AI Redact — Catch Secrets Before AI

Scans your code and intercepts AI chat prompts for PII, secrets, and credentials before they reach any AI model.

Developers accidentally paste API keys, credentials, and personal data into Copilot, ChatGPT, Claude, and other AI tools every day. AI Redact catches it before it leaves your machine.

Features

Real-Time Code Scanning

AI Redact scans your code as you type:

Yellow warnings for PII (emails, phone numbers, SSNs, credit cards)
Red errors for secrets (AWS keys, GitHub tokens, Stripe keys, SSH keys)
Quick-fix actions — click the lightbulb to redact any finding with one click
"Redact All" command — replace every finding in a file at once

AI Prompt Interception

AI Redact wraps your AI models with a protective proxy:

Open the model dropdown in VS Code Chat
Select "AI Redact: [Model] (Protected)"
Chat normally — every message is scanned before it reaches the AI
Sensitive data is warned, redacted, or blocked based on your settings

Three interception modes:

Mode	Behavior
warn (default)	Shows a warning banner, forwards the original prompt
redact	Replaces sensitive values with `[TYPE_REDACTED]` before forwarding
block	Stops the request entirely, shows what was detected

@redact Chat Participant

Use @redact in Copilot Chat for explicit prompt scanning:

@redact Explain this code: const key = "AKIA..."
@redact /scan Check this for secrets before I send it

14 Detection Patterns

Category	Detectors
PII	Email addresses, international phone numbers, Social Security Numbers, credit card numbers (Luhn validated)
Secrets	AWS access keys, AWS secret keys, GitHub PATs, Stripe API keys, Google API keys, Google OAuth secrets, SSH private keys
Tokens	JWT tokens, database connection strings (PostgreSQL, MySQL, MongoDB)
Generic	High-entropy strings (Shannon entropy > 4.5)

Configuration

Setting	Default	Description
`aiRedact.enabled`	`true`	Master on/off switch
`aiRedact.scanOnType`	`true`	Scan as you type
`aiRedact.minSeverity`	`"medium"`	Minimum severity: `low` / `medium` / `high` / `critical`
`aiRedact.enabledDetectors`	`[]` (all)	Specific detectors to enable
`aiRedact.secretSeverity`	`"error"`	Red squiggle for secrets
`aiRedact.piiSeverity`	`"warning"`	Yellow squiggle for PII
`aiRedact.interceptionMode`	`"warn"`	Proxy mode: `warn` / `redact` / `block`

Commands

AI Redact: Scan Current File — manually trigger a scan
AI Redact: Redact All Findings in Current File — replace all findings with placeholders
AI Redact: Toggle Scanner — enable/disable (also via status bar click)

Claude Code Integration

AI Redact includes a CLI tool that integrates with Claude Code's hook system. Add to .claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [{
      "type": "command",
      "command": "ai-redact scan --hook --stdin",
      "timeout": 5
    }]
  }
}

This blocks any prompt containing secrets before it reaches Claude.

Privacy

All processing happens locally. Zero data leaves your machine. No telemetry, no cloud calls, no data collection.

Open Source

Apache 2.0 License — free to use, modify, and distribute.

Built by Liora HQ