A desktop AI assistant that reads your files and answers your questions, and never lets your secrets leave your computer.
What the cloud model would see
A simplified version of our first detection layer, running only in your browser. Nothing you type here is sent anywhere.
Cloud assistants like ChatGPT and Claude are powerful, but whatever you paste into them leaves your machine: API keys, passwords, client names, private code. Companies pay for enterprise contracts to protect this data, yet keeping it safe is still their own responsibility.
Local models avoid the problem but are much weaker. Today you have to pick one or the other for a whole session.
PromptShield sits between you and the model. It decides, per question, where your question can go. Before anything reaches the cloud, it finds sensitive data and masks it, and it keeps a record of every decision.
A local model answers using your local files. Nothing leaves the computer. Best for private notes, sensitive code and personal documents.
A cloud model answers harder questions, but only after secrets are removed from the prompt. The cloud sees the question, not the secrets.
Press Alt+Space and type in plain language.
Relevant passages are found in your own files, on your disk.
Pattern rules, then a small local model, look for secrets and mask them.
Secrets found means it stays local. Otherwise hard questions may go to the cloud.
You get the answer, and every outgoing prompt is written to an audit log.
| Name | Student ID | |
|---|---|---|
| Leyla Abbasova | 22301172 | leyla.abbasova@ug.bilkent.edu.tr |
| Zaeem Masood Sheikh | 22201178 | zaeem.sheikh@ug.bilkent.edu.tr |
| Farrukh Mammadov | 22301186 | farrukh.mammadov@ug.bilkent.edu.tr |
| Shahin Ibrahimli | 22301574 | shahin.ibrahimli@ug.bilkent.edu.tr |
Supervisor
Cevdet Aykanat
Innovation expert
Alper Sarikan
Course instructor
Mert Bıçakcı
Course instructor
Hüseyin Acacak
All of the project's code and documents live in one GitHub repository.
Open the repository on GitHub