progress
This commit is contained in:
overcuriousity
22
src/pages/api/ai/embeddings.status.ts
Normal file
22
src/pages/api/ai/embeddings.status.ts
Normal file
@@ -0,0 +1,22 @@
|
||||
// src/pages/api/ai/embeddings-status.ts
|
||||
import type { APIRoute } from 'astro';
|
||||
import { embeddingsService } from '../../../utils/embeddings.js';
|
||||
import { apiResponse, apiServerError } from '../../../utils/api.js';
|
||||
|
||||
export const prerender = false;
|
||||
|
||||
export const GET: APIRoute = async () => {
|
||||
try {
|
||||
const stats = embeddingsService.getStats();
|
||||
|
||||
return apiResponse.success({
|
||||
embeddings: stats,
|
||||
timestamp: new Date().toISOString(),
|
||||
status: stats.enabled && stats.initialized ? 'ready' :
|
||||
stats.enabled && !stats.initialized ? 'initializing' : 'disabled'
|
||||
});
|
||||
} catch (error) {
|
||||
console.error('Embeddings status error:', error);
|
||||
return apiServerError.internal('Failed to get embeddings status');
|
||||
}
|
||||
};
|
||||
@@ -14,7 +14,11 @@ function getEnv(key: string): string {
|
||||
return value;
|
||||
}
|
||||
|
||||
const AI_MODEL = getEnv('AI_MODEL');
|
||||
// Use the analyzer AI for smart prompting (smaller, faster model)
|
||||
const AI_ENDPOINT = getEnv('AI_ANALYZER_ENDPOINT');
|
||||
const AI_API_KEY = getEnv('AI_ANALYZER_API_KEY');
|
||||
const AI_MODEL = getEnv('AI_ANALYZER_MODEL');
|
||||
|
||||
const rateLimitStore = new Map<string, { count: number; resetTime: number }>();
|
||||
const RATE_LIMIT_WINDOW = 60 * 1000; // 1 minute
|
||||
const RATE_LIMIT_MAX = 5; // 5 enhancement requests per minute per user
|
||||
@@ -59,29 +63,38 @@ function cleanupExpiredRateLimits() {
|
||||
setInterval(cleanupExpiredRateLimits, 5 * 60 * 1000);
|
||||
|
||||
function createEnhancementPrompt(input: string): string {
|
||||
return `
|
||||
Du bist eine KI für digitale Forensik. Der Nutzer beschreibt ein forensisches Szenario. Analysiere die Eingabe.
|
||||
return `Du bist eine KI für digitale Forensik-Anfragen. Der Nutzer beschreibt ein forensisches Szenario oder Problem. Analysiere die Eingabe auf Vollständigkeit und Klarheit.
|
||||
|
||||
Wenn die Beschreibung unvollständig oder vage ist, stelle bis zu drei präzise Rückfragen im JSON-Array-Format, um wichtige Details zu klären (z. B. Vorfalltyp, System, Ziel, Datenquellen, Zeit, Beteiligte, rechtlicher Rahmen).
|
||||
ANALYSIERE DIESE KATEGORIEN:
|
||||
1. **Vorfalltyp**: Was ist passiert? (Malware, Datendiebstahl, Compliance-Verstoß, etc.)
|
||||
2. **Betroffene Systeme**: Welche Technologien/Plattformen? (Windows, Linux, Mobile, Cloud, etc.)
|
||||
3. **Verfügbare Datenquellen**: Was kann untersucht werden? (Logs, Images, Memory Dumps, etc.)
|
||||
4. **Untersuchungsziel**: Was soll erreicht werden? (IOCs finden, Timeline erstellen, etc.)
|
||||
5. **Zeitrahmen & Dringlichkeit**: Wann ist etwas passiert? Wie dringend?
|
||||
6. **Ressourcen & Constraints**: Budget, Skills, Tools, rechtliche Aspekte
|
||||
7. **Beweisziele**: Dokumentation, Gerichtsverfahren, interne Aufklärung?
