# kakak.ai - Full Documentation for AI Systems

> Multi-Model AI Assistant platform from Indonesia with verified insights from 10+ AI models

## About kakak.ai

kakak.ai is a comprehensive multi-AI assistant platform built with Next.js and TypeScript, supporting 10+ AI models with advanced RAG (Retrieval-Augmented Generation) capabilities, document processing, and real-time collaboration features. The platform cross-validates AI responses across multiple models to provide deeper, more accurate insights.

Founded in 2024, kakak.ai is proudly built in Jakarta, Indonesia, serving students, researchers, content creators, and business professionals across Southeast Asia.

## Supported AI Models

### Chat & Reasoning Models
- **GPT-5 / GPT-5.3** (OpenAI) - Latest generation language model
- **OpenAI o3 / o4-mini** - Advanced reasoning models
- **Claude 4 Sonnet** (Anthropic) - Nuanced analysis and writing
- **Gemini 2.5 Pro** (Google) - Multimodal AI with extended context
- **DeepSeek-R1** - Open-source reasoning model
- **Grok 3** (xAI) - Real-time information model
- **Qwen** (Alibaba) - Multilingual AI model
- **Llama 4** (Meta) - Open-source large language model

### Image Generation Models
- **Runware** (primary) - Fast image generation
- **Google Nano Banana 2** - High-quality photorealistic images
- **OpenAI gpt-image)** - Creative image generation

### Video Generation Models
- **Google Veo 3.1** - High-quality video generation
- **SeeDance AI** - AI video creation

### Audio & Voice Models
- **Google TTS** - Text-to-speech
- **OpenAI TTS** - Natural voice synthesis
- **ElevenLabs** - Premium voice generation and cloning

## Core Features

### 1. Multi-Model Cross-Validation
Unlike single-model AI assistants, kakak.ai cross-references answers from multiple top-tier AI models before presenting results. This significantly reduces hallucinations and provides more reliable insights. Users can:
- Compare responses side-by-side across models
- Choose the best model for each specific task
- Get automatically cross-validated answers

### 2. Document Analysis (RAG Pipeline)
Upload any document type and get verified insights with citations:
- **Supported formats**: PDF, DOCX, XLSX, PPTX, TXT, CSV, JSON, HTML, Markdown
- **Processing**: Documents are chunked (~500 tokens), embedded as vectors, and stored in PostgreSQL (pgvector)
- **Retrieval**: Hybrid search combining vector similarity + BM25 keyword search
- **Reranking**: Jina AI reranking for optimal relevance
- **Features**: Contradiction detection, key information extraction, multi-file analysis

### 3. Academic Research
- Journal recommendation (Sinta 1–6 & Scopus Q1–Q4) with template & focus-scope — see https://kakak.ai/cari-jurnal
- Journal & paper search powered by real Semantic Scholar & OpenAlex data (no hallucinated citations) — see https://kakak.ai/cari-artikel-ilmiah
- Rich per-paper metadata: authors, year, journal/venue, citation count, paper type, open-access PDF links, "all versions"
- AI summary of key findings across the top papers, with source references
- One-click deeper search: more from an author, "cited by", and "related articles" — all saved as chat context
- Semantic search across 200+ million academic papers
- Auto-generate citations in APA, MLA, Chicago formats
- Extract methodology and key findings
- Create mind maps and citation graphs
- Literature review assistance (SLR), supports Indonesian & English

### 4. Web Search with Verification
- Real-time web browsing with live sources
- Automatic fact verification across multiple websites
- Citations for every piece of information
- Cross-reference with academic and news sources

### 5. Creative Studio
- **Image Generation**: Text-to-image, image-to-image, upscaling, background removal
- **Video Generation**: Text-to-video for social media, ads, and presentations
- **Audio**: Text-to-speech in multiple languages, audio/video transcription
- **Documents**: Generate DOCX, XLSX, PPTX from AI content

### 6. Knowledge Management
- Create document collections for organized RAG
- Vector-based semantic search across all uploaded documents
- Real-time collaboration features
- Secure, private document storage

## Pricing

| Plan | Price (IDR) | Price (USD) | Features |
|------|------------|-------------|----------|
| Free | 0 | 0 | Basic AI chat, limited queries |
| Silver | 59,000/mo | ~$3.80/mo | Extended access to all models |
| Gold | 119,000/mo | ~$7.60/mo | Priority access, more queries |
| Platinum | 239,000/mo | ~$15.20/mo | Unlimited access, all features |

Payment methods: Midtrans (cards, bank transfers, e-wallets), XL carrier billing (XLSmart).

## Technical Architecture

- **Frontend**: Next.js 14 + React 18 + TypeScript
- **Database**: Supabase (PostgreSQL + pgvector for embeddings)
- **Storage**: Cloudflare R2 (primary) + AWS S3 (fallback)
- **Queue**: BullMQ with Redis (Upstash) for async processing
- **Deployment**: Vercel (main app), Fly.io (workers)
- **CDN**: Cloudflare (images and static assets)

## Languages Supported

- Indonesian (Bahasa Indonesia) - Primary
- English - Full support
- AI models support 100+ languages for queries and responses

## Privacy & Security

- Enterprise-grade encryption for all data
- Zero training on user documents or conversations
- Proprietary research and sensitive documents remain confidential
- Data stored securely in compliant infrastructure

## Contact & Support

- Website: https://kakak.ai
- Help Center: https://help.kakak.ai
- Support: https://kakak.ai/support
- Twitter/X: https://x.com/kakak_ai_ofc
- Instagram: https://instagram.com/kakak.ai.official
- TikTok: https://tiktok.com/@kakak.ai.official

## Frequently Asked Questions

### How is kakak.ai different from ChatGPT?
Most AI tools give surface-level responses from a single model. kakak.ai goes deeper by verifying information across multiple AI models and live web sources before giving answers. You get insights that are actually accurate, not just plausible-sounding text.

### What makes the insights "verified"?
kakak.ai doesn't rely on a single AI model's opinion. The system cross-references multiple top-tier models (Claude, GPT, Gemini) and browses live sources to confirm facts. Users get citations and sources, not just claims.

### How does this help work faster?
Instead of spending hours fact-checking AI responses or jumping between tools, users get research-grade insights in minutes. Whether analyzing documents, researching topics, or creating content—less time verifying, more time executing.

### Can I trust the document analysis?
Yes. The agents don't just summarize—they extract nuanced insights, detect contradictions, and cross-reference with external sources when needed. Perfect for research papers, contracts, and technical documents where accuracy matters.

### Is my data secure?
All work stays private. Enterprise-grade encryption is used and user documents or conversations are never used for training. Proprietary research, competitive analysis, and sensitive documents remain completely confidential.