@kernel.chat/kbot 3.93.0 → 3.94.0

package/dist/tools/research-engine.d.ts

@@ -0,0 +1,58 @@
+/**
+ * research-engine.ts — Autonomous Web Research Pipeline
+ *
+ * kbot discovers new techniques, knowledge, and information from the internet
+ * using its own built-in browser. No Chrome, no Playwright — pure HTML.
+ *
+ * Architecture:
+ *   1. Research Queue — tasks from evolution, brain, narrative, user, or autonomous
+ *   2. Web Search — kbot_search via DuckDuckGo (built-in browser)
+ *   3. Page Reading — kbot_browse for full content extraction
+ *   4. Summarization — local Ollama (kernel:latest) for zero-cost summaries
+ *   5. Result Storage — ~/.kbot/research-results.json (persists across sessions)
+ *   6. Engine Integration — results tagged by applicable engine for routing
+ *
+ * Tools: research_queue, research_results, research_now
+ *
+ * Tick-based: tickResearch() called from the frame loop processes one task
+ * per interval (default 3600 frames = ~10 min at 6 fps).
+ */
+export interface ResearchEngine {
+    queue: ResearchTask[];
+    completed: ResearchResult[];
+    activeTask: ResearchTask | null;
+    lastResearchFrame: number;
+    researchInterval: number;
+    topicsOfInterest: string[];
+}
+export interface ResearchTask {
+    id: string;
+    query: string;
+    purpose: string;
+    source: 'evolution' | 'brain' | 'narrative' | 'user' | 'autonomous';
+    status: 'queued' | 'searching' | 'reading' | 'summarizing' | 'complete' | 'failed';
+    startedAt: number;
+}
+export interface ResearchResult {
+    taskId: string;
+    query: string;
+    summary: string;
+    sources: string[];
+    keyFindings: string[];
+    applicableTo: string[];
+    timestamp: number;
+}
+export interface ResearchAction {
+    type: 'start_research' | 'search_complete' | 'read_complete' | 'summarize_complete' | 'failed';
+    task: ResearchTask;
+    result?: ResearchResult;
+    speech?: string;
+}
+export declare function autonomousResearchTopics(): string[];
+export declare function initResearchEngine(): ResearchEngine;
+export declare function getResearchEngine(): ResearchEngine;
+export declare function queueResearch(engine: ResearchEngine, query: string, purpose: string, source: ResearchTask['source']): ResearchTask;
+export declare function getResearchForEngine(engine: ResearchEngine, engineName: string): ResearchResult[];
+export declare function tickResearch(engine: ResearchEngine, frame: number): Promise<ResearchAction | null>;
+export declare function registerResearchEngineTools(): void;
+//# sourceMappingURL=research-engine.d.ts.map

package/dist/tools/research-engine.js

@@ -0,0 +1,550 @@
+/**
+ * research-engine.ts — Autonomous Web Research Pipeline
+ *
+ * kbot discovers new techniques, knowledge, and information from the internet
+ * using its own built-in browser. No Chrome, no Playwright — pure HTML.
+ *
+ * Architecture:
+ *   1. Research Queue — tasks from evolution, brain, narrative, user, or autonomous
+ *   2. Web Search — kbot_search via DuckDuckGo (built-in browser)
+ *   3. Page Reading — kbot_browse for full content extraction
+ *   4. Summarization — local Ollama (kernel:latest) for zero-cost summaries
+ *   5. Result Storage — ~/.kbot/research-results.json (persists across sessions)
+ *   6. Engine Integration — results tagged by applicable engine for routing
+ *
+ * Tools: research_queue, research_results, research_now
+ *
+ * Tick-based: tickResearch() called from the frame loop processes one task
+ * per interval (default 3600 frames = ~10 min at 6 fps).