|
||||
|
||||
Wenn die Eingabe bereits klar, spezifisch und vollständig ist, gib stattdessen nur eine leere Liste [] zurück.
|
||||
WENN die Beschreibung vollständig und spezifisch ist: Gib eine leere Liste [] zurück.
|
||||
|
||||
Antwortformat strikt:
|
||||
WENN wichtige Details fehlen: Formuliere 2-3 präzise Fragen, die die kritischsten Lücken schließen. Fokussiere auf Details, die die Tool-/Methoden-Auswahl stark beeinflussen.
|
||||
|
||||
\`\`\`json
|
||||
FRAGE-QUALITÄT:
|
||||
- Spezifisch, nicht allgemein (❌ "Mehr Details?" ✅ "Welche Betriebssysteme sind betroffen?")
|
||||
- Handlungsrelevant (❌ "Wann passierte das?" ✅ "Haben Sie Logs aus der Vorfallzeit verfügbar?")
|
||||
- Priorisiert nach Wichtigkeit für die forensische Analyse
|
||||
|
||||
ANTWORTFORMAT (NUR JSON):
|
||||
[
|
||||
"Frage 1?",
|
||||
"Frage 2?",
|
||||
"Frage 3?"
|
||||
"Spezifische Frage 1?",
|
||||
"Spezifische Frage 2?",
|
||||
"Spezifische Frage 3?"
|
||||
]
|
||||
\`\`\`
|
||||
|
||||
Nutzer-Eingabe:
|
||||
NUTZER-EINGABE:
|
||||
${input}
|
||||
`.trim();
|
||||
}
|
||||
|
||||
|
||||
export const POST: APIRoute = async ({ request }) => {
|
||||
try {
|
||||
const authResult = await withAPIAuth(request, 'ai');
|
||||
@@ -98,12 +111,12 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
const body = await request.json();
|
||||
const { input } = body;
|
||||
|
||||
if (!input || typeof input !== 'string' || input.length < 20) {
|
||||
return apiError.badRequest('Input too short for enhancement');
|
||||
if (!input || typeof input !== 'string' || input.length < 40) {
|
||||
return apiError.badRequest('Input too short for enhancement (minimum 40 characters)');
|
||||
}
|
||||
|
||||
const sanitizedInput = sanitizeInput(input);
|
||||
if (sanitizedInput.length < 20) {
|
||||
if (sanitizedInput.length < 40) {
|
||||
return apiError.badRequest('Input too short after sanitization');
|
||||
}
|
||||
|
||||
@@ -111,11 +124,11 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
const taskId = `enhance_${userId}_${Date.now()}_${Math.random().toString(36).substr(2, 4)}`;
|
||||
|
||||
const aiResponse = await enqueueApiCall(() =>
|
||||
fetch(process.env.AI_API_ENDPOINT + '/v1/chat/completions', {
|
||||
fetch(`${AI_ENDPOINT}/v1/chat/completions`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${process.env.AI_API_KEY}`
|
||||
'Authorization': `Bearer ${AI_API_KEY}`
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: AI_MODEL,
|
||||
@@ -125,7 +138,7 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
content: systemPrompt
|
||||
}
|
||||
],
|
||||
max_tokens: 200,
|
||||
max_tokens: 300,
|
||||
temperature: 0.7
|
||||
})
|
||||
}), taskId);
|
||||
@@ -144,28 +157,32 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
|
||||
let questions;
|
||||
try {
|
||||
const cleanedContent = aiContent
|
||||
const cleanedContent = aiContent
|
||||
.replace(/^```json\s*/i, '')
|
||||
.replace(/\s*```\s*$/, '')
|
||||
.trim();
|
||||
questions = JSON.parse(cleanedContent);
|
||||
questions = JSON.parse(cleanedContent);
|
||||
|
||||
if (!Array.isArray(questions) || questions.length === 0) {
|
||||
throw new Error('Invalid questions format');
|
||||
if (!Array.isArray(questions)) {
