PII redaction for Python LLM apps
Protect provider-neutral prompt payloads, tool arguments, results, memory, logs, and async web applications.
Read the guide →MCP PII redaction
Place a local redaction boundary around MCP tool calls, tool results, and resource content without changing protocol shape.
Read the guide →pyveil vs Presidio, NER, and DLP
An honest decision guide: what each tool finds, its operational cost, and when the right answer is to layer them.
Compare approaches →Reproducible detector evaluation
Inspect the public synthetic corpus, exact pass criteria, supported-shape results, and the claims it deliberately does not make.
Inspect the evidence →What pyveil is optimized for
pyveil is a small, local boundary filter for high-confidence PII, credentials, known values, and domain identifiers. It does not pretend to discover every unknown person, address, or organization. When broad semantic recall is required, layer a dedicated NER system upstream and keep pyveil at the final agent boundary.