+ */
+import { registerTool } from './index.js';
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+import { readFileSync, writeFileSync, existsSync, mkdirSync } from 'node:fs';
+import { randomUUID } from 'node:crypto';
+// ─── Constants ──────────────────────────────────────────────────
+const KBOT_DIR = join(homedir(), '.kbot');
+const RESULTS_FILE = join(KBOT_DIR, 'research-results.json');
+const OLLAMA_URL = 'http://localhost:11434';
+const OLLAMA_MODEL = 'kernel:latest';
+const OLLAMA_TIMEOUT = 90_000;
+const MAX_COMPLETED = 200; // keep last N results in memory
+const MAX_PERSISTED = 500; // keep last N results on disk
+const MAX_LINKS_TO_READ = 3; // read top N links per search
+const DEFAULT_RESEARCH_INTERVAL = 3600; // frames between research tasks (~10 min at 6fps)
+// ─── Module State ───────────────────────────────────────────────
+let engineState = null;
+// ─── Persistence ────────────────────────────────────────────────
+function loadResults() {
+    try {
+        if (existsSync(RESULTS_FILE)) {
+            const raw = readFileSync(RESULTS_FILE, 'utf-8');
+            const parsed = JSON.parse(raw);
+            if (Array.isArray(parsed))
+                return parsed.slice(-MAX_PERSISTED);
+        }
+    }
+    catch { /* corrupt file — start fresh */ }
+    return [];
+}
+function saveResults(results) {
+    try {
+        if (!existsSync(KBOT_DIR))
+            mkdirSync(KBOT_DIR, { recursive: true });
+        const trimmed = results.slice(-MAX_PERSISTED);
+        writeFileSync(RESULTS_FILE, JSON.stringify(trimmed, null, 2));
+    }
+    catch { /* disk error — non-fatal */ }
+}
+// ─── Ollama Integration ─────────────────────────────────────────
+async function isOllamaAvailable() {
+    try {
+        const res = await fetch(`${OLLAMA_URL}/api/tags`, { signal: AbortSignal.timeout(3000) });
+        return res.ok;
+    }
+    catch {
+        return false;
+    }
+}
+async function ollamaGenerate(prompt, model = OLLAMA_MODEL) {
+    try {
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), OLLAMA_TIMEOUT);
+        const res = await fetch(`${OLLAMA_URL}/api/generate`, {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({
+                model,
+                prompt,
+                stream: false,
+                options: { temperature: 0.3, num_predict: 2048 },
+            }),
+            signal: controller.signal,
+        });
+        clearTimeout(timer);
+        if (!res.ok)
+            return null;
+        const data = await res.json();
+        return data.response?.trim() || null;
+    }
+    catch {
+        return null;
+    }
+}
+// ─── Link Extraction ────────────────────────────────────────────
+/**
+ * Extract URLs from kbot_search / kbot_browse output.
+ * The browser format is: `[N] link text` with URLs on the page,
+ * plus inline `URL: https://...` lines in the content.
+ */
+function extractLinks(searchOutput) {
+    const urls = [];
+    const seen = new Set();
+    // Match URLs in the output (DuckDuckGo results include them in content)
+    const urlRegex = /https?:\/\/[^\s<>"')\]]+/g;
+    let match;
+    while ((match = urlRegex.exec(searchOutput)) !== null) {
+        const url = match[0].replace(/[.,;:!?]+$/, ''); // trim trailing punctuation
+        if (!seen.has(url) && !url.includes('duckduckgo.com') && !url.includes('127.0.0.1')) {
+            seen.add(url);
+            urls.push(url);
+        }
+    }
+    return urls;
+}
+// ─── Key Findings Extraction ────────────────────────────────────
+function extractKeyFindings(summary) {
+    const findings = [];
+    const lines = summary.split('\n');
+    for (const line of lines) {
+        const trimmed = line.trim();
+        // Bullet points, numbered items, or lines starting with key phrases
+        if (/^[-*]\s+/.test(trimmed) || /^\d+[.)]\s+/.test(trimmed)) {
+            const clean = trimmed.replace(/^[-*\d.)\s]+/, '').trim();
+            if (clean.length > 15 && clean.length < 300) {
+                findings.push(clean);
+            }
+        }
+    }
+    // If no structured findings, split summary into sentence-level findings
+    if (findings.length === 0) {
+        const sentences = summary.split(/[.!?]+/).filter(s => s.trim().length > 20);
+        for (const s of sentences.slice(0, 5)) {
+            findings.push(s.trim());
+        }
+    }
+    return findings.slice(0, 10);
+}
+// ─── Engine Applicability Detection ─────────────────────────────
+const ENGINE_KEYWORDS = {
+    evolution: ['rendering', 'pixel', 'canvas', 'shader', 'graphics', 'animation', 'parallax', 'palette', 'lighting', 'visual', 'sprite', 'tile', 'effect'],
