Sanitize untrusted text before it enters an AI-agent workflow.
Retrieved pages, copied documents, emails, tickets, and tool output can contain hidden Unicode, unusual controls, credential-shaped strings, or suspicious instruction patterns. ScrubMyText provides deterministic preprocessing before model ingestion.
The problem
Raw context can contain zero-width characters, bidi controls, unusual Unicode, embedded secrets, or instruction-like text that deserves inspection before being placed into an agent context window.
One-call preprocessing
prepare_for_agent: inspect → scrub → scan → detect secrets → redact → agent-ready text
One prepare_for_agent invocation counts as one API call.
Available controls
Deterministic normalization
Remove selected hidden/control characters and normalize whitespace without paraphrasing or rewriting the source text.
Exact inspection
Report Unicode and control-character positions so software can see what is actually present.
Risk signals
scan_untrusted_text returns structural and heuristic signals. It is not a guarantee that prompt injection or malicious text has been detected.
Secret handling
Detect likely credential patterns and optionally redact common sensitive-data patterns before passing content downstream.
scrub_text and inspect_text can be called without an API key, up to the current public free allowance.Start with the free API
curl -X POST https://api.scrubmytext.com/v1/scrub -H "Content-Type: application/json" -d '{"text":"Hello world","mode":"standard"}'Common questions
Does ScrubMyText rewrite the user's wording?
The normalization tools are designed to make deterministic character/whitespace changes rather than rewrite prose.
Does the scanner guarantee prompt-injection detection?
No. It surfaces useful structural and heuristic signals; it should be one layer in a broader trust and authorization design.