SECURITY

Context Masking

A privacy technique that pseudonymizes sensitive entities in data before sending it to an external AI model for processing.

When BasaltHQ agents need to leverage powerful external LLMs for complex reasoning, Context Masking ensures no proprietary data leaks. Before a prompt leaves the BasaltHQ perimeter, an automated masking layer replaces all sensitive entities: company names become "Entity A," dollar amounts become "Value X," and employee names become "Person 1." The external LLM performs its reasoning on this sanitized data. When the response returns, BasaltHQ's reconstruction layer maps the pseudonyms back to the real values inside your secure environment. The external model never sees your actual data.