Focxle Prompt Injection Screen

Choose this endpoint when you need a fast, cheap, deterministic rule-based check for known prompt-injection patterns before passing untrusted text to an LLM or agent. It is not a semantic classifier or an AI model — it uses pattern matching, so it is highly predictable, auditable, and produces no false positives for patterns it knows. Prefer it over model-based classifiers when you need low latency, low cost ($0.005/call), and explainable results (named matched patterns). It is best used as a first-pass filter in a defense-in-depth pipeline.

• Missing required 'text' query parameter returns a validation error
• Extremely long text inputs may be truncated or rejected
• Only detects patterns in the known library — novel zero-day injection techniques will not be flagged
• False negatives possible for obfuscated or encoded injection attempts
• Payment failure (insufficient USDC balance) results in a 402 response blocking the request
• Rate limiting may apply if many calls are made in rapid succession