QANATIX Documentation
Data middleware for AI — connect your enterprise data, query it from any AI model with zero hallucination.
What is QANATIX?
QANATIX is a data middleware that sits between your enterprise data sources and AI models. You push in messy data — CSVs, JSON, PDFs, database dumps, SAP exports — and QANATIX normalizes it, generates vector embeddings, and makes it queryable via MCP or REST API.
Your AI agent calls qanatix_search() and gets back ranked, verified results from your actual data — not scraped web content, not hallucinated guesses.
Why QANATIX?
LLMs hallucinate on 61% of business data queries. Pricing, stock levels, compliance data, supplier specs — all locked in ERPs and databases that AI can't reach. QANATIX fixes this.
| Problem | QANATIX solution |
|---|---|
| AI can't see your ERP/database data | 10+ connectors: PDF, CSV, JSON, XML, Postgres, MySQL, MongoDB, Neo4j, SAP IDoc, streaming |
| Hallucinated answers | Every result comes from your verified data with source attribution |
| 800+ tokens per result from web scraping | ~120 tokens per result, structured and compressed |
| No tenant isolation | Every tenant is fully isolated — database, vectors, cache |
| Vendor lock-in | Works with any LLM: Claude, GPT, Gemini, open-source |
Architecture
Your Data Sources QANATIX Pipeline AI Consumers
───────────────── ───────────────────────── ─────────────────
PDF / CSV / JSON ──→ Extract Claude (MCP)
PostgreSQL ──→ Normalize GPT-4o (REST)
MySQL / MongoDB ──→ Validate (JSON Schema) ──→ Gemini (REST)
Neo4j ──→ Embed (dense + sparse) LangChain
SAP IDoc XML ──→ Index (Qdrant hybrid) Your Agent
API / Webhooks ──→ Cursor IDE
NDJSON stream ──→Key features
- 10+ data sources — file upload, database pull, real-time push
- Hybrid search — semantic + keyword + identifier detection, < 200ms
- MCP native — Claude calls
qanatix_search()directly via Streamable HTTP - REST API — works with any LLM or application
- Self-service verticals — define a JSON schema, start ingesting
- Tenant isolation — every query scoped to your tenant_id
- Self-hostable — Docker Compose on your infra, air-gapped support
- EU Cloud — managed deployment in Frankfurt
- Reranking — BGE-reranker-v2-m3 cross-encoder for precision
- Response caching — 30s response cache, 1h query embedding cache
Quick links
- Quickstart — zero to first query in 5 minutes
- Authentication — API keys and scopes
- Data Ingestion — connect your data sources
- Search — how hybrid search works
- MCP Integration — connect to Claude
- Self-hosted — deploy on your infra
- API Reference — all 36 endpoints