Valenova AI Adoption

The Problem

Enterprise teams sit on decades of structured data across Oracle, SAP, PostgreSQL, and Snowflake, but extracting insights requires writing SQL, filing tickets with the data team, or paying for expensive BI dashboards that still can't answer ad-hoc questions.

Cloud-based AI tools require data migration, which enterprise security teams block. What was needed was a way to query legacy databases using natural language, without any data leaving the company's infrastructure and without migrating a single table.

Constraints

• Zero-knowledge was a hard requirement, no raw query results could ever leave the client’s own infrastructure.

• Must connect to existing Oracle, SAP, PostgreSQL, MySQL, and Snowflake instances without any migration or schema changes.

• The local MCP server had to run silently on Windows and macOS without IT team involvement at install time.

• Sub-1-second latency for multi-hop questions across federated databases, the AI agent layer had to be ruthlessly efficient.

• Had to pass enterprise security review: no persistent connections, read-only JDBC/ODBC drivers, encrypted credential storage.

How Nexolve Built It

Valenova is a cross-platform desktop application with a custom Model Context Protocol server. The AI agent layer sits entirely on the client machine, connecting to enterprise databases via read-only drivers, zero data ever leaves the local environment.

1. 1

   Electron desktop app with local MCP server

   The desktop client bundles a React frontend and an Express.js local server running a custom Model Context Protocol (MCP) implementation. All database queries execute locally, raw data never leaves the client's infrastructure.

2. 2

   Custom MCP server for enterprise databases

   The MCP server connects to Oracle, SAP, PostgreSQL, MySQL, and Snowflake via read-only JDBC/ODBC drivers. It exposes a standardised tool interface that the AI agent layer invokes to decompose multi-hop questions into routed sub-queries.

3. 3

   Zero-knowledge architecture

   Supabase stores only encrypted query plans and user metadata, never raw query results or database contents. Google OAuth handles authentication across desktop and web surfaces. The architecture passed enterprise security review without exceptions.

4. 4

   AI agent orchestration

   The AI agent decomposes complex business questions into sub-queries, routes each to the correct data source via MCP, federates partial results, and presents a unified natural language answer, all in under one second.

5. 5

   License-based billing

   LemonSqueezy handles both perpetual and subscription license billing. Webhook handlers update seat entitlements in Supabase within seconds of payment confirmation, with automatic license key validation on app launch.

Tech Stack

Outcomes

• Zero-knowledge architecture, no raw data ever leaves the client's infrastructure.

• Sub-1-second query latency for multi-hop questions across federated enterprise databases.

• No migration required, connects to existing Oracle, SAP, PostgreSQL, MySQL, and Snowflake instances as-is.

• Passed enterprise security review without exceptions due to local-first architecture.

• Custom MCP server provides a standardised tool interface reusable across future AI agent products.