mirror of
https://github.com/TriliumNext/Notes.git
synced 2025-09-24 21:41:30 +08:00
format decomp files
This commit is contained in:
perf3ct
parent
19c36b18a6
commit
d83cce88cb
@ -100,7 +100,8 @@ Format your answer as a valid JSON array without markdown code blocks, like this
|
||||
maxTokens: SEARCH_CONSTANTS.LIMITS.QUERY_PROCESSOR_MAX_TOKENS,
|
||||
bypassFormatter: true,
|
||||
expectsJsonResponse: true,
|
||||
_bypassContextProcessing: true // Prevent recursive calls
|
||||
_bypassContextProcessing: true, // Prevent recursive calls
|
||||
enableTools: false // Explicitly disable tools for this request
|
||||
};
|
||||
|
||||
// Get the response from the LLM
|
||||
@ -167,69 +168,33 @@ Format your answer as a valid JSON array without markdown code blocks, like this
|
||||
// Try to get LLM service if not provided
|
||||
const service = llmService || await this.getLLMService();
|
||||
|
||||
// For when no LLM service is available, use the basic approach
|
||||
// If no LLM service is available, use basic decomposition
|
||||
if (!service) {
|
||||
if (!service) {
|
||||
log.info(`No LLM service available for query decomposition, using original query`);
|
||||
}
|
||||
|
||||
log.info(`Using basic decomposition approach (complexity: ${complexity})`);
|
||||
|
||||
const mainSubQuery = {
|
||||
id: this.generateSubQueryId(),
|
||||
text: query,
|
||||
reason: "Direct question that can be answered without decomposition",
|
||||
isAnswered: false
|
||||
};
|
||||
|
||||
// Add a generic exploration query for context
|
||||
const genericQuery = {
|
||||
id: this.generateSubQueryId(),
|
||||
text: `What information is related to ${query}?`,
|
||||
reason: "General exploration to find related content",
|
||||
isAnswered: false
|
||||
};
|
||||
|
||||
return {
|
||||
originalQuery: query,
|
||||
subQueries: [mainSubQuery, genericQuery],
|
||||
status: 'pending',
|
||||
complexity
|
||||
};
|
||||
log.info(`No LLM service available for query decomposition, using original query`);
|
||||
return this.createBasicDecomposition(query, complexity);
|
||||
}
|
||||
|
||||
// For when the LLM available, we can use more advanced decomposition
|
||||
if (service) {
|
||||
try {
|
||||
// Try to use LLM for advanced decomposition
|
||||
log.info(`Using advanced LLM-based decomposition for complex query (complexity: ${complexity})`);
|
||||
const enhancedSubQueries = await this.createLLMSubQueries(query, context, service);
|
||||
// With LLM service available, always use advanced decomposition regardless of complexity
|
||||
try {
|
||||
log.info(`Using advanced LLM-based decomposition for query (complexity: ${complexity})`);
|
||||
const enhancedSubQueries = await this.createLLMSubQueries(query, context, service);
|
||||
|
||||
if (enhancedSubQueries && enhancedSubQueries.length > 0) {
|
||||
log.info(`LLM decomposed query into ${enhancedSubQueries.length} sub-queries: ${JSON.stringify(enhancedSubQueries)}`);
|
||||
return {
|
||||
originalQuery: query,
|
||||
subQueries: enhancedSubQueries,
|
||||
status: 'pending',
|
||||
complexity
|
||||
};
|
||||
}
|
||||
} catch (error: any) {
|
||||
log.error(`Error during LLM-based decomposition: ${error.message}, falling back to basic decomposition`);
|
||||
// Continue to fallback with basic decomposition
|
||||
if (enhancedSubQueries && enhancedSubQueries.length > 0) {
|
||||
log.info(`LLM decomposed query into ${enhancedSubQueries.length} sub-queries`);
|
||||
return {
|
||||
originalQuery: query,
|
||||
subQueries: enhancedSubQueries,
|
||||
status: 'pending',
|
||||
complexity
|
||||
};
|
||||
}
|
||||
} catch (error: any) {
|
||||
log.error(`Error during LLM-based decomposition: ${error.message}, falling back to basic decomposition`);
|
||||
// Fall through to basic decomposition
|
||||
}
|
||||
|
||||
// Fallback to basic decomposition
|
||||
const subQueries = this.createSubQueries(query, context);
|
||||
log.info(`Decomposed query into ${subQueries.length} sub-queries`);
|
||||
|
||||
return {
|
||||
originalQuery: query,
|
||||
subQueries,
|
||||
status: 'pending',
|
||||
