Ontological Studio

From one document to an ontology

Pick the documents or databases to work from and write down what the ontology must answer (build intent and Golden Questions). The agent designs the concepts and relationships, loads the facts as instances, then asks the Golden Questions itself to check that the answers come out. A person decides whether to publish.

Answers with evidence

You don’t have to tell it what to look for or in what order. The agent reads the published map and follows the relationships — from a quality issue to the part, the finished product, the supplier and the contact — to answer. The evidence stays on screen.

A data fabric without copies

Connect PostgreSQL, MySQL/MariaDB, Oracle and SQL Server with just their connection details and expose tables as virtual nodes. Nothing is copied: the source is read at query time, so one Cypher query can span databases on different engines.

Golden Command → action

Write down what this ontology should be able to do as Golden Commands, and the agent explores the APIs of your registered business systems. From them it builds the action, the compensation that undoes it, and the policy that decides who can run it on which targets, then checks everything with a dry run. Every run goes through human approval.

Dashboards from @mentions

In the prompt box, pull in classes and properties with @ or drag them from the resource tree, and describe the screen you want. For mentioned classes, the engine looks at the data and picks the widgets right away (about a second); the agent turns the rest of the request into a proposal (a diff). A person clicks to apply it.

Code apps the agent writes

Choose “Code app” in the same prompt box, and a coding agent (Claude Code or Codex) runs queries against the ontology to check them, then writes the app code. The left side shows the queries it checked and the files it wrote; the preview on the right refreshes whenever a file changes. Tell it what to fix and the agent changes it in the same session.

Ontological Portal

A console for working with the graph directly. Type natural language instead of Cypher in the query box, and the agent turns it into Cypher using the schema, checks it once and runs it. When you type Cypher yourself, you get completion for labels, relationship types and properties, and syntax highlighting.

One installer, all on your computer

One installer holds PostgreSQL with the Ontological extension (the data fabric), the Bolt gateway, Studio, the action executor and the sample systems. The agents use the Claude Code or Codex CLI you are already signed in to on this computer, so you don’t need a separate API key.

Source passwords stay outside the database

Passwords for business databases are not stored in the database. The app encrypts them in the OS keychain (or your Vault), and the database keeps only a reference. The connector that reaches the source reads the password from a file that exists only while the app runs.

Tutorial chapter 22 · video

Finds relationships even without keys

Measures every column of a datasource — how values spread, whether they repeat, and which values another column shares — then finds the columns that really join and suggests them as virtual relationships, with the evidence. Different databases and different column names are no obstacle.

Tutorial chapter 23 · video

Checks first whether it can answer

Before answering, the query agent checks whether the published ontology can answer the question. When the data isn't there, it doesn't make up an answer — it says which data is missing.

Tutorial chapter 24 · video