result.noteId}`);
log.info(` Similarity: ${Math.round(result.similarity * 100)}%`);
if (result.content) {
const contentPreview = result.content.length > 150
? `${result.content.substring(0, 150)}...`
: result.content;
log.info(` Content preview: ${contentPreview}`);
log.info(` Content length: ${result.content.length} chars`);
} else {
log.info(` Content: None or not loaded`);
}
});
if (filteredResults.length > 5) {
log.info(`... and ${filteredResults.length - 5} more results not shown`);
}
log.info(`========== END VECTOR SEARCH RESULTS ==========`);
} else {
log.info(`No results found that meet the similarity threshold of ${threshold}`);
}
// Log final statistics
log.info(`Vector search statistics:`);
log.info(` Original query: "${query.substring(0, 50)}${query.length > 50 ? '...' : ''}"`);
if (searchQueries.length > 1) {
log.info(` Enhanced with ${searchQueries.length} search queries`);
searchQueries.forEach((q, i) => {
if (i > 0) { // Skip the original query
log.info(` Query ${i}: "${q.substring(0, 50)}${q.length > 50 ? '...' : ''}"`);
}
});
}
log.info(` Final results: ${filteredResults.length} notes`);
log.info(` End timestamp: ${new Date().toISOString()}`);
log.info(`========== END PIPELINE VECTOR SEARCH ==========`);
return {
searchResults: filteredResults,
enhancedQueries: useEnhancedQueries ? searchQueries : undefined
};
} catch (error: any) {
log.error(`PIPELINE VECTOR SEARCH: Error in vector search stage: ${error.message || String(error)}`);
return {
searchResults: [],
enhancedQueries: undefined
};
}
}
}