complexity
|
||||
};
|
||||
return this.createBasicDecomposition(query, complexity);
|
||||
} catch (error: any) {
|
||||
log.error(`Error decomposing query: ${error.message}`);
|
||||
|
||||
@ -248,6 +213,39 @@ Format your answer as a valid JSON array without markdown code blocks, like this
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Create a basic decomposition of a query without using LLM
|
||||
*
|
||||
* @param query The original query
|
||||
* @param complexity The assessed complexity
|
||||
* @returns A basic decomposed query
|
||||
*/
|
||||
private createBasicDecomposition(query: string, complexity: number): DecomposedQuery {
|
||||
log.info(`Using basic decomposition approach (complexity: ${complexity})`);
|
||||
|
||||
const mainSubQuery = {
|
||||
id: this.generateSubQueryId(),
|
||||
text: query,
|
||||
reason: "Direct question that can be answered without decomposition",
|
||||
isAnswered: false
|
||||
};
|
||||
|
||||
// Add a generic exploration query for context
|
||||
const genericQuery = {
|
||||
id: this.generateSubQueryId(),
|
||||
text: `What information is related to ${query}?`,
|
||||
reason: "General exploration to find related content",
|
||||
isAnswered: false
|
||||
};
|
||||
|
||||
return {
|
||||
originalQuery: query,
|
||||
subQueries: [mainSubQuery, genericQuery],
|
||||
status: 'pending',
|
||||
complexity
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Use LLM to create advanced sub-queries from a complex query
|
||||
*
|
||||
@ -267,25 +265,42 @@ Format your answer as a valid JSON array without markdown code blocks, like this
|
||||
}
|
||||
|
||||
try {
|
||||
// Build a prompt from existing templates in the constants
|
||||
const contextPart = context ? `\nContext: ${context}` : '';
|
||||
// Create a much better prompt for more effective query decomposition
|
||||
const prompt = `Decompose the following query into 3-5 specific search queries that would help find comprehensive information.
|
||||
|
||||
// Use existing templates from QUERY_DECOMPOSITION_STRINGS to build the prompt
|
||||
const prompt = `I need to break down a complex query into sub-queries.
|
||||
Query: ${query}${contextPart}
|
||||
Your task is to identify the main concepts and break them down into specific, targeted search queries.
|
||||
|
||||
Please analyze this query and identify the key aspects that need to be addressed.`;
|
||||
DO NOT simply rephrase the original query or create a generic "what's related to X" pattern.
|
||||
DO create specific queries that explore different aspects of the topic.
|
||||
|
||||
For example:
|
||||
If the query is "How does Docker compare to Kubernetes?", good sub-queries would be:
|
||||
- "Docker container architecture and features"
|
||||
- "Kubernetes container orchestration capabilities"
|
||||
- "Docker vs Kubernetes performance comparison"
|
||||
- "When to use Docker versus Kubernetes"
|
||||
|
||||
Format your response as a JSON array of objects with 'text' and 'reason' properties.
|
||||
Example: [
|
||||
{"text": "Docker container architecture", "reason": "Understanding Docker's core technology"},
|
||||
{"text": "Kubernetes orchestration features", "reason": "Exploring Kubernetes' main capabilities"}
|
||||
]
|
||||
|
||||
${context ? `\nContext: ${context}` : ''}
|
||||
|
||||
Query: ${query}`;
|
||||
|
||||
const messages = [
|
||||
{ role: "system" as const, content: prompt }
|
||||
];
|
||||
|
||||
const options = {
|
||||
temperature: SEARCH_CONSTANTS.TEMPERATURE.QUERY_PROCESSOR,
|
||||
temperature: 0.7, // Higher temperature for more creative decomposition
|
||||
maxTokens: SEARCH_CONSTANTS.LIMITS.QUERY_PROCESSOR_MAX_TOKENS,
|
||||
bypassFormatter: true,
|
||||
expectsJsonResponse: true,
|
||||
_bypassContextProcessing: true // Prevent recursive calls
|
||||
_bypassContextProcessing: true, // Prevent recursive calls
|
||||
enableTools: false // Explicitly disable tools for this request
|
||||
};
|
||||
|
||||
// Get the response from the LLM
|
||||
@ -300,6 +315,11 @@ Please analyze this query and identify the key aspects that need to be addressed
