Upload any document and make it instantly searchable.

Your files are automatically split into optimized chunks, indexed for full-text search, and embedded into a vector database for semantic understanding.

• File Upload — Drag and drop PDF or TXT files up to 50 MB. Documents are parsed, chunked, and indexed automatically.
• URL Ingestion — Paste one or multiple web URLs. BABEH scrapes the main content and indexes it alongside your uploaded files.
• Smart Chunking — Configure chunk size (100–10,000 characters) and overlap (0–5,000 characters) to match your content type. Technical manuals need different chunking than marketing copy.
• Document Library — Browse, search, filter by type (file or URL), sort by chunk count, file size, or upload date. Paginate through 10, 25, or 50 records at a time.
• Edit & Re-index — Update document content at any time. BABEH automatically re-chunks and re-embeds the updated content. Re-index a single document or the entire library in one click.
• AI Content Improvement — Let the AI review your document chunks and suggest clearer, more complete rewrites. Review the suggestion, accept it, and save — all without leaving the editor.

Give your AI real-time access to structured pricing data.

A dedicated, schema-driven pricing database that integrates directly into the search pipeline. When a customer asks "how much does a Honda Veloz cost?", the AI pulls live pricing data — not stale document snippets.

• Full CRUD — Add, edit, and delete pricing records through a clean form interface.
• Bulk Import — Import from CSV or JSON. Column headers are automatically mapped to your schema fields.
• Export — Download your entire pricing database as CSV or JSON with schema version metadata included.
• Full-Text Search — Search across product name, SKU, category, and description with weighted relevance (product name weighted 2× higher than other fields).
• Smart Price Detection — Natural language price queries are understood automatically: "between 200 million and 300 million" — range search "under 1 billion" — upper-bound filter "budget 500 million" — lower-bound filter "around 300 million" — approximate search with ±20% tolerance
• VectorDB Sync — Push pricing records into the vector database so they appear in semantic search results alongside document content.

Structure your product data with 30+ fields including 15 boolean feature flags.

Purpose-built for automotive and product-heavy businesses. Every specification is searchable, filterable, and available to the AI during query time.

• Comprehensive Schema — Brand, model, variant, year, transmission, engine type, fuel type, drive system, body type, seating capacity, price, and free-text spec notes.
• 15 Feature Flags — Cooling seat, sunroof, 360 camera, wireless charging, cruise control, collision warning, lane assist, auto parking, Apple CarPlay, Android Auto, keyless entry, push start, rear camera, electric seat, heated seat.
• Three Import Formats — JSON, CSV, and Excel (.xlsx / .xls). The only data type in BABEH that supports Excel import.
• Dual Edit Mode — Edit through a structured form or switch to raw JSON for bulk field editing.
• Full-Text Search — Search across brand, model, variant, transmission, engine type, drive system, body type, and spec notes simultaneously.

Modify your database structure at runtime — no migrations, no downtime.

Add, rename, or remove fields from the pricing database without writing a single line of code. BABEH handles the ALTER TABLE, FTS rebuild, and VectorDB re-sync automatically.

• Runtime Field Management — Add new columns, rename existing ones, or remove fields you no longer need. Changes take effect immediately.
• 11 Configurable Properties Per Field — Name, display label, data type (text / number / integer / boolean), required flag, default value, display format (currency / badge / textarea / boolean badge), FTS indexed flag, FTS weight, VectorDB inclusion flag, VectorDB template, and sort order.
• Automatic Rebuild — When you change FTS or VectorDB settings, BABEH rebuilds the affected indexes in the background. A status indicator shows rebuild progress.
• Version History — Every schema change is versioned. Browse the full history and view a side-by-side diff between any two versions to see exactly what changed.

Superadmin access only.

Three search methods. One unified result.

BABEH doesn't rely on a single search technique. It combines vector similarity, keyword matching, and structured data lookup — then fuses the results with configurable weights.