258
src/services/llm/providers/ollama_service.ts
View File

@ -3,10 +3,26 @@ import { BaseAIService } from '../base_ai_service.js';
import type { Message, ChatCompletionOptions, ChatResponse } from '../ai_interface.js';
import sanitizeHtml from 'sanitize-html';
import { OllamaMessageFormatter } from '../formatters/ollama_formatter.js';
import log from '../../log.js';
import type { ToolCall } from '../tools/tool_interfaces.js';
import toolRegistry from '../tools/tool_registry.js';
interface OllamaFunctionArguments {
[key: string]: any;
}
interface OllamaFunctionCall {
function: {
name: string;
arguments: OllamaFunctionArguments | string;
};
id?: string;
}
interface OllamaMessage {
role: string;
content: string;
tool_calls?: OllamaFunctionCall[];
}
interface OllamaResponse {
@ -14,6 +30,7 @@ interface OllamaResponse {
created_at: string;
message: OllamaMessage;
done: boolean;
done_reason?: string;
total_duration: number;
load_duration: number;
prompt_eval_count: number;
@ -54,7 +71,7 @@ export class OllamaService extends BaseAIService {
if (opts.bypassFormatter) {
// Bypass the formatter entirely - use messages as is
messagesToSend = [...messages];
console.log(`Bypassing formatter for Ollama request with ${messages.length} messages`);
log.info(`Bypassing formatter for Ollama request with ${messages.length} messages`);
} else {
// Use the formatter to prepare messages
messagesToSend = this.formatter.formatMessages(
@ -63,44 +80,156 @@ export class OllamaService extends BaseAIService {
undefined, // context
opts.preserveSystemPrompt
);
console.log(`Sending to Ollama with formatted messages:`, JSON.stringify(messagesToSend, null, 2));
log.info(`Sending to Ollama with formatted messages: ${messagesToSend.length}`);
}
// Check if this is a request that expects JSON response
const expectsJsonResponse = opts.expectsJsonResponse || false;
if (expectsJsonResponse) {
console.log(`Request expects JSON response, adding response_format parameter`);
// Build request body
const requestBody: any = {
model,
messages: messagesToSend,
options: {
temperature,
// Add response_format for requests that expect JSON
...(expectsJsonResponse ? { response_format: { type: "json_object" } } : {})
},
stream: false
};
// Add tools if enabled - put them at the top level for Ollama
if (opts.enableTools !== false) {
// Get tools from registry if not provided in options
if (!opts.tools || opts.tools.length === 0) {
try {
// Get tool definitions from registry
const tools = toolRegistry.getAllToolDefinitions();
requestBody.tools = tools;
log.info(`Adding ${tools.length} tools to request`);
// If no tools found, reinitialize
if (tools.length === 0) {
log.info('No tools found in registry, re-initializing...');
try {
const toolInitializer = await import('../tools/tool_initializer.js');
await toolInitializer.default.initializeTools();
// Try again
requestBody.tools = toolRegistry.getAllToolDefinitions();
log.info(`After re-initialization: ${requestBody.tools.length} tools available`);
} catch (err: any) {
log.error(`Failed to re-initialize tools: ${err.message}`);
}
}
} catch (error: any) {
log.error(`Error getting tools: ${error.message || String(error)}`);
// Create default empty tools array if we couldn't load the tools
requestBody.tools = [];
}
} else {
requestBody.tools = opts.tools;
}
log.info(`Adding ${requestBody.tools.length} tools to Ollama request`);
} else {
log.info('Tools are explicitly disabled for this request');
}
// Log key request details
log.info(`========== OLLAMA API REQUEST ==========`);
log.info(`Model: ${requestBody.model}, Messages: ${requestBody.messages.length}, Tools: ${requestBody.tools ? requestBody.tools.length : 0}`);
log.info(`Temperature: ${temperature}, Stream: ${requestBody.stream}, JSON response expected: ${expectsJsonResponse}`);