|
||||
reason?: string;
|
||||
}
|
||||
|
||||
// Log the response for debugging
|
||||
log.info(`Received response from LLM for query decomposition, extracting JSON...`);
|
||||
|
||||
log.info(`Response: ${responseText}`);
|
||||
|
||||
// Extract JSON from the response
|
||||
const extractedData = JsonExtractor.extract<RawSubQuery[]>(responseText, {
|
||||
extractArrays: true,
|
||||
@ -307,15 +327,74 @@ Please analyze this query and identify the key aspects that need to be addressed
|
||||
useFallbacks: true
|
||||
});
|
||||
|
||||
if (Array.isArray(extractedData) && extractedData.length > 0) {
|
||||
// Convert the raw data to SubQuery objects
|
||||
return extractedData.map(item => ({
|
||||
id: this.generateSubQueryId(),
|
||||
text: item.text,
|
||||
reason: item.reason || "Sub-aspect of the main question",
|
||||
isAnswered: false
|
||||
}));
|
||||
// Validate the extracted data
|
||||
if (!Array.isArray(extractedData)) {
|
||||
log.error(`Failed to extract array from LLM response, got: ${typeof extractedData}`);
|
||||
return this.createSubQueries(query, context);
|
||||
}
|
||||
|
||||
if (extractedData.length === 0) {
|
||||
log.error(`Extracted array is empty, falling back to basic decomposition`);
|
||||
return this.createSubQueries(query, context);
|
||||
}
|
||||
|
||||
log.info(`Successfully extracted ${extractedData.length} items using regex pattern`);
|
||||
|
||||
// Validate each sub-query to ensure it has a text property
|
||||
const validSubQueries = extractedData.filter(item => {
|
||||
if (!item || typeof item !== 'object') {
|
||||
log.error(`Invalid sub-query item: ${JSON.stringify(item)}`);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!item.text || typeof item.text !== 'string') {
|
||||
log.error(`Sub-query missing text property: ${JSON.stringify(item)}`);
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
});
|
||||
|
||||
if (validSubQueries.length === 0) {
|
||||
log.error(`No valid sub-queries found after validation, falling back to basic decomposition`);
|
||||
return this.createSubQueries(query, context);
|
||||
}
|
||||
|
||||
if (validSubQueries.length < extractedData.length) {
|
||||
log.info(`Some invalid sub-queries were filtered out: ${extractedData.length} -> ${validSubQueries.length}`);
|
||||
}
|
||||
|
||||
// Convert the raw data to SubQuery objects
|
||||
let subQueries = validSubQueries.map(item => ({
|
||||
id: this.generateSubQueryId(),
|
||||
text: item.text,
|
||||
reason: item.reason || "Sub-aspect of the main question",
|
||||
isAnswered: false
|
||||
}));
|
||||
|
||||
// Make sure we have at least the original query
|
||||
const hasOriginalQuery = subQueries.some(sq => {
|
||||
// Check if either sq.text or query is null/undefined before using toLowerCase
|
||||
if (!sq.text) return false;
|
||||
const sqText = sq.text.toLowerCase();
|
||||
const originalQuery = query.toLowerCase();
|
||||
|
||||
return sqText.includes(originalQuery) || originalQuery.includes(sqText);
|
||||
});
|
||||
|
||||
if (!hasOriginalQuery) {
|
||||
subQueries.unshift({
|
||||
id: this.generateSubQueryId(),
|
||||
text: query,
|
||||
reason: "Original query",
|
||||
isAnswered: false
|
||||
});
|
||||
}
|
||||
|
||||
// Log the extracted sub-queries for debugging
|
||||
log.info(`Successfully extracted ${subQueries.length} sub-queries from LLM response`);
|
||||
|
||||
return subQueries;
|
||||
} catch (error: any) {
|
||||
log.error(`Error extracting sub-queries from LLM response: ${error.message}`);
|
||||
// Fall through to traditional decomposition
|
||||
|
||||
@ -9,139 +9,139 @@ import type BAttribute from '../../../becca/entities/battribute.js';
|
||||
* Interface for the AI service used by agent tools
|
||||
*/
|
||||
export interface LLMServiceInterface {
|
||||
generateChatCompletion(messages: Array<{
|
||||
role: 'user' | 'assistant' | 'system';
|
||||
content: string;
|
||||
}>, options?: {
|
||||
temperature?: number;
|
||||
maxTokens?: number;
|
||||
model?: string;
|
||||
stream?: boolean;
|
||||
systemPrompt?: string;
|
||||
}): Promise<ChatResponse>;
|