+    narrative: ['story', 'narrative', 'dialogue', 'character', 'quest', 'lore', 'writing', 'plot', 'world-building', 'fiction'],
+    audio: ['audio', 'sound', 'music', 'synthesizer', 'frequency', 'waveform', 'oscillator', 'reverb', 'filter', 'beat'],
+    social: ['community', 'engagement', 'viewer', 'follower', 'chat', 'moderation', 'streaming', 'audience'],
+    brain: ['ai', 'machine learning', 'neural', 'llm', 'model', 'training', 'inference', 'optimization', 'algorithm'],
+    tile: ['tile', 'map', 'biome', 'terrain', 'procedural generation', 'world', 'dungeon', 'level design'],
+    security: ['security', 'vulnerability', 'exploit', 'cve', 'owasp', 'penetration', 'hardening', 'encryption'],
+    research: ['paper', 'arxiv', 'study', 'methodology', 'experiment', 'hypothesis', 'data', 'analysis'],
+};
+function detectApplicableEngines(query) {
+    const lower = query.toLowerCase();
+    const applicable = [];
+    for (const [engine, keywords] of Object.entries(ENGINE_KEYWORDS)) {
+        for (const kw of keywords) {
+            if (lower.includes(kw)) {
+                applicable.push(engine);
+                break;
+            }
+        }
+    }
+    // Always include 'general' as a catch-all
+    if (applicable.length === 0)
+        applicable.push('general');
+    return applicable;
+}
+// ─── Summarize with Ollama ──────────────────────────────────────
+async function summarizeWithOllama(query, content) {
+    const available = await isOllamaAvailable();
+    if (!available) {
+        // Fallback: extract first ~500 chars as basic summary
+        return content.slice(0, 500).trim() + (content.length > 500 ? '...' : '');
+    }
+    const truncatedContent = content.slice(0, 8000); // keep prompt under token limit
+    const prompt = [
+        'You are a research assistant. Summarize the following web content in relation to the research query.',
+        'Extract key findings as bullet points. Be concise and factual.',
+        '',
+        `Research query: "${query}"`,
+        '',
+        'Web content:',
+        truncatedContent,
+        '',
+        'Provide:',
+        '1. A 2-3 sentence summary',
+        '2. Key findings as bullet points (prefix with -)',
+        '3. Any actionable techniques or insights',
+    ].join('\n');
+    const result = await ollamaGenerate(prompt);
+    return result || content.slice(0, 500).trim() + (content.length > 500 ? '...' : '');
+}
+// ─── Autonomous Topic Generation ────────────────────────────────
+const AUTONOMOUS_TOPIC_TEMPLATES = [
+    'Canvas 2D {technique} optimization',
+    '{biome} pixel art reference techniques',
+    'ROM hack {technique} implementation',
+    'procedural {feature} generation algorithms',
+    'indie game {effect} techniques 2026',
+    'retro game {style} rendering methods',
+    'lo-fi aesthetic {element} techniques',
+    'web audio API {feature} tutorial',
+    'terminal UI {technique} best practices',
+    'AI agent {capability} implementation',
+    'real-time {effect} in JavaScript',
+    'pixel art {style} color palettes',
+];
+const TOPIC_FILLS = {
+    technique: ['parallax scrolling', 'palette cycling', 'dithering', 'scanline effects', 'sprite animation', 'tile blending', 'water reflection', 'particle systems'],
+    biome: ['forest', 'ocean', 'desert', 'mountain', 'cave', 'tundra', 'swamp', 'volcanic'],
+    feature: ['terrain', 'vegetation', 'weather', 'rivers', 'clouds', 'fire', 'fog', 'snow'],
+    effect: ['lighting', 'shadow', 'bloom', 'chromatic aberration', 'CRT filter', 'vignette', 'glow', 'distortion'],
+    style: ['SNES', 'GBA', 'NES', 'Amiga', 'PC-98', 'C64', 'MSX', 'Genesis'],
+    element: ['grain', 'noise', 'halftone', 'dot matrix', 'glitch', 'static', 'tape hiss'],
+    capability: ['tool use', 'memory', 'planning', 'self-evaluation', 'context management', 'learning'],
+};
+export function autonomousResearchTopics() {
+    const topics = [];
+    const now = Date.now();
+    for (const template of AUTONOMOUS_TOPIC_TEMPLATES) {
+        let topic = template;
+        const placeholders = template.match(/\{(\w+)\}/g) || [];
+        for (const placeholder of placeholders) {
+            const key = placeholder.replace(/[{}]/g, '');
+            const options = TOPIC_FILLS[key];
+            if (options) {
+                // Deterministic-ish pick based on time + template hash
+                const hash = template.length + key.length + (now % 10000);
+                const pick = options[hash % options.length];
+                topic = topic.replace(placeholder, pick);