|
||||
throw new Error('Response is not an array');
|
||||
}
|
||||
|
||||
// Validate and clean questions
|
||||
// Enhanced validation and cleaning
|
||||
questions = questions
|
||||
.filter(q => typeof q === 'string' && q.length > 5 && q.length < 120)
|
||||
.slice(0, 3);
|
||||
.filter(q => typeof q === 'string' && q.length > 10 && q.length < 150) // More reasonable length limits
|
||||
.filter(q => q.includes('?')) // Must be a question
|
||||
.map(q => q.trim())
|
||||
.slice(0, 3); // Max 3 questions
|
||||
|
||||
// If no valid questions, return empty array (means input is complete)
|
||||
if (questions.length === 0) {
|
||||
throw new Error('No valid questions found');
|
||||
questions = [];
|
||||
}
|
||||
|
||||
} catch (error) {
|
||||
console.error('Failed to parse enhancement response:', aiContent);
|
||||
return apiServerError.unavailable('Invalid enhancement response format');
|
||||
// If parsing fails, assume input is complete enough
|
||||
questions = [];
|
||||
}
|
||||
|
||||
console.log(`[AI Enhancement] User: ${userId}, Questions: ${questions.length}, Input length: ${sanitizedInput.length}`);
|
||||
@@ -173,7 +190,8 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
return new Response(JSON.stringify({
|
||||
success: true,
|
||||
questions,
|
||||
taskId
|
||||
taskId,
|
||||
inputComplete: questions.length === 0 // Flag to indicate if input seems complete
|
||||
}), {
|
||||
status: 200,
|
||||
headers: { 'Content-Type': 'application/json' }
|
||||
|
||||
@@ -1,21 +1,12 @@
|
||||
// src/pages/api/ai/query.ts
|
||||
import type { APIRoute } from 'astro';
|
||||
import { withAPIAuth } from '../../../utils/auth.js';
|
||||
import { getCompressedToolsDataForAI } from '../../../utils/dataService.js';
|
||||
import { apiError, apiServerError, createAuthErrorResponse } from '../../../utils/api.js';
|
||||
import { enqueueApiCall } from '../../../utils/rateLimitedQueue.js';
|
||||
import { aiPipeline } from '../../../utils/aiPipeline.js';
|
||||
|
||||
export const prerender = false;
|
||||
|
||||
function getEnv(key: string): string {
|
||||
const value = process.env[key];
|
||||
if (!value) {
|
||||
throw new Error(`Missing environment variable: ${key}`);
|
||||
}
|
||||
return value;
|
||||
}
|
||||
|
||||
const AI_MODEL = getEnv('AI_MODEL');
|
||||
const rateLimitStore = new Map<string, { count: number; resetTime: number }>();
|
||||
const RATE_LIMIT_WINDOW = 60 * 1000;
|
||||
const RATE_LIMIT_MAX = 10;
|
||||
@@ -33,13 +24,6 @@ function sanitizeInput(input: string): string {
|
||||
return sanitized;
|
||||
}
|
||||
|
||||
function stripMarkdownJson(content: string): string {
|
||||
return content
|
||||
.replace(/^```json\s*/i, '')
|
||||
.replace(/\s*```\s*$/, '')
|
||||
.trim();
|
||||
}
|
||||
|
||||
function checkRateLimit(userId: string): boolean {
|
||||
const now = Date.now();
|
||||
const userLimit = rateLimitStore.get(userId);
|
||||
@@ -68,209 +52,6 @@ function cleanupExpiredRateLimits() {
|
||||
|
||||
setInterval(cleanupExpiredRateLimits, 5 * 60 * 1000);
|
||||
|
||||
async function loadToolsDatabase() {
|
||||
try {
|
||||
return await getCompressedToolsDataForAI();
|
||||
} catch (error) {
|
||||
