What PromptBrake Does

PromptBrake automatically tests your LLM API endpoint for the most common release‑blocking failures. It runs a fixed suite of over 60 real‑world attack prompts against the live endpoint you ship and returns evidence‑backed findings in minutes.

Core Security Checks (13)

• Prompt Injection – direct, indirect, nested encoding, adversarial suffixes
• Data Exposure – system prompt leaks, cross‑user leakage, sensitive echo tests
• Tool Abuse – unauthorized function calls, fake tool transcripts, parameter tampering
• Output Controls – schema escape, output sanitization bypass, structured response failures
• Memory Leak – extraction of conversation history
• Context/History Checks – multi‑turn escalation, long‑context refusal decay

How It Works

1. Connect: Enter the API URL and credentials for the route your application already uses.
2. Run the Scan: PromptBrake sends a fixed attack suite to the live endpoint.
3. Fix with Evidence: Review PASS, WARN, FAIL outcomes, use Fix Advisor guidance, then re‑run after deploying fixes.

Why Teams Choose PromptBrake

• No SDK rewrite or source code required.
• Fast scans (3–8 min) that fit into CI/CD pipelines.
• Evidence‑backed findings with remediation guidance.
• Replay packs of recent real‑world attacks for continuous testing.

Pricing

PromptBrake offers a free endpoint scan with no credit card required. No paid tiers are listed on the site.