// Check message structure and log detailed information about each message
requestBody.messages.forEach((msg: any, index: number) => {
const keys = Object.keys(msg);
log.info(`Message ${index}, Role: ${msg.role}, Keys: ${keys.join(', ')}`);
// Log message content preview
if (msg.content && typeof msg.content === 'string') {
const contentPreview = msg.content.length > 200
? `${msg.content.substring(0, 200)}...`
: msg.content;
log.info(`Message ${index} content: ${contentPreview}`);
}
// Log tool-related details
if (keys.includes('tool_calls')) {
log.info(`Message ${index} has ${msg.tool_calls.length} tool calls:`);
msg.tool_calls.forEach((call: any, callIdx: number) => {
log.info(` Tool call ${callIdx}: ${call.function?.name || 'unknown'}, ID: ${call.id || 'unspecified'}`);
if (call.function?.arguments) {
const argsPreview = typeof call.function.arguments === 'string'
? call.function.arguments.substring(0, 100)
: JSON.stringify(call.function.arguments).substring(0, 100);
log.info(` Arguments: ${argsPreview}...`);
}
});
}
if (keys.includes('tool_call_id')) {
log.info(`Message ${index} is a tool response for tool call ID: ${msg.tool_call_id}`);
}
if (keys.includes('name') && msg.role === 'tool') {
log.info(`Message ${index} is from tool: ${msg.name}`);
}
});
// Log tool definitions
if (requestBody.tools && requestBody.tools.length > 0) {
log.info(`Sending ${requestBody.tools.length} tool definitions:`);
requestBody.tools.forEach((tool: any, toolIdx: number) => {
log.info(` Tool ${toolIdx}: ${tool.function?.name || 'unnamed'}`);
if (tool.function?.description) {
log.info(` Description: ${tool.function.description.substring(0, 100)}...`);
}
if (tool.function?.parameters) {
const paramNames = tool.function.parameters.properties
? Object.keys(tool.function.parameters.properties)
: [];
log.info(` Parameters: ${paramNames.join(', ')}`);
}
});
}
// Log full request body (this will create large logs but is helpful for debugging)
const requestStr = JSON.stringify(requestBody);
log.info(`Full Ollama request (truncated): ${requestStr.substring(0, 1000)}...`);
log.info(`========== END OLLAMA REQUEST ==========`);
// Make API request
const response = await fetch(`${apiBase}/api/chat`, {
method: 'POST',
headers: {
'Content-Type': 'application/json'
},
body: JSON.stringify({
model,
messages: messagesToSend,
options: {
temperature,
// Add response_format for requests that expect JSON
...(expectsJsonResponse ? { response_format: { type: "json_object" } } : {})
},
stream: false
})
body: JSON.stringify(requestBody)
});
if (!response.ok) {
const errorBody = await response.text();
console.error(`Ollama API error: ${response.status} ${response.statusText}`, errorBody);
log.error(`Ollama API error: ${response.status} ${response.statusText} - ${errorBody}`);
throw new Error(`Ollama API error: ${response.status} ${response.statusText}`);
}
const data: OllamaResponse = await response.json();
console.log('Raw response from Ollama:', JSON.stringify(data, null, 2));
console.log('Parsed Ollama response:', JSON.stringify(data, null, 2));
return {
// Log response details
log.info(`========== OLLAMA API RESPONSE ==========`);
log.info(`Model: ${data.model}, Content length: ${data.message.content.length} chars`);
log.info(`Tokens: ${data.prompt_eval_count} prompt, ${data.eval_count} completion, ${data.prompt_eval_count + data.eval_count} total`);
log.info(`Duration: ${data.total_duration}ns total, ${data.prompt_eval_duration}ns prompt, ${data.eval_duration}ns completion`);
log.info(`Done: ${data.done}, Reason: ${data.done_reason || 'not specified'}`);
// Log content preview
const contentPreview = data.message.content.length > 300
? `${data.message.content.substring(0, 300)}...`
: data.message.content;
log.info(`Response content: ${contentPreview}`);
// Handle the response and extract tool calls if present
const chatResponse: ChatResponse = {