||||
generateChatCompletion(messages: Array<{
|
||||
role: 'user' | 'assistant' | 'system';
|
||||
content: string;
|
||||
}>, options?: {
|
||||
temperature?: number;
|
||||
maxTokens?: number;
|
||||
model?: string;
|
||||
stream?: boolean;
|
||||
systemPrompt?: string;
|
||||
}): Promise<ChatResponse>;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for tool initialization
|
||||
*/
|
||||
export interface AgentToolInitializationParams {
|
||||
aiServiceManager: LLMServiceInterface;
|
||||
aiServiceManager: LLMServiceInterface;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for agent tool manager
|
||||
*/
|
||||
export interface IAgentToolsManager {
|
||||
initialize(aiServiceManager: LLMServiceInterface): Promise<void>;
|
||||
isInitialized(): boolean;
|
||||
getAllTools(): {
|
||||
vectorSearch: IVectorSearchTool;
|
||||
noteNavigator: INoteNavigatorTool;
|
||||
queryDecomposition: IQueryDecompositionTool;
|
||||
contextualThinking: IContextualThinkingTool;
|
||||
};
|
||||
getVectorSearchTool(): IVectorSearchTool;
|
||||
getNoteNavigatorTool(): INoteNavigatorTool;
|
||||
getQueryDecompositionTool(): IQueryDecompositionTool;
|
||||
getContextualThinkingTool(): IContextualThinkingTool;
|
||||
initialize(aiServiceManager: LLMServiceInterface): Promise<void>;
|
||||
isInitialized(): boolean;
|
||||
getAllTools(): {
|
||||
vectorSearch: IVectorSearchTool;
|
||||
noteNavigator: INoteNavigatorTool;
|
||||
queryDecomposition: IQueryDecompositionTool;
|
||||
contextualThinking: IContextualThinkingTool;
|
||||
};
|
||||
getVectorSearchTool(): IVectorSearchTool;
|
||||
getNoteNavigatorTool(): INoteNavigatorTool;
|
||||
getQueryDecompositionTool(): IQueryDecompositionTool;
|
||||
getContextualThinkingTool(): IContextualThinkingTool;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for context service used by vector search
|
||||
*/
|
||||
export interface IContextService {
|
||||
findRelevantNotesMultiQuery(queries: string[], contextNoteId: string | null, limit: number): Promise<VectorSearchResult[]>;
|
||||
processQuery(userQuestion: string, llmService: LLMServiceInterface, contextNoteId: string | null, showThinking: boolean): Promise<{
|
||||
context: string;
|
||||
sources: Array<{
|
||||
noteId: string;
|
||||
title: string;
|
||||
similarity: number;
|
||||
findRelevantNotesMultiQuery(queries: string[], contextNoteId: string | null, limit: number): Promise<VectorSearchResult[]>;
|
||||
processQuery(userQuestion: string, llmService: LLMServiceInterface, contextNoteId: string | null, showThinking: boolean): Promise<{
|
||||
context: string;
|
||||
sources: Array<{
|
||||
noteId: string;
|
||||
title: string;
|
||||
similarity: number;
|
||||
}>;
|
||||
thinking?: string;
|
||||
}>;
|
||||
thinking?: string;
|
||||
}>;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for vector search tool
|
||||
*/
|
||||
export interface IVectorSearchTool {
|
||||
setContextService(contextService: IContextService): void;
|
||||
search(
|
||||
query: string,
|
||||
contextNoteId?: string,
|
||||
searchOptions?: {
|
||||
limit?: number;
|
||||
threshold?: number;
|
||||
includeContent?: boolean;
|
||||
}
|
||||
): Promise<VectorSearchResult[]>;
|
||||
searchNotes(query: string, options?: {
|
||||
parentNoteId?: string;
|
||||
maxResults?: number;
|
||||
similarityThreshold?: number;
|
||||
}): Promise<VectorSearchResult[]>;
|
||||
searchContentChunks(query: string, options?: {
|
||||
noteId?: string;
|
||||
maxResults?: number;
|
||||
similarityThreshold?: number;
|
||||
}): Promise<VectorSearchResult[]>;
|
||||
explainResults(query: string, results: VectorSearchResult[]): string;
|
||||
setContextService(contextService: IContextService): void;
|
||||
search(
|
||||
query: string,
|
||||
contextNoteId?: string,
|
||||
searchOptions?: {
|
||||
limit?: number;
|
||||
threshold?: number;
|
||||
includeContent?: boolean;
|
||||
}
|
||||
): Promise<VectorSearchResult[]>;
|
||||
searchNotes(query: string, options?: {
|
||||
parentNoteId?: string;
|
||||
maxResults?: number;
|
||||
similarityThreshold?: number;
|
||||