+            }
+        }
+        topics.push(topic);
+    }
+    return topics;
+}
+// ─── Core Engine Functions ──────────────────────────────────────
+export function initResearchEngine() {
+    const completed = loadResults();
+    const engine = {
+        queue: [],
+        completed,
+        activeTask: null,
+        lastResearchFrame: 0,
+        researchInterval: DEFAULT_RESEARCH_INTERVAL,
+        topicsOfInterest: autonomousResearchTopics().slice(0, 5),
+    };
+    engineState = engine;
+    return engine;
+}
+export function getResearchEngine() {
+    if (!engineState)
+        return initResearchEngine();
+    return engineState;
+}
+export function queueResearch(engine, query, purpose, source) {
+    const task = {
+        id: randomUUID().slice(0, 8),
+        query,
+        purpose,
+        source,
+        status: 'queued',
+        startedAt: 0,
+    };
+    engine.queue.push(task);
+    return task;
+}
+export function getResearchForEngine(engine, engineName) {
+    return engine.completed.filter(r => r.applicableTo.includes(engineName));
+}
+// ─── Research Execution ─────────────────────────────────────────
+async function executeResearch(task) {
+    const { ensureLazyToolsLoaded, executeTool } = await import('./index.js');
+    await ensureLazyToolsLoaded();
+    // Search
+    task.status = 'searching';
+    task.startedAt = Date.now();
+    const searchResult = await executeTool({
+        id: `research_search_${task.id}`,
+        name: 'kbot_search',
+        arguments: { query: task.query },
+    });
+    // Extract links from search results
+    const links = extractLinks(searchResult.result);
+    // Read top results
+    task.status = 'reading';
+    let content = searchResult.result; // use search results as baseline
+    const readContents = [];
+    for (const url of links.slice(0, MAX_LINKS_TO_READ)) {
+        try {
+            const readResult = await executeTool({
+                id: `research_read_${task.id}_${readContents.length}`,
+                name: 'kbot_browse',
+                arguments: { url },
+            });
+            if (!readResult.error) {
+                readContents.push(readResult.result);
+            }
+        }
+        catch {
+            // Skip failed reads — non-fatal
+        }
+    }
+    if (readContents.length > 0) {
+        // Combine: search overview + page content (trimmed)
+        content = [
+            '## Search Results Overview',
+            searchResult.result.slice(0, 2000),
+            '',
+            ...readContents.map((c, i) => [
+                `## Source ${i + 1}`,
+                c.slice(0, 4000),
+            ].join('\n')),
+        ].join('\n');
+    }
+    // Summarize using local Ollama
+    task.status = 'summarizing';
+    const summary = await summarizeWithOllama(task.query, content);
+    task.status = 'complete';
+    return {
+        taskId: task.id,
+        query: task.query,
+        summary,
+        sources: links.slice(0, 5),
+        keyFindings: extractKeyFindings(summary),
+        applicableTo: detectApplicableEngines(task.query),
+        timestamp: Date.now(),
+    };
+}
+// ─── Tick (Frame Loop Integration) ──────────────────────────────
+export async function tickResearch(engine, frame) {
+    // Not time yet
+    if (frame - engine.lastResearchFrame < engine.researchInterval)
+        return null;
+    // Already working on something
+    if (engine.activeTask)
+        return null;
+    // Auto-queue autonomous topics if queue is empty
+    if (engine.queue.length === 0) {
+        const topics = autonomousResearchTopics();
+        if (topics.length > 0) {
+            // Pick one topic we haven't researched recently
+            const recentQueries = new Set(engine.completed.slice(-20).map(r => r.query));
+            const fresh = topics.find(t => !recentQueries.has(t));
+            if (fresh) {
+                queueResearch(engine, fresh, 'Autonomous discovery of new techniques', 'autonomous');
+            }
+        }
+    }
+    // Nothing to do
+    if (engine.queue.length === 0)
+        return null;
+    // Pick next task
+    const task = engine.queue.shift();
+    engine.activeTask = task;
+    engine.lastResearchFrame = frame;
+    try {
+        const result = await executeResearch(task);
+        // Store result
+        engine.completed.push(result);
+        if (engine.completed.length > MAX_COMPLETED) {
+            engine.completed = engine.completed.slice(-MAX_COMPLETED);
+        }
+        saveResults(engine.completed);
+        engine.activeTask = null;
+        return {
