Dockery/Trilium
1
0
Fork 0
mirror of https://github.com/zadam/trilium.git synced 2025-10-31 02:16:05 +01:00
Code Issues Packages Projects Releases Activity

feat(llm): remove everything to do with embeddings, part 3

Browse Source
This commit is contained in:
perf3ct
2025-06-07 18:30:46 +00:00
parent 44a2e7df21
commit 4550c12c6e
17 changed files with 44 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
apps/client/src/services/load_results.ts
View File

@@ -44,18 +44,7 @@ interface OptionRow {}
interface NoteReorderingRow {}
interface NoteEmbeddingRow {
embedId: string;
noteId: string;
providerId: string;
modelId: string;
dimension: number;
version: number;
dateCreated: string;
utcDateCreated: string;
dateModified: string;
utcDateModified: string;
}
type EntityRowMappings = {
notes: NoteRow;

2
apps/client/src/translations/en/translation.json
View File

@@ -1195,7 +1195,7 @@
"restore_provider": "Restore provider to search",
"similarity_threshold": "Similarity Threshold",
"similarity_threshold_description": "Minimum similarity score (0-1) for notes to be included in context for LLM queries",
"reprocess_started": "Embedding reprocessing started in the background",
"reprocess_index": "Rebuild Search Index",
"reprocessing_index": "Rebuilding...",
"reprocess_index_started": "Search index optimization started in the background",

4
apps/client/src/widgets/type_widgets/options/ai_settings/providers.ts
View File

@@ -6,7 +6,7 @@ import type { OpenAIModelResponse, AnthropicModelResponse, OllamaModelResponse }
export class ProviderService {
constructor(private $widget: JQuery<HTMLElement>) {
// Embedding functionality removed
// AI provider settings
}
/**
@@ -204,7 +204,7 @@ export class ProviderService {
try {
// Use the general Ollama base URL
const ollamaBaseUrl = this.$widget.find('.ollama-base-url').val() as string;
const response = await server.get<OllamaModelResponse>(`llm/providers/ollama/models?baseUrl=${encodeURIComponent(ollamaBaseUrl)}`);
if (response && response.success && response.models && response.models.length > 0) {

2
apps/client/src/widgets/type_widgets/options/ai_settings/template.ts
View File

@@ -16,7 +16,7 @@ export const TPL = `
</div>
</div>
<!-- Embedding statistics section removed -->
<!-- AI settings template -->
<div class="ai-providers-section options-section">
<h4>${t("ai_llm.provider_configuration")}</h4>

17
apps/server/src/routes/api/anthropic.ts
View File

@@ -48,17 +48,6 @@ interface AnthropicModel {
* type: string
* type:
* type: string
* embeddingModels:
* type: array
* items:
* type: object
* properties:
* id:
* type: string
* name:
* type: string
* type:
* type: string
* '500':
* description: Error listing models
* security:
@@ -90,14 +79,10 @@ async function listModels(req: Request, res: Response) {
type: 'chat'
}));
// Anthropic doesn't currently have embedding models
const embeddingModels: AnthropicModel[] = [];
// Return the models list
return {
success: true,
chatModels,
embeddingModels
chatModels
};
} catch (error: any) {
log.error(`Error listing Anthropic models: ${error.message || 'Unknown error'}`);

29
apps/server/src/routes/api/openai.ts
View File

@@ -40,17 +40,6 @@ import OpenAI from "openai";
* type: string
* type:
* type: string
* embeddingModels:
* type: array
* items:
* type: object
* properties:
* id:
* type: string
* name:
* type: string
* type:
* type: string
* '500':
* description: Error listing models
* security:
@@ -82,8 +71,7 @@ async function listModels(req: Request, res: Response) {
// Filter and categorize models
const allModels = response.data || [];
// Include all models as chat models, without filtering by specific model names
// This allows models from providers like OpenRouter to be displayed
// Include all models as chat models, excluding embedding models
const chatModels = allModels
.filter((model) =>
// Exclude models that are explicitly for embeddings
@@ -96,23 +84,10 @@ async function listModels(req: Request, res: Response) {
type: 'chat'
}));
const embeddingModels = allModels
.filter((model) =>
// Only include embedding-specific models
model.id.includes('embedding') ||
model.id.includes('embed')
)
.map((model) => ({
id: model.id,
name: model.id,
type: 'embedding'
}));
// Return the models list
return {
success: true,
chatModels,
embeddingModels
chatModels
};
} catch (error: any) {
log.error(`Error listing OpenAI models: ${error.message || 'Unknown error'}`);

