Questions in everyday language
Users ask about tables, columns and relationships in plain words and get a direct answer, including the SQL for a view when they need it.
Platform
AI
AI, governed like everything else.
SchemaVortex has two AI features, and both follow the rules its users follow. AI Chat answers questions inside the product from metadata only, on your own Azure OpenAI. The AI Assistant connects a coding agent on a user's computer to the platform, with that user's own permissions. The agent never sees more than the user, and everything it does is logged.
AI Chat
AI Chat answers questions about your schema.
A chat assistant for your data model. Users ask which tables exist, how they connect, where a field comes from, and how to query it in SQL.
Runs on your own Azure OpenAI
The models run in your own Azure subscription. Nothing is sent to an outside service, and Microsoft does not train its models on what is sent to them.
Metadata only
AI Chat gets schema and table names, column names and types, keys, notes, tags, view SQL and view history. It cannot read the data itself.
The Vault, the Mart and the Sandbox
AI Chat works on the Vault, the Mart and the Sandbox views. It does not see your Origins or Source tables.
AI Assistant
A coding agent with the user's permissions.
A coding agent on the user's own computer, such as Claude Code, Codex, Gemini CLI or GitHub Copilot, connects to SchemaVortex with a personal key. It works with the permissions that user has.
SchemaVortex
Catalog and proposals
- The agent reads the catalog as a set of files: Source tables, Vault tables, Mart views, Sandbox views, relationships, lineage, view SQL and view history. Forbidden columns are marked.
- The agent can write Mart Plans, Vault drafts, Sandbox drafts and developer notes, within the user's permissions. It cannot publish or approve anything. A person does that in SchemaVortex.
- Every action is recorded in the AI Assistant log under the user's name: keys, logins, plans, drafts, notes and samples.
The user's computer
Coding agent and personal key
- Any coding agent that can read a file and call a web API: Claude Code, Codex, Gemini CLI, GitHub Copilot.
- A personal key identifies the user to the platform. It is valid for one week, and it stops working when an administrator revokes the user's sessions.
Your data
A temporary, limited login
- The login carries the user's own database access. Masked columns and columns hidden from AI come back empty.
- It is valid for one hour. Its database sessions are listed under the user's name.
If the deployment allows it, the agent can also ask for a profile of a Source table: row counts, value ranges, the most common values and detected patterns, per column. The profile goes through the same governance gates as everything else, and masked columns and columns hidden from AI are left out.
AI data access
The agent sees at most what the user sees.
If your deployment turns on data access, an agent can read data through a temporary login. The login has the user's own access and nothing more. A Data Steward can also hide single columns from it.
Narrower than the user's access
Every masked column comes back empty for the agent, even where the user may read it.
AI Gate
A Data Steward can hide a column from AI. It comes back empty in every AI session. People keep reading it as before.
Valid for one hour
After one hour the login expires. If the user's permissions change, it is revoked automatically. Administrators can revoke it at any time.
See governed AI in action.
A short live demo: ask AI Chat about the schema, then watch a coding agent propose a Mart Plan and a person publish it.