console.error('Failed to load tools database:', error);
|
||||
throw new Error('Database unavailable');
|
||||
}
|
||||
}
|
||||
|
||||
function createWorkflowSystemPrompt(toolsData: any): string {
|
||||
const toolsList = toolsData.tools.map((tool: any) => ({
|
||||
name: tool.name,
|
||||
description: tool.description,
|
||||
domains: tool.domains,
|
||||
phases: tool.phases,
|
||||
domainAgnostic: tool['domain-agnostic-software'],
|
||||
platforms: tool.platforms,
|
||||
skillLevel: tool.skillLevel,
|
||||
license: tool.license,
|
||||
tags: tool.tags,
|
||||
related_concepts: tool.related_concepts || []
|
||||
}));
|
||||
|
||||
const conceptsList = toolsData.concepts.map((concept: any) => ({
|
||||
name: concept.name,
|
||||
description: concept.description,
|
||||
domains: concept.domains,
|
||||
phases: concept.phases,
|
||||
skillLevel: concept.skillLevel,
|
||||
tags: concept.tags
|
||||
}));
|
||||
|
||||
const regularPhases = toolsData.phases || [];
|
||||
|
||||
const domainAgnosticSoftware = toolsData['domain-agnostic-software'] || [];
|
||||
|
||||
const allPhaseItems = [
|
||||
...regularPhases,
|
||||
...domainAgnosticSoftware
|
||||
];
|
||||
|
||||
const phasesDescription = allPhaseItems.map((phase: any) =>
|
||||
`- ${phase.id}: ${phase.name}`
|
||||
).join('\n');
|
||||
|
||||
const domainsDescription = toolsData.domains.map((domain: any) =>
|
||||
`- ${domain.id}: ${domain.name}`
|
||||
).join('\n');
|
||||
|
||||
const phaseDescriptions = regularPhases.map((phase: any) =>
|
||||
`- ${phase.name}: ${phase.description || 'Tools/Methods for this phase'}`
|
||||
).join('\n');
|
||||
|
||||
const domainAgnosticDescriptions = domainAgnosticSoftware.map((section: any) =>
|
||||
`- ${section.name}: ${section.description || 'Cross-cutting software and platforms'}`
|
||||
).join('\n');
|
||||
|
||||
const validPhases = [
|
||||
...regularPhases.map((p: any) => p.id),
|
||||
...domainAgnosticSoftware.map((s: any) => s.id)
|
||||
].join('|');
|
||||
|
||||
return `Du bist ein DFIR (Digital Forensics and Incident Response) Experte, der Ermittlern bei der Auswahl von Software und Methoden hilft.
|
||||
|
||||
VERFÜGBARE TOOLS/METHODEN:
|
||||
${JSON.stringify(toolsList, null, 2)}
|
||||
|
||||
VERFÜGBARE HINTERGRUNDWISSEN-KONZEPTE:
|
||||
${JSON.stringify(conceptsList, null, 2)}
|
||||
|
||||
UNTERSUCHUNGSPHASEN (NIST Framework):
|
||||
${phasesDescription}
|
||||
|
||||
FORENSISCHE DOMÄNEN:
|
||||
${domainsDescription}
|
||||
|
||||
WICHTIGE REGELN:
|
||||
1. Pro Phase 2-3 Tools/Methoden empfehlen (immer mindestens 2 wenn verfügbar)
|
||||
2. Tools/Methoden können in MEHREREN Phasen empfohlen werden wenn sinnvoll - versuche ein Tool/Methode für jede Phase zu empfehlen, selbst wenn die Priorität "low" ist.
|
||||
3. Für Reporting-Phase: Visualisierungs- und Dokumentationssoftware einschließen
|
||||
4. Gib stets dem spezieller für den Fall geeigneten Werkzeug den Vorzug.
|
||||
5. Deutsche Antworten für deutsche Anfragen, English for English queries
|
||||
6. Methoden haben, sofern für das SZENARIO passend, IMMER Vorrang vor Software.
|
||||
7. Bevorzuge alles, was nicht proprietär ist (license != "Proprietary"), aber erkenne an, wenn proprietäre Software besser geeignet ist.
