How do I keep secrets safe in AI coding tools?
Swap them for stand-ins on the local machine before the prompt is sent. The model sees realistic placeholders, and the real values stay on the local machine.
The short answer
Route the AI coding tool through Pretense. It swaps detected secrets and PII for realistic stand-ins on the local machine before each prompt is sent, so the model works on stand-ins and the mapping back to real values stays local.
How it works
Requests route through a local proxy
Point the AI coding tool's base URL at Pretense. Every request routed through it is checked on the local machine before it is sent.
Secrets and PII become stand-ins
Detected API keys, credentials, customer data, patient records and SSNs are swapped for realistic stand-ins with the same structure, so the model keeps the context it needs.
Code names are renamed in code blocks
Function, class and variable names inside code blocks are renamed on the way out and restored byte-exactly on the local machine when the reply comes back.
The mapping stays local
Secrets and PII stay as stand-ins in the reply. The mapping back to the real values stays on the local machine.
Why stand-ins beat redaction
Redaction deletes the sensitive parts, which strips the context the model relies on. Stand-ins keep the same shape, so the code still reads correctly and the model keeps the context it needs to give a useful answer.
Compliance status
Audit trails and policy presets are available today. No auditor has been engaged. No SOC 2 report or HIPAA certification has been issued.
Setup takes about 30 seconds. Read the quickstart or see the FAQ.