PISIGuard: Secret-Safe AI Chats

AI chat tools are now part of daily operations work. Engineers paste stack traces, logs, deploy notes, screenshots, customer symptoms, and shell snippets into assistants to move faster. That can help during an incident, but it also creates a new data path for secrets and personal information.

PISIGuard is trending on Hacker News because it solves a very practical problem: masking sensitive values locally before they are sent to AI chat platforms.

What Is PISIGuard?

PISIGuard stands for Personally Identifiable and Sensitive Information Guard. It is a browser extension that detects common sensitive data, replaces it with placeholders before submission, then restores the original values in the AI response.

The project README lists support for names, email addresses, phone numbers, credit card numbers, passwords, API keys, and similar values. The important design choice is locality. Detection, masking, and restoration happen in the browser. The extension does not route prompts through another server.

For operators, that matters. A local extension is easier to reason about when the goal is to reduce exposure.

Key Features

• Local masking: sensitive values are replaced before the AI platform receives the message.
• Reply restoration: placeholders are swapped back in the model response so the conversation still reads naturally.
• Common chat support: the README mentions ChatGPT, Claude, and DeepSeek.
• No telemetry path: processing happens on device, with no analytics or server calls.
• Custom rules: advanced users can add detection rules for local needs.

Installation

Firefox users can install PISIGuard from Mozilla Add-ons. You can also build it from source:

# Build from source
git clone https://github.com/mohamed--abdel-maksoud/PISIGuard.git
cd PISIGuard
pnpm install
pnpm test
pnpm run build

Then open chrome://extensions, enable Developer Mode, choose Load unpacked, and select the generated extension folder under dist. Firefox users can load a temporary add-on from about:debugging.

SRE Workflow

Use PISIGuard as a guardrail for browser-based AI analysis. It is most useful when engineers ask for help with:

• alert descriptions that include customer names or tenant IDs
• log excerpts that contain emails, tokens, or session identifiers
• support escalations with personal information
• incident timelines copied from chat or ticketing tools
• shell snippets that accidentally include credentials

Do not treat it as a compliance boundary by itself. Teams still need log scrubbing, secret scanning, least-privilege tokens, and an approved AI usage policy.

Operational Tips

Test masking rules with real internal patterns before rollout. Tenant IDs, trace IDs, and cloud resource names often look harmless to generic detectors but still carry operational sensitivity.

Keep incident prompts short and scoped. Redaction works best when engineers paste only the data needed for the current question.

For production incidents, prefer read-only AI workflows first. Let the assistant explain logs, compare deploys, draft hypotheses, and prepare commands. Keep execution behind human approval.

Conclusion

PISIGuard is a small tool with a serious operational lesson: AI assistance should not require leaking raw incident context by default.

For SRE teams adopting AI in production workflows, this kind of guardrail is becoming table stakes.

Looking to automate infrastructure operations with strong context and approval flows? Akmatori helps SRE teams build AI-assisted incident response with production guardrails. For reliable global infrastructure, check out Gcore.