Skip to content

Releases: Unknown-086/Semantic-Database-Engine

Release list

Semantic Database Engine v1.0.0

Semantic Database Engine v1.0.0 Pre-release
Pre-release

Choose a tag to compare

View all tags
@Unknown-086 Unknown-086 released this 02 Mar 15:27
v1.0.0
97245c9

Semantic Database Engine v1.0.0

A self-contained Windows console app that lets you query any MSSQL or MySQL database
in plain English — no SQL knowledge required. It uses ONNX embeddings for local
semantic search and Gemini as the reasoning layer to generate natural-language answers.

What's in the zip

File Description
SemanticDb.Console.exe Self-contained .NET 8 executable (no runtime install needed)
appsettings.json Configuration file — fill in your API key

Requirements

  • Windows x64
  • A running MSSQL or MySQL database
  • A Gemini API key (free tier works — get one here)

Setup

  1. Extract the zip into any folder
  2. Open appsettings.json and fill in:
    • "GeminiApiKey" → your Gemini API key
  3. Run SemanticDb.Console.exe
  4. On first run, it will download the ONNX embeddings model and then type index to embed your database
  5. Then ask questions: ask How many customers do we have?

Commands

Command Description
index Scans and indexes your database (incremental — safe to re-run)
ask <question> Ask anything in plain English
status Show index stats (tables, document count, last indexed)
test-connection Verify your database connection
help List all available commands

What it can answer

  • Counts — "How many orders do we have?"
  • Filters — "Show me completed orders from London"
  • Aggregates — "What is the total revenue from all orders?"
  • Semantic search — "Find customers who complained about delivery"
  • Overviews — "Give me a summary of what's in this database"
  • Cross-table reasoning — handled gracefully with honest scope limits

Security

Sensitive tables and columns are automatically excluded during indexing.
Passwords, tokens, API keys, payment data, PII fields, and audit logs are
never indexed or exposed to the LLM. See appsettings.json → Security
to customize exclusion rules for your schema.

Known limitations in this version

  • Verbless filter queries (e.g. customers from Liverpool) fall back to semantic search instead of applying a city filter — add a verb prefix to work around: show me customers from Liverpool
  • Adjective-qualified entities (e.g. VIP customers) are not yet split into entity + segment filter

These will be addressed in v1.1.0.

Loading