Reviewed context in. Safe answers out.
Redactorr sanitises sensitive identifiers inside your browser runtime before anything reaches the model. Ask complex questions, draft legal briefs, and analyse documents without leaking raw customer PII to cloud providers.
POST /v1/chat/completions HTTP/1.1
Host: api.anthropic.com
Authorization: Bearer [REDACTED_API_KEY]
Content-Type: application/json
{
"model": "claude-3-5-sonnet-20241022",
"messages": [
{
"role": "user",
"content": "PATIENT CLINICAL INTAKE:\nPatient: [PERSON_1] (DOB: [DATE_1])\nMedicare No: [MEDICARE_1]\nPrescribing Clinician: Dr. [PERSON_2] (AHPRA: [AHPRA_1])\nClinical Assessment: Prescribed 50mg Sertraline daily following acute stress consultation at [LOCATION_1].\n\nSummarize clinical indications and potential contraindications."
}
]
}The 4-Stage AI Isolation Pipeline
How Redactorr enables safe generative AI workflows without exposing unredacted source material.
LOCAL PATTERN EXTRACTION
Browser-side AEGIS engine identifies names, Medicare, TFNs, and credentials before any network transmission.
TOKENISED DISPATCH
Document is converted to structured tokens ([PERSON_1], [MEDICARE_1]) and transmitted to the LLM.
ISOLATED MODEL REASONING
Claude, ChatGPT, or DeepSeek analyses syntax, relationships, and context with zero access to raw PII.
LOCAL RE-IDENTIFICATION
AI output is returned to the client and re-hydrated locally behind your corporate firewall.
Compatible with Any Foundation Model
Redactorr acts as a zero-trust governance layer between your sensitive data and any cloud or on-premise AI.
Anthropic Claude Models
Deep legal and financial analysis over extended context windows with full entity tokenisation.
OpenAI & ChatGPT Enterprise
High-speed synthesis and drafting without exposing sensitive customer records to model logs.
DeepSeek Reasoning Models
Complex mathematical and logical reasoning over fully sanitised engineering and clinical logs.
Google Gemini Models
Massive multi-document cross-referencing and multimodal ingestion with zero raw identity leakage.
Local & Private LLMs
On-premises model execution paired with client-side token masking for maximum isolation.
Internal APIs & Enterprise RAG
Automate high-volume redaction pipelines before sending to proprietary internal vector databases.
Enterprise Privacy Guarantees
No Foundation Model Training
- Model providers (Anthropic, OpenAI, Google) receive only tokenised strings, making it impossible to memorise raw PII.
- Zero retention agreements enforced across enterprise API endpoints.
- Australian Privacy Principles (APP 11) data security compliance.
Client-Side Map Governance
- Pseudonymization maps remain in local browser memory under client-held encryption keys.
- Re-identification occurs entirely on the client after the LLM completes inference.
- SHA-256 integrity logs prove exactly what context was shared with third-party LLMs.
Stop blocking AI adoption over privacy fears.
Test our browser-side pattern engine today, or deploy Redactorr Pro for your entire legal and analytics team.