text: data.message.content,
model: data.model,
provider: this.getName(),
@ -110,8 +239,97 @@ export class OllamaService extends BaseAIService {
totalTokens: data.prompt_eval_count + data.eval_count
}
};
} catch (error) {
console.error('Ollama service error:', error);
// Add tool calls if present
if (data.message.tool_calls && data.message.tool_calls.length > 0) {
log.info(`Ollama response includes ${data.message.tool_calls.length} tool calls`);
// Log detailed information about each tool call
const transformedToolCalls: ToolCall[] = [];
// Log detailed information about the tool calls in the response
log.info(`========== OLLAMA TOOL CALLS IN RESPONSE ==========`);
data.message.tool_calls.forEach((toolCall, index) => {
log.info(`Tool call ${index + 1}:`);
log.info(` Name: ${toolCall.function?.name || 'unknown'}`);
log.info(` ID: ${toolCall.id || `auto-${index + 1}`}`);
// Generate a unique ID if none is provided
const id = toolCall.id || `tool-call-${Date.now()}-${index}`;
// Handle arguments based on their type
let processedArguments: Record<string, any> | string;
if (typeof toolCall.function.arguments === 'string') {
// Log raw string arguments in full for debugging
log.info(` Raw string arguments: ${toolCall.function.arguments}`);
// Try to parse JSON string arguments
try {
processedArguments = JSON.parse(toolCall.function.arguments);
log.info(` Successfully parsed arguments to object with keys: ${Object.keys(processedArguments).join(', ')}`);
log.info(` Parsed argument values:`);
Object.entries(processedArguments).forEach(([key, value]) => {
const valuePreview = typeof value === 'string'
? (value.length > 100 ? `${value.substring(0, 100)}...` : value)
: JSON.stringify(value);
log.info(` ${key}: ${valuePreview}`);
});
} catch (e) {
// If parsing fails, keep as string and log the error
processedArguments = toolCall.function.arguments;
log.info(` Could not parse arguments as JSON: ${e.message}`);
log.info(` Keeping as string: ${processedArguments.substring(0, 200)}${processedArguments.length > 200 ? '...' : ''}`);
// Try to clean and parse again with more aggressive methods
try {
const cleaned = toolCall.function.arguments
.replace(/^['"]|['"]$/g, '') // Remove surrounding quotes
.replace(/\\"/g, '"') // Replace escaped quotes
.replace(/([{,])\s*'([^']+)'\s*:/g, '$1"$2":') // Replace single quotes around property names
.replace(/([{,])\s*(\w+)\s*:/g, '$1"$2":'); // Add quotes around unquoted property names
log.info(` Attempting to parse cleaned argument: ${cleaned}`);
const reparseArg = JSON.parse(cleaned);
log.info(` Successfully parsed cleaned argument with keys: ${Object.keys(reparseArg).join(', ')}`);
} catch (cleanErr) {
log.info(` Failed to parse cleaned arguments: ${cleanErr.message}`);
}
}
} else {
// If it's already an object, use it directly and log details
processedArguments = toolCall.function.arguments;
log.info(` Object arguments with keys: ${Object.keys(processedArguments).join(', ')}`);
log.info(` Argument values:`);
Object.entries(processedArguments).forEach(([key, value]) => {
const valuePreview = typeof value === 'string'
? (value.length > 100 ? `${value.substring(0, 100)}...` : value)
: JSON.stringify(value);
log.info(` ${key}: ${valuePreview}`);
});
}
// Convert to our standard ToolCall format
transformedToolCalls.push({
id,
type: 'function',
function: {
name: toolCall.function.name,
arguments: processedArguments
}
});
});
// Add transformed tool calls to response
chatResponse.tool_calls = transformedToolCalls;
log.info(`Transformed ${transformedToolCalls.length} tool calls for execution`);
log.info(`========== END OLLAMA TOOL CALLS ==========`);
}
log.info(`========== END OLLAMA RESPONSE ==========`);
return chatResponse;
} catch (error: any) {
log.error(`Ollama service error: ${error.message || String(error)}`);
throw error;
}
}