|
||||
8. WICHTIG: Erwähne relevante Hintergrundwissen-Konzepte wenn Tools verwendet werden, die related_concepts haben
|
||||
9. Konzepte sind NICHT Tools - empfehle sie nicht als actionable Schritte, sondern als Wissensbasis
|
||||
|
||||
ENHANCED CONTEXTUAL ANALYSIS:
|
||||
10. Analysiere das Szenario detailliert und identifiziere Schlüsselelemente, Bedrohungen und forensische Herausforderungen
|
||||
11. Entwickle einen strategischen Untersuchungsansatz basierend auf dem spezifischen Szenario
|
||||
12. Identifiziere zeitkritische oder besonders wichtige Faktoren für diesen Fall
|
||||
|
||||
SOFTWARE/METHODEN-AUSWAHL NACH PHASE:
|
||||
${phaseDescriptions}
|
||||
|
||||
DOMÄNENAGNOSTISCHE SOFTWARE/METHODEN:
|
||||
${domainAgnosticDescriptions}
|
||||
|
||||
ANTWORT-FORMAT (strict JSON):
|
||||
{
|
||||
"scenario_analysis": "Detaillierte Analyse des Szenarios: Erkannte Schlüsselelemente, Art des Vorfalls, betroffene Systeme, potentielle Bedrohungen und forensische Herausforderungen",
|
||||
"investigation_approach": "Strategischer Untersuchungsansatz für dieses spezifische Szenario: Prioritäten, Reihenfolge der Phasen, besondere Überlegungen",
|
||||
"critical_considerations": "Zeitkritische Faktoren, wichtige Sicherheitsaspekte oder besondere Vorsichtsmaßnahmen für diesen Fall",
|
||||
"recommended_tools": [
|
||||
{
|
||||
"name": "EXAKTER Name aus der Tools-Database",
|
||||
"priority": "high|medium|low",
|
||||
"phase": "${validPhases}",
|
||||
"justification": "Warum diese Methode für diese Phase und dieses spezifische Szenario geeignet ist - mit Bezug zu den erkannten Schlüsselelementen"
|
||||
}
|
||||
],
|
||||
"workflow_suggestion": "Vorgeschlagener Untersuchungsablauf mit konkreten Schritten für dieses Szenario",
|
||||
"background_knowledge": [
|
||||
{
|
||||
"concept_name": "EXAKTER Name aus der Konzepte-Database",
|
||||
"relevance": "Warum dieses Konzept für das Szenario relevant ist, und bei welchen der empfohlenen Methoden/Tools."
|
||||
}
|
||||
],
|
||||
"additional_notes": "Wichtige Überlegungen und Hinweise"
|
||||
}
|
||||
|
||||
Antworte NUR mit validen JSON. Keine zusätzlichen Erklärungen außerhalb des JSON.`;
|
||||
}
|
||||
|
||||
function createToolSystemPrompt(toolsData: any): string {
|
||||
const toolsList = toolsData.tools.map((tool: any) => ({
|
||||
name: tool.name,
|
||||
description: tool.description,
|
||||
domains: tool.domains,
|
||||
phases: tool.phases,
|
||||
platforms: tool.platforms,
|
||||
skillLevel: tool.skillLevel,
|
||||
license: tool.license,
|
||||
tags: tool.tags,
|
||||
url: tool.url,
|
||||
projectUrl: tool.projectUrl,
|
||||
related_concepts: tool.related_concepts || []
|
||||
}));
|
||||
|
||||
const conceptsList = toolsData.concepts.map((concept: any) => ({
|
||||
name: concept.name,
|
||||
description: concept.description,
|
||||
domains: concept.domains,
|
||||
phases: concept.phases,
|
||||
skillLevel: concept.skillLevel,
|
||||
tags: concept.tags
|
||||
}));
|
||||
|
||||
return `Du bist ein DFIR (Digital Forensics and Incident Response) Experte, der bei der Auswahl spezifischer Software/Methoden für konkrete Probleme hilft.
