QANATIX

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.

ProblemQANATIX solution
AI can't see your ERP/database data10+ connectors: PDF, CSV, JSON, XML, Postgres, MySQL, MongoDB, Neo4j, SAP IDoc, streaming
Hallucinated answersEvery 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 isolationEvery tenant is fully isolated — database, vectors, cache
Vendor lock-inWorks 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

On this page