+            type: 'summarize_complete',
+            task,
+            result,
+            speech: `Research complete: "${task.query}" — found ${result.keyFindings.length} key findings.`,
+        };
+    }
+    catch (err) {
+        task.status = 'failed';
+        engine.activeTask = null;
+        return {
+            type: 'failed',
+            task,
+            speech: `Research failed for "${task.query}": ${err instanceof Error ? err.message : String(err)}`,
+        };
+    }
+}
+// ─── Tool Registration ──────────────────────────────────────────
+export function registerResearchEngineTools() {
+    // ── research_queue ──────────────────────────────────────────
+    registerTool({
+        name: 'research_queue',
+        description: 'Queue a web research task for the Research Engine. The engine will search the web, read relevant pages, and summarize findings using local Ollama. Results are persisted to ~/.kbot/research-results.json.',
+        parameters: {
+            query: { type: 'string', description: 'Search query / research topic', required: true },
+            purpose: { type: 'string', description: 'Why this research is needed (guides summarization)' },
+            source: { type: 'string', description: 'Source of the request: evolution | brain | narrative | user | autonomous (default: user)' },
+        },
+        tier: 'free',
+        async execute(args) {
+            const query = String(args.query);
+            const purpose = String(args.purpose || 'User-requested research');
+            const source = (String(args.source || 'user'));
+            const validSources = ['evolution', 'brain', 'narrative', 'user', 'autonomous'];
+            const safeSource = validSources.includes(source) ? source : 'user';
+            const engine = getResearchEngine();
+            const task = queueResearch(engine, query, purpose, safeSource);
+            const queuePos = engine.queue.length;
+            const ollamaUp = await isOllamaAvailable();
+            return [
+                `# Research Queued`,
+                ``,
+                `**Task ID**: ${task.id}`,
+                `**Query**: ${task.query}`,
+                `**Purpose**: ${purpose}`,
+                `**Source**: ${safeSource}`,
+                `**Queue position**: ${queuePos}`,
+                `**Ollama**: ${ollamaUp ? 'available (will summarize with ' + OLLAMA_MODEL + ')' : 'unavailable (will use raw extraction)'}`,
+                ``,
+                engine.activeTask
+                    ? `Currently researching: "${engine.activeTask.query}" (${engine.activeTask.status})`
+                    : 'No active research — will start on next tick.',
+                ``,
+                `Completed research: ${engine.completed.length} results on file.`,
+            ].join('\n');
+        },
+    });
+    // ── research_results ────────────────────────────────────────
+    registerTool({
+        name: 'research_results',
+        description: 'View completed research results. Filter by engine, query, or show all. Results are persisted across sessions in ~/.kbot/research-results.json.',
+        parameters: {
+            engine: { type: 'string', description: 'Filter results applicable to a specific engine (evolution, narrative, audio, social, brain, tile, security, research, general)' },
+            query: { type: 'string', description: 'Filter results by query substring' },
+            limit: { type: 'number', description: 'Max results to return (default 10)' },
+        },
+        tier: 'free',
+        async execute(args) {
+            const engineFilter = args.engine ? String(args.engine).toLowerCase() : undefined;
+            const queryFilter = args.query ? String(args.query).toLowerCase() : undefined;
+            const limit = Number(args.limit) || 10;
+            const engine = getResearchEngine();
+            let results = engine.completed;
+            if (engineFilter) {
+                results = results.filter(r => r.applicableTo.includes(engineFilter));
+            }
+            if (queryFilter) {
+                results = results.filter(r => r.query.toLowerCase().includes(queryFilter));
+            }
+            results = results.slice(-limit);
+            if (results.length === 0) {
+                return [
+                    '# Research Results',
+                    '',
+                    'No research results found.',
+                    engineFilter ? `Filter: engine=${engineFilter}` : '',
+                    queryFilter ? `Filter: query contains "${queryFilter}"` : '',
+                    '',