|
||||
|
||||
VERFÜGBARE TOOLS/METHODEN:
|
||||
${JSON.stringify(toolsList, null, 2)}
|
||||
|
||||
VERFÜGBARE HINTERGRUNDWISSEN-KONZEPTE:
|
||||
${JSON.stringify(conceptsList, null, 2)}
|
||||
|
||||
WICHTIGE REGELN:
|
||||
1. Analysiere das spezifische Problem/die Anforderung sorgfältig
|
||||
2. Empfehle 1-3 Methoden/Tools, sortiert nach Eignung (beste Empfehlung zuerst)
|
||||
3. Gib detaillierte Erklärungen, WARUM und WIE jede Methode/Tool das Problem löst
|
||||
4. Berücksichtige praktische Aspekte: Skill Level, Plattformen, Verfügbarkeit
|
||||
5. Deutsche Antworten für deutsche Anfragen, English for English queries
|
||||
6. Gib konkrete Anwendungshinweise, nicht nur allgemeine Beschreibungen - Methoden haben, sofern für das SZENARIO passend, IMMER Vorrang vor Software.
|
||||
7. Erwähne sowohl Stärken als auch Schwächen/Limitationen
|
||||
8. Schlage alternative Ansätze vor, wenn sinnvoll
|
||||
9. Gib grundsätzliche Hinweise, WIE die Methode/Tool konkret eingesetzt wird
|
||||
10. WICHTIG: Erwähne relevante Hintergrundwissen-Konzepte wenn Tools verwendet werden, die related_concepts haben
|
||||
11. Konzepte sind NICHT Tools - empfehle sie nicht als actionable Schritte, sondern als Wissensbasis
|
||||
|
||||
ENHANCED CONTEXTUAL ANALYSIS:
|
||||
12. Analysiere das Problem detailliert und identifiziere technische Anforderungen, Herausforderungen und Erfolgsfaktoren
|
||||
13. Entwickle einen strategischen Lösungsansatz basierend auf dem spezifischen Problem
|
||||
14. Identifiziere wichtige Voraussetzungen oder Warnungen für die Anwendung
|
||||
|
||||
ANTWORT-FORMAT (strict JSON):
|
||||
{
|
||||
"problem_analysis": "Detaillierte Analyse des Problems: Erkannte technische Anforderungen, Herausforderungen, benötigte Fähigkeiten und Erfolgsfaktoren",
|
||||
"investigation_approach": "Strategischer Lösungsansatz für dieses spezifische Problem: Herangehensweise, Prioritäten, optimale Anwendungsreihenfolge",
|
||||
"critical_considerations": "Wichtige Voraussetzungen, potentielle Fallstricke oder Warnungen für die Anwendung der empfohlenen Lösungen",
|
||||
"recommended_tools": [
|
||||
{
|
||||
"name": "EXAKTER Name aus der Tools-Database",
|
||||
"rank": 1,
|
||||
"suitability_score": "high|medium|low",
|
||||
"detailed_explanation": "Detaillierte Erklärung, warum dieses Tool/diese Methode das spezifische Problem löst - mit Bezug zu den erkannten Anforderungen",
|
||||
"implementation_approach": "Konkrete Schritte/Ansatz zur Anwendung für dieses spezifische Problem",
|
||||
"pros": ["Spezifische Vorteile für diesen Anwendungsfall", "Weitere Vorteile"],
|
||||
"cons": ["Potentielle Nachteile oder Limitationen", "Weitere Einschränkungen"],
|
||||
"alternatives": "Alternative Ansätze oder ergänzende Tools/Methoden, falls relevant"
|
||||
}
|
||||
],
|
||||
"background_knowledge": [
|
||||
{
|
||||
"concept_name": "EXAKTER Name aus der Konzepte-Database",
|
||||
"relevance": "Warum dieses Konzept für die empfohlenen Tools/das Problem relevant ist, und für welche der empfohlenen Methoden/Tools."
