Privacy layer for AI

Protect the prompt.
Preserve the meaning.

Marai detects sensitive entities before your prompt reaches an AI provider—without flattening the context that makes it useful.

01

Original prompt

Live preview calls only this self-hosted Marai service—not an LLM provider
Try an exampleBrowse sample prompts
Sample promptsSelecting one replaces the prompt above

Loading examples…

Examples: nvidia/Nemotron-PII · CC BY 4.0

85 characters
02

Safe for the LLM

waiting
M

Your protected prompt appears here

Only detected entities change. Intent, grammar, and context remain intact.

Prompt→Detect→Tokenize→LLM provider

PII stays inside your trust boundary