Skip to main content
Prophecy agents operate within your selected deployment environment and respects existing access controls. The agents are assistive rather than autonomous: they do not have autonomy to connect to external systems or perform any action without approval. Prophecy agents are considered limited risk according to the EU AI Act. This page explains how data is accessed, executed, and stored when using agents.

​
Key principles

  • Agents generate transformation logic and suggestions, but you review and approve all changes before pipelines are run in production.
  • Agents operate only within your Prophecy deployment and connected data platforms.
  • You can inspect all generated transformations in visual, SQL, or documentation form before execution.
  • You maintain full control over whether changes are applied.

​
Storage and environment

Agents run inside your Prophecy deployment environment. Prophecy stores:
  • Agent chat history.
  • Project metadata (pipelines, analyses, schemas, configurations).
  • Version-controlled project artifacts.
These items remain within your Prophecy deployment environment.

​
LLM interaction

When generating suggestions, agents may include project metadata in prompts sent to the configured language model. This metadata may include:
  • Table and column names
  • Schemas and data types
  • Transformation code within the current project
The agent does not include sample data by default. Using sample data can only be enabled by an administrator.

​
Execution model

Agents generate transformation logic within your Prophecy deployment environment. When validating transformations or running pipelines, execution occurs deterministically on your connected data platform and under the permissions of the user who invoked the agent. All reads and writes respect existing warehouse access controls, role-based permissions, and organizational governance policies. Agents do not independently access external systems or perform actions outside your configured data platforms.