A drop-in proxy that strips PII from LLM requests before they leave your network. Change your API base URL and you're protected. Full streaming support, no code changes needed for OpenAI or Anthropic SDKs.
Anonamoose is a drop-in proxy for AI workloads that contain personal data. You point your existing OpenAI or Anthropic SDK at it instead of the public endpoint, and it strips identifying information from requests before they leave your network. Streaming responses work as usual; nothing about your code changes.
Detection runs in layers — dictionary rules for known names and IDs, local NER for surfaces it has not seen before, regex for structured patterns and a name detector for the awkward cases. The point is to fail closed: anything ambiguous is redacted before the request goes anywhere.
It is fully self-hosted with SQLite for local state. Nothing about the request leaves your environment beyond what you have explicitly allowed. Audit logs cover every request, redaction decision and downstream call so you can show a security team or an auditor exactly what was sent.
We map the deployment guidance to HIPAA Safe Harbor, GDPR, SOC 2 and ISO 27001 patterns. There is also a standalone /api/v1/redact endpoint if you want to use the redaction engine in your own pipelines or log scrubbing.
Multi-layer Detection
Dictionary rules, local NER, regex, and name detection.
Self-hosted
Runs locally with SQLite — no data leaves your environment.
Compliance Ready
HIPAA Safe Harbor, GDPR, SOC 2 and ISO 27001 mappings.
Redaction API
Standalone /api/v1/redact endpoint for pipelines and logs.