}): Promise<VectorSearchResult[]>;
|
||||
searchContentChunks(query: string, options?: {
|
||||
noteId?: string;
|
||||
maxResults?: number;
|
||||
similarityThreshold?: number;
|
||||
}): Promise<VectorSearchResult[]>;
|
||||
explainResults(query: string, results: VectorSearchResult[]): string;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for note navigator tool
|
||||
*/
|
||||
export interface INoteNavigatorTool {
|
||||
getNoteInfo(noteId: string): NoteInfo | null;
|
||||
getNotePathsFromRoot(noteId: string): NotePathInfo[];
|
||||
getNoteHierarchy(noteId: string, depth?: number): NoteHierarchyLevel | null;
|
||||
getNoteAttributes(noteId: string): BAttribute[];
|
||||
findPathBetweenNotes(fromNoteId: string, toNoteId: string): NotePathInfo | null;
|
||||
searchNotesByTitle(searchTerm: string, limit?: number): NoteInfo[];
|
||||
getNoteClones(noteId: string): Promise<NoteInfo[]>;
|
||||
getNoteContextDescription(noteId: string): Promise<string>;
|
||||
getNoteStructure(noteId: string): Promise<{
|
||||
noteId: string;
|
||||
title: string;
|
||||
type: string;
|
||||
childCount: number;
|
||||
attributes: Array<{name: string, value: string}>;
|
||||
parentPath: Array<{title: string, noteId: string}>;
|
||||
}>;
|
||||
getChildNotes(noteId: string, limit?: number): Promise<Array<{noteId: string, title: string}>>;
|
||||
getParentNotes(noteId: string): Promise<Array<{noteId: string, title: string}>>;
|
||||
getLinkedNotes(noteId: string, limit?: number): Promise<Array<{noteId: string, title: string, direction: 'from'|'to'}>>;
|
||||
getNotePath(noteId: string): Promise<string>;
|
||||
getNoteInfo(noteId: string): NoteInfo | null;
|
||||
getNotePathsFromRoot(noteId: string): NotePathInfo[];
|
||||
getNoteHierarchy(noteId: string, depth?: number): NoteHierarchyLevel | null;
|
||||
getNoteAttributes(noteId: string): BAttribute[];
|
||||
findPathBetweenNotes(fromNoteId: string, toNoteId: string): NotePathInfo | null;
|
||||
searchNotesByTitle(searchTerm: string, limit?: number): NoteInfo[];
|
||||
getNoteClones(noteId: string): Promise<NoteInfo[]>;
|
||||
getNoteContextDescription(noteId: string): Promise<string>;
|
||||
getNoteStructure(noteId: string): Promise<{
|
||||
noteId: string;
|
||||
title: string;
|
||||
type: string;
|
||||
childCount: number;
|
||||
attributes: Array<{ name: string, value: string }>;
|
||||
parentPath: Array<{ title: string, noteId: string }>;
|
||||
}>;
|
||||
getChildNotes(noteId: string, limit?: number): Promise<Array<{ noteId: string, title: string }>>;
|
||||
getParentNotes(noteId: string): Promise<Array<{ noteId: string, title: string }>>;
|
||||
getLinkedNotes(noteId: string, limit?: number): Promise<Array<{ noteId: string, title: string, direction: 'from' | 'to' }>>;
|
||||
getNotePath(noteId: string): Promise<string>;
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for query decomposition tool
|
||||
*/
|
||||
export interface IQueryDecompositionTool {
|
||||
decomposeQuery(query: string, context?: string): DecomposedQuery;
|
||||
updateSubQueryAnswer(decomposedQuery: DecomposedQuery, subQueryId: string, answer: string): DecomposedQuery;
|
||||
synthesizeAnswer(decomposedQuery: DecomposedQuery): string;
|
||||
getQueryStatus(decomposedQuery: DecomposedQuery): string;
|
||||
assessQueryComplexity(query: string): number;
|
||||
generateSubQueryId(): string;
|
||||
createSubQueries(query: string, context?: string): SubQuery[];
|
||||
decomposeQuery(query: string, context?: string): DecomposedQuery;
|
||||
updateSubQueryAnswer(decomposedQuery: DecomposedQuery, subQueryId: string, answer: string): DecomposedQuery;
|
||||
synthesizeAnswer(decomposedQuery: DecomposedQuery): string;
|
||||
getQueryStatus(decomposedQuery: DecomposedQuery): string;
|
||||
assessQueryComplexity(query: string): number;
|
||||
generateSubQueryId(): string;
|
||||
createSubQueries(query: string, context?: string): SubQuery[];
|
||||
}
|
||||
|
||||
/**
|
||||
* Interface for contextual thinking tool
|
||||
*/
|
||||
export interface IContextualThinkingTool {
|
||||
startThinking(query: string): string;