2
apps/server/src/routes/api/options.ts
View File

@@ -92,7 +92,7 @@ const ALLOWED_OPTIONS = new Set<OptionNames>([
"showLoginInShareTheme",
"splitEditorOrientation",
// AI/LLM integration options (embedding options removed)
// AI/LLM integration options
"aiEnabled",
"aiTemperature",
"aiSystemPrompt",

78
apps/server/src/services/llm/constants/provider_constants.ts
View File

@@ -1,24 +1,17 @@
/**
* Configuration constants for LLM providers
*/
export const PROVIDER_CONSTANTS = {
ANTHROPIC: {
API_VERSION: '2023-06-01',
BETA_VERSION: 'messages-2023-12-15',
BASE_URL: 'https://api.anthropic.com',
DEFAULT_MODEL: 'claude-3-haiku-20240307',
// Model mapping for simplified model names to their full versions
MODEL_MAPPING: {
'claude-3.7-sonnet': 'claude-3-7-sonnet-20250219',
'claude-3.5-sonnet': 'claude-3-5-sonnet-20241022',
'claude-3.5-haiku': 'claude-3-5-haiku-20241022',
'claude-3-opus': 'claude-3-opus-20240229',
'claude-3-sonnet': 'claude-3-sonnet-20240229',
'claude-3-haiku': 'claude-3-haiku-20240307',
'claude-2': 'claude-2.1'
},
// These are the currently available models from Anthropic
DEFAULT_MODEL: 'claude-3-5-sonnet-20241022',
API_VERSION: '2023-06-01',
BETA_VERSION: undefined,
CONTEXT_WINDOW: 200000,
AVAILABLE_MODELS: [
{
id: 'claude-3-7-sonnet-20250219',
name: 'Claude 3.7 Sonnet',
id: 'claude-3-5-sonnet-20250106',
name: 'Claude 3.5 Sonnet (New)',
description: 'Most intelligent model with hybrid reasoning capabilities',
maxTokens: 8192
},
@@ -64,12 +57,7 @@ export const PROVIDER_CONSTANTS = {
OPENAI: {
BASE_URL: 'https://api.openai.com/v1',
DEFAULT_MODEL: 'gpt-3.5-turbo',
DEFAULT_EMBEDDING_MODEL: 'text-embedding-ada-002',
CONTEXT_WINDOW: 16000,
EMBEDDING_DIMENSIONS: {
ADA: 1536,
DEFAULT: 1536
},
AVAILABLE_MODELS: [
{
id: 'gpt-4o',
@@ -132,51 +120,6 @@ export const LLM_CONSTANTS = {
DEFAULT: 6000
},
// Embedding dimensions (verify these with your actual models)
EMBEDDING_DIMENSIONS: {
OLLAMA: {
DEFAULT: 384,
NOMIC: 768,
MISTRAL: 1024
},
OPENAI: {
ADA: 1536,
DEFAULT: 1536
},
ANTHROPIC: {
CLAUDE: 1024,
DEFAULT: 1024
},
VOYAGE: {
DEFAULT: 1024
}
},
// Model-specific embedding dimensions for Ollama models
OLLAMA_MODEL_DIMENSIONS: {
"llama3": 8192,
"llama3.1": 8192,
"mistral": 8192,
"nomic": 768,
"mxbai": 1024,
"nomic-embed-text": 768,
"mxbai-embed-large": 1024,
"default": 384
},
// Model-specific context windows for Ollama models
OLLAMA_MODEL_CONTEXT_WINDOWS: {
"llama3": 8192,
"llama3.1": 8192,
"llama3.2": 8192,
"mistral": 8192,
"nomic": 32768,
"mxbai": 32768,
"nomic-embed-text": 32768,
"mxbai-embed-large": 32768,
"default": 8192
},
// Batch size configuration
BATCH_SIZE: {
OPENAI: 10, // OpenAI can handle larger batches efficiently
@@ -189,8 +132,7 @@ export const LLM_CONSTANTS = {
CHUNKING: {
DEFAULT_SIZE: 1500,
OLLAMA_SIZE: 1000,
DEFAULT_OVERLAP: 100,
MAX_SIZE_FOR_SINGLE_EMBEDDING: 5000
DEFAULT_OVERLAP: 100
},
// Search/similarity thresholds

2
apps/server/src/services/llm/context/modules/cache_manager.ts
View File

@@ -6,7 +6,7 @@ import type { ICacheManager, CachedNoteData, CachedQueryResults } from '../../in
* Provides a centralized caching system to avoid redundant operations
*/
export class CacheManager implements ICacheManager {
// Cache for recently used context to avoid repeated embedding lookups
// Cache for recently used context to avoid repeated lookups
private noteDataCache = new Map<string, CachedNoteData<unknown>>();
// Cache for recently used queries