157
src/services/llm/rest_chat_service.ts
View File

@ -609,6 +609,56 @@ class RestChatService {
// Use the correct method name: generateChatCompletion
const response = await service.generateChatCompletion(aiMessages, chatOptions);
// Check for tool calls in the response
if (response.tool_calls && response.tool_calls.length > 0) {
log.info(`========== STREAMING TOOL CALLS DETECTED ==========`);
log.info(`Response contains ${response.tool_calls.length} tool calls, executing them...`);
try {
// Execute the tools
const toolResults = await this.executeToolCalls(response);
// Make a follow-up request with the tool results
const toolMessages = [...aiMessages, {
role: 'assistant',
content: response.text || '',
tool_calls: response.tool_calls
}, ...toolResults];
log.info(`Making follow-up request with ${toolResults.length} tool results`);
// Send partial response to let the client know tools are being processed
if (!res.writableEnded) {
res.write(`data: ${JSON.stringify({ content: "Processing tools... " })}\n\n`);
}
// Use non-streaming for the follow-up to get a complete response
const followUpOptions = {...chatOptions, stream: false, enableTools: false}; // Prevent infinite loops
const followUpResponse = await service.generateChatCompletion(toolMessages, followUpOptions);
messageContent = followUpResponse.text || "";
// Send the complete response as a single chunk
if (!res.writableEnded) {
res.write(`data: ${JSON.stringify({ content: messageContent })}\n\n`);
res.write('data: [DONE]\n\n');
res.end();
}
// Store the full response for the session
session.messages.push({
role: 'assistant',
content: messageContent,
timestamp: new Date()
});
return; // Skip the rest of the processing
} catch (toolError) {
log.error(`Error executing tools: ${toolError}`);
// Continue with normal streaming response as fallback
}
}
// Handle streaming if the response includes a stream method
if (response.stream) {
await response.stream((chunk: { text: string; done: boolean }) => {
@ -667,6 +717,113 @@ class RestChatService {
}
}
/**
* Execute tool calls from the LLM response
* @param response The LLM response containing tool calls
*/
private async executeToolCalls(response: any): Promise<Message[]> {
if (!response.tool_calls || response.tool_calls.length === 0) {
return [];
}
log.info(`Executing ${response.tool_calls.length} tool calls from REST chat service`);
// Import tool registry directly to avoid circular dependencies
const toolRegistry = (await import('./tools/tool_registry.js')).default;
// Check if tools are available
const availableTools = toolRegistry.getAllTools();
if (availableTools.length === 0) {
log.error('No tools available in registry for execution');
// Try to initialize tools
try {
const toolInitializer = await import('./tools/tool_initializer.js');
await toolInitializer.default.initializeTools();
log.info(`Initialized ${toolRegistry.getAllTools().length} tools`);
} catch (error) {
log.error(`Failed to initialize tools: ${error}`);
throw new Error('Tool execution failed: No tools available');
}
}
// Execute each tool call and collect results
const toolResults = await Promise.all(response.tool_calls.map(async (toolCall: any) => {
try {
log.info(`Executing tool: ${toolCall.function.name}, ID: ${toolCall.id || 'unknown'}`);
// Get the tool from registry
const tool = toolRegistry.getTool(toolCall.function.name);
if (!tool) {
throw new Error(`Tool not found: ${toolCall.function.name}`);
}
// Parse arguments
let args;
if (typeof toolCall.function.arguments === 'string') {
try {
args = JSON.parse(toolCall.function.arguments);
} catch (e) {
log.error(`Failed to parse tool arguments: ${e.message}`);
// Try cleanup and retry
try {
const cleaned = toolCall.function.arguments
.replace(/^['"]|['"]$/g, '') // Remove surrounding quotes
.replace(/\\"/g, '"') // Replace escaped quotes
.replace(/([{,])\s*'([^']+)'\s*:/g, '$1"$2":') // Replace single quotes around property names
.replace(/([{,])\s*(\w+)\s*:/g, '$1"$2":'); // Add quotes around unquoted property names
args = JSON.parse(cleaned);
} catch (cleanErr) {
// If all parsing fails, use as-is
args = { text: toolCall.function.arguments };
}
}
} else {
args = toolCall.function.arguments;
}
// Log what we're about to execute
log.info(`Executing tool with arguments: ${JSON.stringify(args)}`);
// Execute the tool and get result
const startTime = Date.now();
const result = await tool.execute(args);
const executionTime = Date.now() - startTime;
log.info(`Tool execution completed in ${executionTime}ms`);
// Log the result
const resultPreview = typeof result === 'string'
? result.substring(0, 100) + (result.length > 100 ? '...' : '')
: JSON.stringify(result).substring(0, 100) + '...';
log.info(`Tool result: ${resultPreview}`);
// Format result as a proper message
return {
role: 'tool',
content: typeof result === 'string' ? result : JSON.stringify(result),
name: toolCall.function.name,
tool_call_id: toolCall.id || `tool-${Date.now()}-${Math.random().toString(36).substring(2, 9)}`
};
} catch (error: any) {
log.error(`Error executing tool ${toolCall.function.name}: ${error.message}`);
// Return error as tool result
return {
role: 'tool',
content: `Error: ${error.message}`,
name: toolCall.function.name,
tool_call_id: toolCall.id || `tool-${Date.now()}-${Math.random().toString(36).substring(2, 9)}`
};
}
}));
log.info(`Completed execution of ${toolResults.length} tools`);
return toolResults;
}
/**
* Build context from relevant notes
*/

