Scan your AI history for leaked secrets
One command. 15 seconds. Checks Claude, Cursor, Copilot, ChatGPT, and 35+ more AI tools for AWS keys, database passwords, API tokens, and private keys.
Then run citrusglaze-scan. Requires Python 3.8+. View source on GitHub
AI Tools Scanned
Claude Code, Claude Desktop, Cursor, GitHub Copilot, ChatGPT, Gemini, Windsurf, Cody, Continue, Aider, and 29+ more
Secrets Detected
AWS keys, GCP service accounts, Azure tokens, database URIs, private keys, API tokens, Slack/GitHub/GitLab tokens, JWTs, and 200+ more patterns
How it works
Install
pip install citrusglaze-scan
Scan
Finds AI tool chat histories on your machine and scans for 254+ secret patterns
Review
Get an HTML report with every secret found, severity, and which AI tool it was sent to
The average developer leaks 5 secrets per week to AI
We scanned real developer workstations. 13% of AI prompts contain sensitive data — AWS keys, database passwords, API tokens. Most developers never know.
100% Local
Zero network requests. Your data never leaves your machine. Verify in the source code.
Open Source
Read every line. No obfuscation, no telemetry, no hidden calls. MIT licensed.
No Signup Required
pip install and run. No account, no API key, no credit card. Results in 15 seconds.
What real scans find
These are real secret types found across developer workstations. Names and values redacted.
AWS Access Key pasted into Cursor chat
Developer asked AI to debug an IAM policy and included the key. Found in 3 separate conversations over 2 weeks.
PostgreSQL connection string with production credentials
Full URI including username, password, and hostname sent to Claude Code while debugging a migration.
GitHub personal access token in Copilot context
Token embedded in a .env file that was included as context when asking for CI/CD help.
Slack webhook URL shared with ChatGPT
Asking AI to help build a notification bot — webhook URL allows anyone to post to internal channels.
RSA private key included in SSH debugging session
Developer pasted their private key into an AI prompt asking why SSH connections were failing.
Not ready to scan? Get the report first.
See what we found scanning thousands of real AI prompts — secrets, patterns, and the tools that leak the most.
No spam. Unsubscribe anytime. Your data stays private.
Found secrets? Block them in real-time.
The scanner shows what's already leaked. The CitrusGlaze proxy catches secrets before they reach AI providers — every request, zero latency.
FAQ
Does it send my data anywhere? ▾
No. The scanner runs 100% locally. No network requests, no telemetry, no phone-home. Your data never leaves your machine. Check the source code yourself.
What AI tools does it support? ▾
39+ tools including Claude Code, Claude Desktop, Cursor, GitHub Copilot, ChatGPT Desktop, Gemini, Windsurf, Cody, Continue, Aider, pip, npm, and more.
What's the difference between the scanner and the proxy? ▾
The scanner checks your existing AI chat history for secrets that have already been sent. The proxy sits between your AI tools and the internet to block secrets in real-time before they're sent. Scan first, then install the proxy to prevent future leaks.
What Python version do I need? ▾
Python 3.8 or newer. Works on macOS, Linux, and Windows. No compiled dependencies — pure Python with zero external packages.
How do I read the results? ▾
The scanner generates an HTML report at ~/citrusglaze-report.html. Open it in your browser to see every secret found, its severity (Critical, High, Medium), which AI tool it was sent to, and the date.
Does it work on Windows and Linux? ▾
Yes. The scanner works on macOS, Linux, and Windows. It's pure Python with zero compiled dependencies. On each platform it automatically finds the correct AI tool data directories.
Ready to find out?
One command. 15 seconds. No account, no signup, no data leaves your machine.
Then run citrusglaze-scan. Results in your browser.