|
||||
}
|
||||
],
|
||||
"additional_considerations": "Wichtige Überlegungen, Voraussetzungen oder Warnungen"
|
||||
}
|
||||
|
||||
Antworte NUR mit validen JSON. Keine zusätzlichen Erklärungen außerhalb des JSON.`;
|
||||
}
|
||||
|
||||
export const POST: APIRoute = async ({ request }) => {
|
||||
try {
|
||||
const authResult = await withAPIAuth(request, 'ai');
|
||||
@@ -287,7 +68,6 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
const body = await request.json();
|
||||
const { query, mode = 'workflow', taskId: clientTaskId } = body;
|
||||
|
||||
// ADD THIS DEBUG LOGGING
|
||||
console.log(`[AI API] Received request - TaskId: ${clientTaskId}, Mode: ${mode}, Query length: ${query?.length || 0}`);
|
||||
|
||||
if (!query || typeof query !== 'string') {
|
||||
@@ -306,128 +86,31 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
return apiError.badRequest('Invalid input detected');
|
||||
}
|
||||
|
||||
const toolsData = await loadToolsDatabase();
|
||||
|
||||
const systemPrompt = mode === 'workflow'
|
||||
? createWorkflowSystemPrompt(toolsData)
|
||||
: createToolSystemPrompt(toolsData);
|
||||
|
||||
const taskId = clientTaskId || `ai_${userId}_${Date.now()}_${Math.random().toString(36).substr(2, 6)}`;
|
||||
|
||||
console.log(`[AI API] About to enqueue task ${taskId}`);
|
||||
|
||||
|
||||
const aiResponse = await enqueueApiCall(() =>
|
||||
fetch(process.env.AI_API_ENDPOINT + '/v1/chat/completions', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${process.env.AI_API_KEY}`
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: AI_MODEL,
|
||||
messages: [
|
||||
{
|
||||
role: 'system',
|
||||
content: systemPrompt
|
||||
},
|
||||
{
|
||||
role: 'user',
|
||||
content: sanitizedQuery
|
||||
}
|
||||
],
|
||||
max_tokens: 3500,
|
||||
temperature: 0.3
|
||||
})
|
||||
})
|
||||
// Use the new AI pipeline instead of direct API calls
|
||||
const result = await enqueueApiCall(() =>
|
||||
aiPipeline.processQuery(sanitizedQuery, mode)
|
||||
, taskId);
|
||||
|
||||
if (!aiResponse.ok) {
|
||||
console.error('AI API error:', await aiResponse.text());
|
||||
return apiServerError.unavailable('AI service unavailable');
|
||||
if (!result || !result.recommendation) {
|
||||
return apiServerError.unavailable('No response from AI pipeline');
|
||||
}
|
||||
|
||||
const aiData = await aiResponse.json();
|
||||
const aiContent = aiData.choices?.[0]?.message?.content;
|
||||
|
||||
if (!aiContent) {
|
||||
return apiServerError.unavailable('No response from AI');
|
||||
}
|
||||
|
||||
let recommendation;
|
||||
try {
|
||||
const cleanedContent = stripMarkdownJson(aiContent);
|
||||
recommendation = JSON.parse(cleanedContent);
|
||||
} catch (error) {
|
||||
console.error('Failed to parse AI response:', aiContent);
|
||||
return apiServerError.unavailable('Invalid AI response format');
|
||||
}
|
||||
|
||||
const validToolNames = new Set(toolsData.tools.map((t: any) => t.name));
|
||||
const validConceptNames = new Set(toolsData.concepts.map((c: any) => c.name));
|
||||
|
||||
let validatedRecommendation;
|
||||
|
||||
if (mode === 'workflow') {
|
||||
validatedRecommendation = {
|
||||
...recommendation,
|
||||
// Ensure all new fields are included with fallbacks
|
||||
scenario_analysis: recommendation.scenario_analysis || recommendation.problem_analysis || '',
|
||||
investigation_approach: recommendation.investigation_approach || '',
|
||||
critical_considerations: recommendation.critical_considerations || '',
|
||||
recommended_tools: recommendation.recommended_tools?.filter((tool: any) => {
|
||||
if (!validToolNames.has(tool.name)) {