101
src/services/llm/tools/read_note_tool.ts Normal file
View File

@ -0,0 +1,101 @@
/**
* Read Note Tool
*
* This tool allows the LLM to read the content of a specific note.
*/
import type { Tool, ToolHandler } from './tool_interfaces.js';
import log from '../../log.js';
import becca from '../../../becca/becca.js';
/**
* Definition of the read note tool
*/
export const readNoteToolDefinition: Tool = {
type: 'function',
function: {
name: 'read_note',
description: 'Read the content of a specific note by its ID',
parameters: {
type: 'object',
properties: {
noteId: {
type: 'string',
description: 'The ID of the note to read'
},
includeAttributes: {
type: 'boolean',
description: 'Whether to include note attributes in the response (default: false)'
}
},
required: ['noteId']
}
}
};
/**
* Read note tool implementation
*/
export class ReadNoteTool implements ToolHandler {
public definition: Tool = readNoteToolDefinition;
/**
* Execute the read note tool
*/
public async execute(args: { noteId: string, includeAttributes?: boolean }): Promise<string | object> {
try {
const { noteId, includeAttributes = false } = args;
log.info(`Executing read_note tool - NoteID: "${noteId}", IncludeAttributes: ${includeAttributes}`);
// Get the note from becca
const note = becca.notes[noteId];
if (!note) {
log.info(`Note with ID ${noteId} not found - returning error`);
return `Error: Note with ID ${noteId} not found`;
}
log.info(`Found note: "${note.title}" (Type: ${note.type})`);
// Get note content
const startTime = Date.now();
const content = await note.getContent();
const duration = Date.now() - startTime;
log.info(`Retrieved note content in ${duration}ms, content length: ${content?.length || 0} chars`);
// Prepare the response
const response: any = {
noteId: note.noteId,
title: note.title,
type: note.type,
content: content || ''
};
// Include attributes if requested
if (includeAttributes) {
const attributes = note.getOwnedAttributes();
log.info(`Including ${attributes.length} attributes in response`);
response.attributes = attributes.map(attr => ({
name: attr.name,
value: attr.value,
type: attr.type
}));
if (attributes.length > 0) {
// Log some example attributes
attributes.slice(0, 3).forEach((attr, index) => {
log.info(`Attribute ${index + 1}: ${attr.name}=${attr.value} (${attr.type})`);
});
}
}
return response;
} catch (error: any) {
log.error(`Error executing read_note tool: ${error.message || String(error)}`);
return `Error: ${error.message || String(error)}`;
}
}
}

95
src/services/llm/tools/search_notes_tool.ts Normal file
View File

@ -0,0 +1,95 @@
/**
* Search Notes Tool
*
* This tool allows the LLM to search for notes using semantic search.
*/
import type { Tool, ToolHandler } from './tool_interfaces.js';
import log from '../../log.js';
import aiServiceManager from '../ai_service_manager.js';
/**
* Definition of the search notes tool
*/
export const searchNotesToolDefinition: Tool = {
type: 'function',
function: {
name: 'search_notes',
description: 'Search for notes in the database using semantic search. Returns notes most semantically related to the query.',
parameters: {
type: 'object',
properties: {
query: {
type: 'string',
description: 'The search query to find semantically related notes'
},
parentNoteId: {
type: 'string',
description: 'Optional parent note ID to restrict search to a specific branch'
},
maxResults: {
type: 'number',
description: 'Maximum number of results to return (default: 5)'
}
},
required: ['query']
}
}
};
/**
* Search notes tool implementation
*/
export class SearchNotesTool implements ToolHandler {
public definition: Tool = searchNotesToolDefinition;
/**
* Execute the search notes tool
*/
public async execute(args: { query: string, parentNoteId?: string, maxResults?: number }): Promise<string | object> {
try {
const { query, parentNoteId, maxResults = 5 } = args;
log.info(`Executing search_notes tool - Query: "${query}", ParentNoteId: ${parentNoteId || 'not specified'}, MaxResults: ${maxResults}`);
// Get the vector search tool from the AI service manager
const vectorSearchTool = aiServiceManager.getVectorSearchTool();
log.info(`Retrieved vector search tool from AI service manager`);
// Execute the search
log.info(`Performing semantic search for: "${query}"`);
const searchStartTime = Date.now();
const results = await vectorSearchTool.searchNotes(query, {
parentNoteId,
maxResults
});
const searchDuration = Date.now() - searchStartTime;
log.info(`Search completed in ${searchDuration}ms, found ${results.length} matching notes`);
if (results.length > 0) {
// Log top results
results.slice(0, 3).forEach((result, index) => {
log.info(`Result ${index + 1}: "${result.title}" (similarity: ${Math.round(result.similarity * 100)}%)`);
});
} else {
log.info(`No matching notes found for query: "${query}"`);
}
// Format the results
return {
count: results.length,
results: results.map(result => ({
noteId: result.noteId,
title: result.title,
preview: result.contentPreview,
similarity: Math.round(result.similarity * 100) / 100,
parentId: result.parentId
}))
};
} catch (error: any) {
log.error(`Error executing search_notes tool: ${error.message || String(error)}`);
return `Error: ${error.message || String(error)}`;
}
}
}