|
||||
addThinkingStep(
|
||||
processId: string,
|
||||
step: Omit<ThinkingStep, 'id'>,
|
||||
parentId?: string
|
||||
): string;
|
||||
completeThinking(processId?: string): ThinkingProcess | null;
|
||||
getThinkingProcess(processId: string): ThinkingProcess | null;
|
||||
getActiveThinkingProcess(): ThinkingProcess | null;
|
||||
visualizeThinking(thinkingId: string): string;
|
||||
getThinkingSummary(thinkingId: string): string;
|
||||
resetActiveThinking(): void;
|
||||
startThinking(query: string): string;
|
||||
addThinkingStep(
|
||||
processId: string,
|
||||
step: Omit<ThinkingStep, 'id'>,
|
||||
parentId?: string
|
||||
): string;
|
||||
completeThinking(processId?: string): ThinkingProcess | null;
|
||||
getThinkingProcess(processId: string): ThinkingProcess | null;
|
||||
getActiveThinkingProcess(): ThinkingProcess | null;
|
||||
visualizeThinking(thinkingId: string): string;
|
||||
getThinkingSummary(thinkingId: string): string;
|
||||
resetActiveThinking(): void;
|
||||
}
|
||||
|
||||
@ -5,19 +5,19 @@ import type { ToolCall } from '../tools/tool_interfaces.js';
|
||||
* Model metadata interface to track provider information
|
||||
*/
|
||||
export interface ModelMetadata {
|
||||
// The provider that supports this model
|
||||
provider: 'openai' | 'anthropic' | 'ollama' | 'local';
|
||||
// The actual model identifier used by the provider's API
|
||||
modelId: string;
|
||||
// Display name for UI (optional)
|
||||
displayName?: string;
|
||||
// Model capabilities
|
||||
capabilities?: {
|
||||
contextWindow?: number;
|
||||
supportsTools?: boolean;
|
||||
supportsVision?: boolean;
|
||||
supportsStreaming?: boolean;
|
||||
};
|
||||
// The provider that supports this model
|
||||
provider: 'openai' | 'anthropic' | 'ollama' | 'local';
|
||||
// The actual model identifier used by the provider's API
|
||||
modelId: string;
|
||||
// Display name for UI (optional)
|
||||
displayName?: string;
|
||||
// Model capabilities
|
||||
capabilities?: {
|
||||
contextWindow?: number;
|
||||
supportsTools?: boolean;
|
||||
supportsVision?: boolean;
|
||||
supportsStreaming?: boolean;
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
@ -25,196 +25,196 @@ export interface ModelMetadata {
|
||||
* but not necessarily sent directly to APIs
|
||||
*/
|
||||
export interface ProviderConfig {
|
||||
// Internal configuration
|
||||
systemPrompt?: string;
|
||||
// Provider metadata for model routing
|
||||
providerMetadata?: ModelMetadata;
|
||||
// Internal configuration
|
||||
systemPrompt?: string;
|
||||
// Provider metadata for model routing
|
||||
providerMetadata?: ModelMetadata;
|
||||
}
|
||||
|
||||
/**
|
||||
* OpenAI-specific options, structured to match the OpenAI API
|
||||
*/
|
||||
export interface OpenAIOptions extends ProviderConfig {
|
||||
// Connection settings (not sent to API)
|
||||
apiKey: string;
|
||||
baseUrl: string;
|
||||
// Connection settings (not sent to API)
|
||||
apiKey: string;
|
||||
baseUrl: string;
|
||||
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: Message[];
|
||||
temperature?: number;
|
||||
max_tokens?: number;
|
||||
stream?: boolean;
|
||||
top_p?: number;
|
||||
frequency_penalty?: number;
|
||||
presence_penalty?: number;
|
||||
tools?: any[];
|
||||
tool_choice?: string | object;
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: Message[];
|
||||
temperature?: number;
|
||||
max_tokens?: number;
|
||||
stream?: boolean;
|
||||
top_p?: number;
|
||||
frequency_penalty?: number;
|
||||
presence_penalty?: number;
|
||||
tools?: any[];
|
||||
tool_choice?: string | object;
|
||||
|
||||
// Internal control flags (not sent directly to API)
|
||||
enableTools?: boolean;
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
// Internal control flags (not sent directly to API)
|
||||
enableTools?: boolean;
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
}
|
||||
|
||||
/**
|
||||
* Anthropic-specific options, structured to match the Anthropic API
|
||||
*/
|
||||
export interface AnthropicOptions extends ProviderConfig {
|
||||
