Skip to content
Intruva

Privacy & data

Your prompt is your data.

Prompts often contain source code, work details, personal data, and sometimes credentials. Intruva is designed so that, in this pilot, none of it leaves your device.

What's processed, and where

  • Only the text you type or paste into Intruva.
  • Demo mode (current): analysis, optimization and the security scan all run in your browser. No network request contains your prompt.
  • When the AI engine is connected: the indicator in the workspace header will say so. The security scan will still run on your device first, so you can redact secrets before anything is sent.

What's retained

  • Prompts are never stored: no history, no logs, no account.
  • A prompt you start on the home page is passed to the workspace through this tab's session storage, then deleted immediately.
  • The only thing saved on your device is your animation preference.
  • No analytics or tracking scripts are included in this pilot build.

What the security scan covers

It looks for:

  • API keys and tokens (OpenAI-style sk-, Anthropic, AWS, GitHub, Slack, Google, Stripe, JWT)
  • Private key blocks and credentials inside connection strings
  • Passwords and secrets assigned to variables
  • Card numbers (Luhn-checked), US SSN, Aadhaar, PAN, emails, phone numbers
  • Prompt-injection phrases: instruction overrides, system-prompt extraction, jailbreak personas
  • Instructions to send data elsewhere, unbounded agent permissions, destructive commands
  • Conflicting instructions (e.g. “brief” and “detailed”)

Its limits:

  • Pattern-based: novel secret formats and paraphrased injections can be missed.
  • Runs only on the text in the editor. Linked files and URLs are not fetched.
  • A clean result is not a guarantee of safety.

Masked evidence is shown in the UI (e.g. sk-p•••••••• (35 chars hidden)), so detected secrets aren't echoed back on screen.

Coming next

  • AI-assisted engine. Deeper diagnosis via a model provider, with explicit disclosure before any prompt leaves your device.
  • Team policies. Shared rules for what must never be sent to an AI tool.
  • The Chrome extension, developer API and MCP server follow the same rules: on-device by default, no prompt storage, explicit disclosure before anything is sent.