Semantic Toolkit

Turn any database into a
deployable knowledge layer.

Upload a schema, teach it in plain English, and ship what it learns as a dashboard, an embeddable widget, or a fully offline executable — no AI calls required once it's trained.

Try it out now / Login → See how it works
The pipeline

Six stages. One growing knowledge layer.

Every answer it gets right becomes something the next question can reuse — instead of being re-derived from scratch, every time.

1
Upload & profile
Upload a SQL dump, or connect a live SQLite, MySQL, PostgreSQL, SQL Server, or MongoDB database. Every table, column, type, and foreign key is parsed automatically — no manual schema entry.
schema_profile.json
2
AI interview
One focused question at a time about what each table means, which fields are ambiguous, and what status values actually represent. When an answer pins down something concrete, it's captured as a reusable Rule automatically.
interview_state.jsonrules learned
3
Teaching — question & answer
Around 100 candidate questions generated from the schema. For each one, a step-by-step path to the answer with real data at every step — corrected, refined in plain English, or walked through tiny confirmed steps until it's right.
knowledge_store.json
4
Intents & Rules accumulate
Every confirmed answer becomes an Intent — a reusable pattern with its fields, filters, and every phrasing it should match. Every concept pinned down along the way becomes a Rule the next question reuses directly.
↺ Both feed back into stage 3 — every future question gets answered using what's already been taught.
5
Actions, dashboards & insights
Automated actions and triggers, drag-and-resize dashboard panels, and AI-grounded narrative insights — all built directly on the same trained intents and rules.
6
Export
Package the finished knowledge layer as a standalone Windows .exe — fully offline, no network or AI calls at runtime — or as an embeddable widget for an existing site or product.
Architecture

Train once, deploy three ways

The same knowledge layer powers the tool itself, your dashboards, and whatever you ship to end users.

Intents + Rules
knowledge_store.json
→
The tool itself
Teaching & Explore — where the layer keeps getting taught and corrected.
→
Dashboards & insights
In-app charts, panels, and AI-grounded narrative analysis.
→
The exported product
A standalone .exe or embeddable widget — deterministic, no AI calls at runtime.
Capabilities

What the finished product can do

Natural-language querying
Plain-English questions answered deterministically, every time.
Scheduling & monitoring
Automated triggers on a schedule or a condition — email, export, alert.
Embeddable widget
Drop a trained query tool into an existing site or product.
Air-gapped executable
Fully offline once trained — no network or AI calls at runtime.
Business insights
AI-grounded narrative analysis of trends, gaps, and recommendations.
Dashboards
Drag-and-resize panels of charts and tables, backed by real intents.
Consistent answers
Rules guarantee the same business concept is handled the same way every time.
Schema documentation
AI interview builds a structured semantic map of any undocumented database.

Contact us

Questions about Semantic Toolkit? Email us at info@semantictoolkit.com
We read every message and aim to reply within one working day.