+                    `Total results on file: ${engine.completed.length}`,
+                    `Queue: ${engine.queue.length} tasks pending`,
+                    engine.activeTask ? `Active: "${engine.activeTask.query}" (${engine.activeTask.status})` : 'No active research.',
+                ].filter(Boolean).join('\n');
+            }
+            const lines = ['# Research Results', ''];
+            for (const r of results) {
+                lines.push(`## ${r.query}`);
+                lines.push(`*${new Date(r.timestamp).toISOString().slice(0, 16)}* | Sources: ${r.sources.length} | Engines: ${r.applicableTo.join(', ')}`);
+                lines.push('');
+                lines.push(r.summary.slice(0, 800));
+                if (r.keyFindings.length > 0) {
+                    lines.push('');
+                    lines.push('**Key findings:**');
+                    for (const f of r.keyFindings.slice(0, 5)) {
+                        lines.push(`- ${f}`);
+                    }
+                }
+                if (r.sources.length > 0) {
+                    lines.push('');
+                    lines.push('**Sources:**');
+                    for (const s of r.sources.slice(0, 3)) {
+                        lines.push(`- ${s}`);
+                    }
+                }
+                lines.push('');
+                lines.push('---');
+                lines.push('');
+            }
+            lines.push(`Showing ${results.length} of ${engine.completed.length} total results.`);
+            lines.push(`Queue: ${engine.queue.length} pending | Active: ${engine.activeTask ? `"${engine.activeTask.query}"` : 'none'}`);
+            return lines.join('\n');
+        },
+    });
+    // ── research_now ────────────────────────────────────────────
+    registerTool({
+        name: 'research_now',
+        description: 'Execute a research task immediately (bypasses the queue). Searches the web, reads top results, summarizes with Ollama, and returns findings. Use for urgent research needs.',
+        parameters: {
+            query: { type: 'string', description: 'Search query / research topic', required: true },
+            purpose: { type: 'string', description: 'Why this research is needed' },
+        },
+        tier: 'free',
+        timeout: 120_000, // 2 min — research involves multiple web requests + Ollama
+        async execute(args) {
+            const query = String(args.query);
+            const purpose = String(args.purpose || 'Immediate research request');
+            const engine = getResearchEngine();
+            const task = {
+                id: randomUUID().slice(0, 8),
+                query,
+                purpose,
+                source: 'user',
+                status: 'queued',
+                startedAt: 0,
+            };
+            try {
+                const result = await executeResearch(task);
+                // Store result
+                engine.completed.push(result);
+                if (engine.completed.length > MAX_COMPLETED) {
+                    engine.completed = engine.completed.slice(-MAX_COMPLETED);
+                }
+                saveResults(engine.completed);
+                const lines = [
+                    '# Research Complete',
+                    '',
+                    `**Query**: ${result.query}`,
+                    `**Sources found**: ${result.sources.length}`,
+                    `**Applicable engines**: ${result.applicableTo.join(', ')}`,
+                    '',
+                    '## Summary',
+                    '',
+                    result.summary,
+                    '',
+                ];
+                if (result.keyFindings.length > 0) {
+                    lines.push('## Key Findings', '');
+                    for (const f of result.keyFindings) {
+                        lines.push(`- ${f}`);
+                    }
+                    lines.push('');
+                }
+                if (result.sources.length > 0) {
+                    lines.push('## Sources', '');
+                    for (const s of result.sources) {
+                        lines.push(`- ${s}`);
+                    }
+                    lines.push('');
+                }
+                lines.push(`*Task ${task.id} completed in ${((Date.now() - task.startedAt) / 1000).toFixed(1)}s*`);
+                return lines.join('\n');
+            }
+            catch (err) {
+                return `Research failed: ${err instanceof Error ? err.message : String(err)}`;
+            }
+        },
+    });
+}
+//# sourceMappingURL=research-engine.js.map