36
src/services/llm/tools/tool_initializer.ts Normal file
View File

@ -0,0 +1,36 @@
/**
* Tool Initializer
*
* This module initializes all available tools for the LLM to use.
*/
import toolRegistry from './tool_registry.js';
import { SearchNotesTool } from './search_notes_tool.js';
import { ReadNoteTool } from './read_note_tool.js';
import log from '../../log.js';
/**
* Initialize all tools for the LLM
*/
export async function initializeTools(): Promise<void> {
try {
log.info('Initializing LLM tools...');
// Register basic notes tools
toolRegistry.registerTool(new SearchNotesTool());
toolRegistry.registerTool(new ReadNoteTool());
// More tools can be registered here
// Log registered tools
const toolCount = toolRegistry.getAllTools().length;
log.info(`Successfully registered ${toolCount} LLM tools`);
} catch (error: any) {
log.error(`Error initializing LLM tools: ${error.message || String(error)}`);
// Don't throw, just log the error to prevent breaking the pipeline
}
}
export default {
initializeTools
};

57
src/services/llm/tools/tool_interfaces.ts Normal file
View File

@ -0,0 +1,57 @@
/**
* Tool Interfaces
*
* This file defines the interfaces for the LLM tool calling system.
*/
/**
* Interface for a tool definition to be sent to the LLM
*/
export interface Tool {
type: 'function';
function: {
name: string;
description: string;
parameters: {
type: 'object';
properties: Record<string, ToolParameter>;
required: string[];
};
};
}
/**
* Interface for a tool parameter
*/
export interface ToolParameter {
type: string;
description: string;
enum?: string[];
}
/**
* Interface for a tool call from the LLM
*/
export interface ToolCall {
id?: string;
type?: string;
function: {
name: string;
arguments: Record<string, any> | string;
};
}
/**
* Interface for a tool handler that executes a tool
*/
export interface ToolHandler {
/**
* Tool definition to be sent to the LLM
*/
definition: Tool;
/**
* Execute the tool with the given arguments
*/
execute(args: Record<string, any>): Promise<string | object>;
}

69
src/services/llm/tools/tool_registry.ts Normal file
View File

@ -0,0 +1,69 @@
/**
* Tool Registry
*
* This file defines the registry for tools that can be called by LLMs.
*/
import type { Tool, ToolHandler } from './tool_interfaces.js';
import log from '../../log.js';
/**
* Registry for tools that can be called by LLMs
*/
export class ToolRegistry {
private static instance: ToolRegistry;
private tools: Map<string, ToolHandler> = new Map();
private constructor() {}
/**
* Get singleton instance of the tool registry
*/
public static getInstance(): ToolRegistry {
if (!ToolRegistry.instance) {
ToolRegistry.instance = new ToolRegistry();
}
return ToolRegistry.instance;
}
/**
* Register a tool with the registry
*/
public registerTool(handler: ToolHandler): void {
const name = handler.definition.function.name;
if (this.tools.has(name)) {
log.info(`Tool '${name}' already registered, replacing...`);
}
this.tools.set(name, handler);
log.info(`Registered tool: ${name}`);
}
/**
* Get a tool by name
*/
public getTool(name: string): ToolHandler | undefined {
return this.tools.get(name);
}
/**
* Get all registered tools
*/
public getAllTools(): ToolHandler[] {
return Array.from(this.tools.values());
}
/**
* Get all tool definitions for sending to LLM
*/
public getAllToolDefinitions(): Tool[] {
const toolDefs = Array.from(this.tools.values()).map(handler => handler.definition);
return toolDefs;
}
}
// Export singleton instance
const toolRegistry = ToolRegistry.getInstance();
export default toolRegistry;