// Connection settings (not sent to API)
|
||||
apiKey: string;
|
||||
baseUrl: string;
|
||||
apiVersion?: string;
|
||||
betaVersion?: string;
|
||||
// Connection settings (not sent to API)
|
||||
apiKey: string;
|
||||
baseUrl: string;
|
||||
apiVersion?: string;
|
||||
betaVersion?: string;
|
||||
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: any[];
|
||||
system?: string;
|
||||
temperature?: number;
|
||||
max_tokens?: number;
|
||||
stream?: boolean;
|
||||
top_p?: number;
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: any[];
|
||||
system?: string;
|
||||
temperature?: number;
|
||||
max_tokens?: number;
|
||||
stream?: boolean;
|
||||
top_p?: number;
|
||||
|
||||
// Internal parameters (not sent directly to API)
|
||||
formattedMessages?: { messages: any[], system: string };
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
// Internal parameters (not sent directly to API)
|
||||
formattedMessages?: { messages: any[], system: string };
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
}
|
||||
|
||||
/**
|
||||
* Ollama-specific options, structured to match the Ollama API
|
||||
*/
|
||||
export interface OllamaOptions extends ProviderConfig {
|
||||
// Connection settings (not sent to API)
|
||||
baseUrl: string;
|
||||
// Connection settings (not sent to API)
|
||||
baseUrl: string;
|
||||
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: Message[];
|
||||
stream?: boolean;
|
||||
options?: {
|
||||
temperature?: number;
|
||||
num_ctx?: number;
|
||||
top_p?: number;
|
||||
top_k?: number;
|
||||
num_predict?: number; // equivalent to max_tokens
|
||||
response_format?: { type: string };
|
||||
};
|
||||
tools?: any[];
|
||||
// Direct API parameters as they appear in requests
|
||||
model: string;
|
||||
messages?: Message[];
|
||||
stream?: boolean;
|
||||
options?: {
|
||||
temperature?: number;
|
||||
num_ctx?: number;
|
||||
top_p?: number;
|
||||
top_k?: number;
|
||||
num_predict?: number; // equivalent to max_tokens
|
||||
response_format?: { type: string };
|
||||
};
|
||||
tools?: any[];
|
||||
|
||||
// Internal control flags (not sent directly to API)
|
||||
enableTools?: boolean;
|
||||
bypassFormatter?: boolean;
|
||||
preserveSystemPrompt?: boolean;
|
||||
expectsJsonResponse?: boolean;
|
||||
toolExecutionStatus?: any[];
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
// Internal control flags (not sent directly to API)
|
||||
enableTools?: boolean;
|
||||
bypassFormatter?: boolean;
|
||||
preserveSystemPrompt?: boolean;
|
||||
expectsJsonResponse?: boolean;
|
||||
toolExecutionStatus?: any[];
|
||||
// Streaming callback handler
|
||||
streamCallback?: (text: string, isDone: boolean, originalChunk?: any) => Promise<void> | void;
|
||||
}
|
||||
|
||||
/**
|
||||
* Create OpenAI options from generic options and config
|
||||
*/
|
||||
export function createOpenAIOptions(
|
||||
opts: ChatCompletionOptions = {},
|
||||
apiKey: string,
|
||||
baseUrl: string,
|
||||
defaultModel: string
|
||||
opts: ChatCompletionOptions = {},
|
||||
apiKey: string,
|
||||
baseUrl: string,
|
||||
defaultModel: string
|
||||
): OpenAIOptions {
|
||||
return {
|
||||
// Connection settings
|
||||
apiKey,
|
||||
baseUrl,
|
||||
return {
|
||||
// Connection settings
|
||||
apiKey,
|
||||
baseUrl,
|
||||
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
temperature: opts.temperature,
|
||||
max_tokens: opts.maxTokens,
|
||||
stream: opts.stream,
|
||||
top_p: opts.topP,
|
||||
frequency_penalty: opts.frequencyPenalty,
|
||||
presence_penalty: opts.presencePenalty,
|
||||
tools: opts.tools,
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
temperature: opts.temperature,
|
||||
max_tokens: opts.maxTokens,
|
||||
stream: opts.stream,
|
||||
top_p: opts.topP,
|
||||
frequency_penalty: opts.frequencyPenalty,
|
||||
presence_penalty: opts.presencePenalty,
|
||||
tools: opts.tools,
|
||||
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
enableTools: opts.enableTools,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