38
apps/server/src/services/llm/context/modules/provider_manager.ts
View File

@@ -1,37 +1 @@
import log from '../../../log.js';
/**
* Manages embedding providers for context services
* Simplified since embedding functionality has been removed
*/
export class ProviderManager {
/**
* Get the selected embedding provider based on user settings
* Returns null since embeddings have been removed
*/
async getSelectedEmbeddingProvider(): Promise<null> {
log.info('Embedding providers have been removed - returning null');
return null;
}
/**
* Get all enabled embedding providers
* Returns empty array since embeddings have been removed
*/
async getEnabledEmbeddingProviders(): Promise<never[]> {
log.info('Embedding providers have been removed - returning empty array');
return [];
}
/**
* Check if embedding providers are available
* Returns false since embeddings have been removed
*/
isEmbeddingAvailable(): boolean {
return false;
}
}
// Export singleton instance
export const providerManager = new ProviderManager();
export default providerManager;
// This file has been removed as embedding functionality has been completely removed from the codebase

23
apps/server/src/services/llm/context/services/context_service.ts
View File

@@ -11,7 +11,6 @@
*/
import log from '../../../log.js';
import providerManager from '../modules/provider_manager.js';
import cacheManager from '../modules/cache_manager.js';
import queryProcessor from './query_processor.js';
import contextFormatter from '../modules/context_formatter.js';
@@ -56,17 +55,11 @@ export class ContextService {
this.initPromise = (async () => {
try {
// Initialize provider
const provider = await providerManager.getSelectedEmbeddingProvider();
if (!provider) {
throw new Error(`No embedding provider available. Could not initialize context service.`);
}
// Agent tools are already initialized in the AIServiceManager constructor
// No need to initialize them again
this.initialized = true;
log.info(`Context service initialized - embeddings disabled`);
log.info(`Context service initialized`);
} catch (error: unknown) {
const errorMessage = error instanceof Error ? error.message : String(error);
log.error(`Failed to initialize context service: ${errorMessage}`);
@@ -177,10 +170,9 @@ export class ContextService {
}
}
// Step 3: Find relevant notes using basic text search (since embeddings are removed)
// This will use traditional note search instead of vector similarity
log.info("Using traditional search instead of embedding-based search");
// Step 3: Find relevant notes using traditional search
log.info("Using traditional search for note discovery");
// Use fallback context based on the context note if provided
if (contextNoteId) {
try {
@@ -194,7 +186,7 @@ export class ContextService {
similarity: 1.0,
content: content || ""
}];
// Add child notes as additional context
const childNotes = contextNote.getChildNotes().slice(0, maxResults - 1);
for (const child of childNotes) {
@@ -215,13 +207,10 @@ export class ContextService {
log.info(`Final combined results: ${relevantNotes.length} relevant notes`);
// Step 4: Build context from the notes
const provider = await providerManager.getSelectedEmbeddingProvider();
const providerId = 'default'; // Provider is always null since embeddings removed
const context = await contextFormatter.buildContextFromNotes(
relevantNotes,
userQuestion,
providerId
'default'
);
// Step 5: Add agent tools context if requested

1
apps/server/src/services/llm/interfaces/context_interfaces.ts
View File

@@ -60,7 +60,6 @@ export interface IContextFormatter {
*/
export interface ILLMService {
sendMessage(message: string, options?: Record<string, unknown>): Promise<string>;
generateEmbedding?(text: string): Promise<number[]>;
streamMessage?(message: string, callback: (text: string) => void, options?: Record<string, unknown>): Promise<string>;
}

10
apps/server/src/services/llm/interfaces/error_interfaces.ts
View File

@@ -36,16 +36,6 @@ export interface OllamaError extends LLMServiceError {
code?: string;
}
/**
* Embedding-specific error interface
*/
export interface EmbeddingError extends LLMServiceError {
provider: string;
model?: string;
batchSize?: number;
isRetryable: boolean;
}
/**
* Guard function to check if an error is a specific type of error
*/

