IndividualScan. Detect. Sanitize.
Real-time protection for pros & small teams. Guardrail scans what you type or upload, catches sensitive data and risky prompts, and sanitizes it in real time — before it reaches ChatGPT, Copilot, or any AI tool.
Scan, detect, sanitize — in the moment, on your device
No data leaves your computer to be checked. Every step happens locally, in real time.
Scan
As you type, paste, upload a file, or an AI agent acts, Guardrail scans it — prompts, clipboard, screen, local files, and agent tool calls alike.
Detect
Meaning-aware detection identifies sensitive data and risky intent by context — not just keyword matching — so it catches what simpler tools miss.
Sanitize
Sensitive values are masked, anonymized, or blocked before they reach the AI. Safe content passes straight through, untouched.
Scan → Detect → Sanitize — silently, in real time, with nothing uploaded to the cloud.
Where Guardrail steps in, in real time
The everyday moments where risky prompts, agent actions, and data leaks happen — and how Guardrail quietly stops them, right on your device.
Catch risky prompts
Hidden or manipulative instructions — in a message, a pasted document, or a line of code — are spotted and stopped before the AI ever acts on them.
- Flags jailbreaks & injected commands
- Stopped on-device, in real time
Control what AI agents do
Guardrail watches the hidden hand-offs between AI agents and blocks unsafe transfers or commands.
- Audits agent-to-agent activity
- Blocks unauthorized transfers & escalation
Stop commands that overreach
When an agent tries to grab system-level permissions it was never granted — or paste a customer list straight into a chatbot — Guardrail steps in before it happens.
- Flags privilege-escalation attempts
- Blocks unauthorized system commands
10 categories of risky intent Guardrail watches for
Examined across plain text, code, documents, and system commands.
Sanitized, without losing meaning
Names and amounts are detected and replaced with placeholders before your message reaches the AI model — the sentence stays just as useful.
“Scott owes Leo $1M due in 10 days”
“[Person 1] owes [Person 2] [Amount 1] due in 10 days”
What Guardrail protects, and where
47+ file formats and multiple languages — Guardrail understands context, not just keywords.
Data it protects
Where it works
No noticeable delay. Guardrail is optimized for instant performance using efficient AI models, so you keep working at full speed while it scans, detects, and sanitizes in the background.
Start free, upgrade when you need more
Coming soon: custom sensitive-entity categories and a redaction process certificate for every file.
Download Guardrail — free for a limited time
Pick your device and install. No credit card, no setup headache — protection starts the moment it's running.
Prefer the command line?
Power users on Windows can install with a single PowerShell command.
iwr -useb https://idox.ai/install-guardrail.ps1 | iexQuestions, answered
Stop AI data leaks before they happen
Join the pros and small teams keeping their data safe while they move fast with AI. Free for a limited time.