enableTools: opts.enableTools,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Create Anthropic options from generic options and config
|
||||
*/
|
||||
export function createAnthropicOptions(
|
||||
opts: ChatCompletionOptions = {},
|
||||
apiKey: string,
|
||||
baseUrl: string,
|
||||
defaultModel: string,
|
||||
apiVersion: string,
|
||||
betaVersion: string
|
||||
opts: ChatCompletionOptions = {},
|
||||
apiKey: string,
|
||||
baseUrl: string,
|
||||
defaultModel: string,
|
||||
apiVersion: string,
|
||||
betaVersion: string
|
||||
): AnthropicOptions {
|
||||
return {
|
||||
// Connection settings
|
||||
apiKey,
|
||||
baseUrl,
|
||||
apiVersion,
|
||||
betaVersion,
|
||||
return {
|
||||
// Connection settings
|
||||
apiKey,
|
||||
baseUrl,
|
||||
apiVersion,
|
||||
betaVersion,
|
||||
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
temperature: opts.temperature,
|
||||
max_tokens: opts.maxTokens,
|
||||
stream: opts.stream,
|
||||
top_p: opts.topP,
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
temperature: opts.temperature,
|
||||
max_tokens: opts.maxTokens,
|
||||
stream: opts.stream,
|
||||
top_p: opts.topP,
|
||||
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Create Ollama options from generic options and config
|
||||
*/
|
||||
export function createOllamaOptions(
|
||||
opts: ChatCompletionOptions = {},
|
||||
baseUrl: string,
|
||||
defaultModel: string,
|
||||
contextWindow: number
|
||||
opts: ChatCompletionOptions = {},
|
||||
baseUrl: string,
|
||||
defaultModel: string,
|
||||
contextWindow: number
|
||||
): OllamaOptions {
|
||||
return {
|
||||
// Connection settings
|
||||
baseUrl,
|
||||
return {
|
||||
// Connection settings
|
||||
baseUrl,
|
||||
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
stream: opts.stream,
|
||||
options: {
|
||||
temperature: opts.temperature,
|
||||
num_ctx: contextWindow,
|
||||
num_predict: opts.maxTokens,
|
||||
response_format: opts.expectsJsonResponse ? { type: "json_object" } : undefined
|
||||
},
|
||||
tools: opts.tools,
|
||||
// API parameters
|
||||
model: opts.model || defaultModel,
|
||||
stream: opts.stream,
|
||||
options: {
|
||||
temperature: opts.temperature,
|
||||
num_ctx: contextWindow,
|
||||
num_predict: opts.maxTokens,
|
||||
response_format: opts.expectsJsonResponse ? { type: "json_object" } : undefined
|
||||
},
|
||||
tools: opts.tools,
|
||||
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
enableTools: opts.enableTools,
|
||||
bypassFormatter: opts.bypassFormatter,
|
||||
preserveSystemPrompt: opts.preserveSystemPrompt,
|
||||
expectsJsonResponse: opts.expectsJsonResponse,
|
||||
toolExecutionStatus: opts.toolExecutionStatus,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
// Internal configuration
|
||||
systemPrompt: opts.systemPrompt,
|
||||
enableTools: opts.enableTools,
|
||||
bypassFormatter: opts.bypassFormatter,
|
||||
preserveSystemPrompt: opts.preserveSystemPrompt,
|
||||
expectsJsonResponse: opts.expectsJsonResponse,
|
||||
toolExecutionStatus: opts.toolExecutionStatus,
|
||||
// Pass through streaming callback
|
||||
streamCallback: opts.streamCallback,
|
||||
// Include provider metadata
|
||||
providerMetadata: opts.providerMetadata,
|
||||
};
|
||||
}
|
||||
|
||||
@ -153,7 +153,7 @@ export async function createEmbeddingProviderConfig(
|
||||
dateCreated, utcDateCreated, dateModified, utcDateModified)
|
||||
VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,
|
||||
[providerId, name, priority, JSON.stringify(config),
|
||||
now, utcNow, now, utcNow]
|
||||
now, utcNow, now, utcNow]
|
||||
);
|
||||
|
||||
return providerId;
|
||||
@ -460,8 +460,8 @@ export function getAnthropicOptions(
|
||||
supportsStreaming: true,
|
||||
// Anthropic models typically have large context windows
|
||||
contextWindow: modelName.includes('claude-3-opus') ? 200000 :
|
||||
modelName.includes('claude-3-sonnet') ? 180000 :
|
||||
modelName.includes('claude-3.5-sonnet') ? 200000 : 100000
|
||||
modelName.includes('claude-3-sonnet') ? 180000 :
|
||||
modelName.includes('claude-3.5-sonnet') ? 200000 : 100000
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user