Skip to content
Entropy Data

Features using AI

Entropy Data offers a number of features that use AI. All AI features are optional and only active when an AI model is configured. See AI Configuration for supported providers and how to enable AI.

AI Models

AI features rely on a Large Language Model. You can choose between:

  • Managed (Cloud only) — Entropy Data hosts the model. AI features work out of the box.
  • Bring Your Own Model — Connect any OpenAI-API-compatible deployment, or an Anthropic API endpoint. You keep control of cost, region, and data residency.

Features

Entropy Intelligence

Conversational interface to discover, evaluate, request access to, and query data products in natural language. AI generates SQL that runs against output ports the user has access to, using the user's own credentials. See Chat with your data.

Search AI

Natural-language search across data products, data contracts, teams, and business definitions, so that searches match meaning, not just keywords.

Data Governance AI

Applies organization-defined, text-based governance policies at three points:

  • Resource validation — checks whether a data product, data contract, team, or tag complies with the policy.
  • Access request evaluation — checks whether an access request meets the policy and the terms of use of the data contract. See Access Approval Workflow.

In both cases the AI produces a recommendation; a human reviewer remains responsible for the final decision. Reviewers can flag false positives, and this feedback is taken into account on subsequent runs.

Query validation at execution time works differently — see Purpose-based Access Control below.

Purpose-based Access Control

Enforces the purpose a user stated when requesting access. When AI is enabled, a query submitted through Entropy Intelligence, the MCP server, or the access evaluation API is checked against the terms of use of the data contract and the purpose recorded in the user's access agreement. If the query does not fit, it is rejected before it runs and the user is told which agreement or term it conflicted with.

Unlike the other AI features, this check is enforcing rather than advisory: no human confirms the decision at query time. Three properties bound it:

  • It can only restrict, never widen. It runs after the regular access-grant check, so it never gives a user access they do not already hold.
  • It fails closed. If the model is unavailable, the query is denied rather than allowed.
  • It is optional, like every AI feature, and inactive unless AI is enabled for the organization.

See Entropy Intelligence Security for the full request flow.

Data Quality Check Generation

Generates executable SQL data quality checks from a natural-language description, turning text-based quality expectations into runnable code. The author reviews and edits the generated SQL before saving.

Data Contract Assistant

Inline assistant in the ODCS Data Contract Editor. Helps draft and refine data contracts and can generate custom code snippets (SQL queries, validation scripts, transformations) from the contract. Suggestions are inserted into the editor for the author to review before applying.

Semantic Linking Suggestions

Recommends semantic concepts for the properties of a data contract — either one property at a time, or for the whole contract at once via Suggest semantics for all properties. Suggestions are presented on a review screen before any link is saved. The whole-contract flow is experimental and off by default. See Semantics.

Data Contract Versioning

Recommends the next semantic version (major, minor, patch) by analyzing the differences between the previous and the current data contract. See Version Data Contracts.

External AI agents

Beyond features inside Entropy Data, AI agents can interact with the platform from the outside. The AI itself runs in the user's tooling, not in Entropy Data.

  • MCP — Claude, ChatGPT, and other MCP clients query the marketplace and, when permitted, execute SQL using the user's access rights.
  • Data Product Builder — Skills for AI coding agents (Claude Code, OpenAI Codex, GitHub Copilot CLI) that bootstrap data products against your organization's conventions.

Classification

Under the EU AI Act, these features are classified as limited risk. See EU AI Act.

Transparency

Entropy Data makes AI usage visible in three ways:

  • When AI is configured — Organization owners see and manage the active model under Organization Settings → AI. Each AI feature has its own toggle so it is clear what is enabled.
  • When the user interacts with AI — AI-driven surfaces (Entropy Intelligence chat, governance recommendations, generated SQL, version suggestions, contract assistant) are explicitly labeled as AI output and presented for review. Where a decision is enforced rather than suggested, as in Purpose-based Access Control, the user is shown the reason and the agreement or term it conflicted with.
  • Where AI is applied — Each AI-generated artifact shows what produced it: the data product used, the prompt or input, the policy that was evaluated, or the contract diff. Generated SQL and version recommendations remain editable before they take effect.