8
apps/server/src/services/llm/model_capabilities_service.ts
View File

@@ -6,7 +6,7 @@ import aiServiceManager from './ai_service_manager.js';
/**
* Service for fetching and caching model capabilities
* Simplified to only handle chat models since embeddings have been removed
* Handles chat model capabilities
*/
export class ModelCapabilitiesService {
// Cache model capabilities
@@ -24,10 +24,10 @@ export class ModelCapabilitiesService {
// Get from static definitions or service
const capabilities = await this.fetchChatModelCapabilities(modelName);
// Cache the result
this.capabilitiesCache.set(`chat:${modelName}`, capabilities);
return capabilities;
}
@@ -82,4 +82,4 @@ export class ModelCapabilitiesService {
// Export singleton instance
export const modelCapabilitiesService = new ModelCapabilitiesService();
export default modelCapabilitiesService;
export default modelCapabilitiesService;

8
apps/server/src/services/llm/pipeline/chat_pipeline.ts
View File

@@ -8,7 +8,7 @@ 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';
// VectorSearchStage removed along with embedding functionality
// Traditional search is used instead of vector search
import toolRegistry from '../tools/tool_registry.js';
import toolInitializer from '../tools/tool_initializer.js';
import log from '../../log.js';
@@ -29,7 +29,7 @@ export class ChatPipeline {
llmCompletion: LLMCompletionStage;
responseProcessing: ResponseProcessingStage;
toolCalling: ToolCallingStage;
// vectorSearch removed with embedding functionality
// traditional search is used instead of vector search
};
config: ChatPipelineConfig;
@@ -50,7 +50,7 @@ export class ChatPipeline {
llmCompletion: new LLMCompletionStage(),
responseProcessing: new ResponseProcessingStage(),
toolCalling: new ToolCallingStage(),
// vectorSearch removed with embedding functionality
// traditional search is used instead of vector search
};
// Set default configuration values
@@ -307,7 +307,7 @@ export class ChatPipeline {
// Forward to callback with original chunk data in case it contains additional information
streamCallback(processedChunk.text, processedChunk.done, chunk);
// Mark that we have streamed content to prevent duplication
hasStreamedContent = true;
});

18
apps/server/src/services/llm/provider_validation.ts
View File

@@ -1,6 +1,6 @@
/**
* Provider Validation Service
*
*
* Validates AI provider configurations before initializing the chat system.
* This prevents startup errors when AI is enabled but providers are misconfigured.
*/
@@ -58,7 +58,7 @@ async function checkChatProviderConfigs(result: ProviderValidationResult): Promi
// Check OpenAI chat provider
const openaiApiKey = await options.getOption('openaiApiKey');
const openaiBaseUrl = await options.getOption('openaiBaseUrl');
if (openaiApiKey || openaiBaseUrl) {
result.validChatProviders.push('openai');
log.info("OpenAI chat provider configuration available");
@@ -83,16 +83,6 @@ async function checkChatProviderConfigs(result: ProviderValidationResult): Promi
}
}
/**
* Check if we have at least one valid embedding provider available
* Returns false since embeddings have been removed
*/
export async function getEmbeddingProviderAvailability(): Promise<boolean> {
log.info("Embedding providers have been removed, returning false");
return false;
}
export default {
validateProviders,
getEmbeddingProviderAvailability
};
validateProviders
};

10
apps/server/src/services/llm/tools/relationship_tool.ts
View File

@@ -11,7 +11,7 @@ import attributes from '../../attributes.js';
import aiServiceManager from '../ai_service_manager.js';
import { SEARCH_CONSTANTS } from '../constants/search_constants.js';
import searchService from '../../search/services/search.js';
// Define types locally since embeddings are no longer available
// Define types locally for relationship tool
interface Backlink {
noteId: string;
title: string;
@@ -278,7 +278,7 @@ export class RelationshipTool implements ToolHandler {
// Create search queries from the note title and content
const searchQueries = [title];
// Extract key terms from content if available
if (content && typeof content === 'string') {
// Extract meaningful words from content (filter out common words)
@@ -288,7 +288,7 @@ export class RelationshipTool implements ToolHandler {
.filter(word => word.length > 3)
.filter(word => !/^(the|and|but|for|are|from|they|been|have|this|that|with|will|when|where|what|how)$/.test(word))
.slice(0, 10); // Take first 10 meaningful words
if (contentWords.length > 0) {
searchQueries.push(contentWords.join(' '));
}
@@ -301,11 +301,11 @@ export class RelationshipTool implements ToolHandler {
for (const query of searchQueries) {
try {
const results = searchService.searchNotes(query, {
const results = searchService.searchNotes(query, {
includeArchivedNotes: false,
fastSearch: false // Use full search for better results
});
// Add results to our map (avoiding duplicates)
for (const note of results.slice(0, limit * 2)) { // Get more to account for duplicates
if (note.noteId !== sourceNote.noteId && !note.isDeleted) {