Data Privacy

Summary

• Purpose: Clarify what ProofChat sends to AI providers and how to control it
• Audience: Implementers, security reviewers
• Prereqs: See /docs

What data is sent

• Database schema: The schema as configured via DDL extraction and schema labeling, limited to the portions you include in configuration
• User prompts: Anything typed into the chat input
• Tool-shared data: Outputs from tools that you explicitly enable (e.g., summarized query results, component payloads) when necessary to refine SQL queries, searches, or otherwise improve the answer

FileMaker Script Steps and Data Transmission

FileMaker script steps that communicate with AI models may send data back to refine queries and improve responses. While it's technically possible to restrict this, it's practically challenging. Unless you're very familiar with how FileMaker communicates with large language models through its Generate Response for Model, Perform SQL Query with Natural Language, and Perform Find with Natural Language script steps, you should assume that some data will be sent to the models you have configured.

If you'd like our help analyzing the specific configurations and tools you create with ProofChat, we can provide that service. See the support page for contact information.

Ways to reduce exposure

• Minimize context: Label only the schema and fields required; avoid sensitive fields unless strictly needed
• Review your FileMaker script configurations carefully if data sensitivity is a concern

Provider configuration

• Prefer providers/models that meet your compliance needs
• Private Inference: We can provide completely private infrastructure where you control the models and data never leaves servers under your control. Note that private inference is more expensive than using providers like OpenAI. Contact us via the support page to learn more.

Related

• Configuration → AI Accounts and Models
• Configuration → Schema Labeling
• Reference → /docs/reference/settings/ai-accounts