• Vector Search — Semantic similarity via ChromaDB. Understands meaning, not just keywords. "affordable family car" matches "budget-friendly minivan" even though they share no words.
• BM25 Keyword Search — Full-text search via SQLite FTS5. Fast, precise, and excellent for exact product names, SKUs, and model numbers.
• Hybrid Fusion — Combine vector and BM25 results with configurable weights (default 50/50). Tune the balance to match your use case.
• Query Decomposition — Multi-topic queries like "compare Honda Veloz vs Toyota Avanza" are automatically split into sub-queries for more diverse, complete retrieval.
• Automatic Data Source Detection — The engine analyzes each query for pricing keywords (13 terms in Indonesian and English) and specification keywords (25+ terms). When detected, it automatically searches the relevant structured database in addition to the knowledge base.
• Relevance Threshold — Results below the configured threshold (default 0.25) are filtered out. Score visualization uses color coding: green for strong matches, yellow for moderate, red for weak.
• Citation System — Every AI response includes numbered source citations. Users can verify the information source for any claim in the response.
• Streaming Responses — Answers stream in real-time via Server-Sent Events. Users see the response forming word by word instead of waiting for the full answer.

Understand how your knowledge base is being used.

Five analytics views give you full visibility into queries, performance, token consumption, and user behavior.

• Summary Dashboard — Total queries, 7-day query volume, active days, average processing time, total token usage, model usage breakdown, and data volume in MB.
• LLM Usage Metrics — Track input tokens, output tokens, and total tokens over configurable periods (7, 14, 30, 60, or 90 days). Visualized as line charts for token trends and bar charts for daily query volume.
• Search Insights — Unique search terms, average words per query, top queries ranked by frequency, and an interactive word cloud generated from all search history.
• User Analytics — Browser distribution, operating system breakdown, device type split (desktop / mobile / tablet), and geographic distribution of users.
• Query History — Full searchable log of every query with detail view. Filter by date range. Export to CSV for offline analysis.

Add AI-powered search to any website with two lines of code.

A self-contained JavaScript widget that runs inside a Shadow DOM — completely isolated from your site's CSS and JavaScript. No dependencies. No conflicts.

• Two-Line Installation — Add a script tag and call KBaaSWidget.init(). That's it.
• Zero Dependencies — No jQuery, no React, no framework required. Works on any website.
• Shadow DOM Isolation — The widget's styles and scripts are fully encapsulated. Your site's CSS cannot break the widget, and the widget cannot break your site.
• Programmatic API — search(), clear(), destroy(), setTheme(), configure(), getConfig() — full control from your application code.
• Auto Theme Detection — Automatically matches the user's OS preference (light or dark). Manual override available via API.
• Responsive Design — Adapts to desktop, tablet, and mobile viewports. Four breakpoints ensure the widget looks right on every screen size.
• Markdown Rendering — AI responses render with full Markdown support: headers, lists, bold, code blocks, tables.
• Keyboard Shortcut — Ctrl+Enter to submit a query.

Four roles. Granular permissions. Full audit trail.

• Four Roles — Superadmin (full access), Manager (content + settings), Editor (content only), Analyst (read-only analytics and settings).
• User Management — Create, edit, deactivate, and delete user accounts. Assign roles at creation or change them later. Superadmin only.
• Activity Log — Every POST, PUT, and DELETE request is automatically logged with timestamp, user, IP address, HTTP method, endpoint, status code, and user agent. Login passwords are automatically redacted. 120-day retention. Filterable by search term, HTTP method, and date range. Export to CSV.

Fine-tune every aspect of the AI behavior without touching code.

When something doesn't look right, see exactly what the search engine is doing.

• Search Debugger — Run any query through the full pipeline and inspect all 9 stages: configuration, keyword detection, result summary, LLM context window, threshold filtering, hybrid fusion scores, raw vector results, raw BM25 results, and structured data (pricing + specs) results. Each result shows its relevance score with color-coded bars.
• VectorDB Viewer — Browse every record in the vector database. View embeddings, metadata, and source documents. Filter by source type. Superadmin only.