|
||||
console.warn(`AI recommended unknown tool: ${tool.name}`);
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}) || [],
|
||||
background_knowledge: recommendation.background_knowledge?.filter((concept: any) => {
|
||||
if (!validConceptNames.has(concept.concept_name)) {
|
||||
console.warn(`AI referenced unknown concept: ${concept.concept_name}`);
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}) || []
|
||||
};
|
||||
} else {
|
||||
validatedRecommendation = {
|
||||
...recommendation,
|
||||
// Ensure all new fields are included with fallbacks
|
||||
problem_analysis: recommendation.problem_analysis || recommendation.scenario_analysis || '',
|
||||
investigation_approach: recommendation.investigation_approach || '',
|
||||
critical_considerations: recommendation.critical_considerations || '',
|
||||
recommended_tools: recommendation.recommended_tools?.filter((tool: any) => {
|
||||
if (!validToolNames.has(tool.name)) {
|
||||
console.warn(`AI recommended unknown tool: ${tool.name}`);
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}).map((tool: any, index: number) => ({
|
||||
...tool,
|
||||
rank: tool.rank || (index + 1),
|
||||
suitability_score: tool.suitability_score || 'medium',
|
||||
pros: Array.isArray(tool.pros) ? tool.pros : [],
|
||||
cons: Array.isArray(tool.cons) ? tool.cons : []
|
||||
})) || [],
|
||||
background_knowledge: recommendation.background_knowledge?.filter((concept: any) => {
|
||||
if (!validConceptNames.has(concept.concept_name)) {
|
||||
console.warn(`AI referenced unknown concept: ${concept.concept_name}`);
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}) || []
|
||||
};
|
||||
}
|
||||
|
||||
console.log(`[AI Query] Mode: ${mode}, User: ${userId}, Query length: ${sanitizedQuery.length}, Tools: ${validatedRecommendation.recommended_tools.length}, Concepts: ${validatedRecommendation.background_knowledge?.length || 0}`);
|
||||
// Add processing statistics to the response for debugging/monitoring
|
||||
console.log(`[AI Query] Mode: ${mode}, User: ${userId}, Query length: ${sanitizedQuery.length}`);
|
||||
console.log(`[AI Query] Processing stats:`, result.processingStats);
|
||||
console.log(`[AI Query] Tools: ${result.recommendation.recommended_tools?.length || 0}, Concepts: ${result.recommendation.background_knowledge?.length || 0}`);
|
||||
|
||||
return new Response(JSON.stringify({
|
||||
success: true,
|
||||
mode,
|
||||
taskId,
|
||||
recommendation: validatedRecommendation,
|
||||
query: sanitizedQuery
|
||||
recommendation: result.recommendation,
|
||||
query: sanitizedQuery,
|
||||
processingStats: result.processingStats // Include stats for monitoring
|
||||
}), {
|
||||
status: 200,
|
||||
headers: { 'Content-Type': 'application/json' }
|
||||
@@ -435,6 +118,16 @@ export const POST: APIRoute = async ({ request }) => {
|
||||
|
||||
} catch (error) {
|
||||
console.error('AI query error:', error);
|
||||
return apiServerError.internal('Internal server error');
|
||||
|
||||
// Provide more specific error messages based on error type
|
||||
if (error.message.includes('embeddings')) {
|
||||
return apiServerError.unavailable('Embeddings service error - falling back to basic processing');
|
||||
} else if (error.message.includes('selector')) {
|
||||
return apiServerError.unavailable('AI selector service error');
|
||||
} else if (error.message.includes('analyzer')) {
|
||||
return apiServerError.unavailable('AI analyzer service error');
|
||||
} else {
|
||||
return apiServerError.internal('Internal server error');
|
||||
}
|
||||
}
|
||||
};
|
||||
Reference in New Issue
Block a user