nivq
This install section is for Enterprise / on-prem customers who self-host nivq.

What is nivq?

The governed, on-prem agentic AI data assistant for regulated enterprises — read-only by design, audit-first, and yours to run.

nivq is a governed agentic AI platform for regulated enterprise data, by Nivorbit. People ask questions in plain language; nivq reasons across multiple steps, writes and runs the SQL, charts the result, and learns from past queries — all while staying read-only by design and keeping your data on your infrastructure.

It sits on the governed side of the 2026 agentic-AI split: the agent acts across many tool calls and gets smarter over time, but it never mutates your data on its own. Any change is routed through an explicit human approval queue, and every validation, execution, and permission denial is written to a per-workspace audit log.

Why teams choose nivq

The one-sentence version

Ask your data anything, in your own words — and stay in control. Your data never leaves your servers, nothing runs without your approval, and every query is logged.

  • Read-only by default. The baseline agent can only read. Writes require explicit, audited human approval — a wrong answer can never silently train tomorrow's wrong answer.
  • Your data stays put. Run nivq entirely on-prem (or in your own cloud). With the MCP-native deployment mode, nivq never opens a direct database connection at all — useful when policy forbids any vendor-initiated DB access.
  • Bring your own LLM. Each agent is wired to its own database and its own LLM provider — OpenAI, Anthropic, Azure, or a fully local model via Ollama for air-gapped sites.
  • Compliance-first architecture. Five-layer multi-tenant isolation, AES-256 encryption, OIDC/SSO, and an architecture built to be KVKK / BDDK / GDPR / EU AI Act compatible.
  • Auditable. Every SQL validation, execution, and denial is captured in a partitioned, per-workspace audit log.

How it works, in brief

  1. You create a workspace and configure one or more agents. Each agent connects to an external database and uses its chosen LLM provider.
  2. A user chats with an agent in natural language. nivq generates SQL, validates it against a guard, executes the read, and renders charts or tables.
  3. The agent's pattern memory grows behind an approval gate, so it gets better at your data over time — without ever learning an unapproved answer.
  4. Everything is logged; entitlements are enforced by a signed, on-prem licence.

Ready to run it? Head to Requirements and then the Quickstart.