To Run Locally

Prerequisites: Node.js (for frontend) and Python 3.10+ (for the optional local API)

Frontend (web UI)

1. Install dependencies: npm install
2. Set the GEMINI_API_KEY in .env.local to your Gemini API key (if using Gemini features)
3. Run the app: npm run dev

Backend (optional Python FastAPI service)

1. Create and activate a Python virtual environment:

   python -m venv .venv
   source .venv/bin/activate

2. Install the (minimal) dependencies:

   pip install fastapi uvicorn pydantic python-multipart requests

3. Run the API server (module path):

   uvicorn src.api.redact_api_main:app --reload --host 0.0.0.0 --port 8000

Usage Example

Quick local workflow:

• Start the frontend: npm run dev (open http://localhost:5173)
• Optionally start the Python API: uvicorn api.redact_api_main:app --reload --app-dir src --port 8000
• Upload a PDF in the UI and download the sanitized/exported artifacts.

Python API (FastAPI) — Endpoints & Examples

The local API entrypoint is src/api/redact_api_main.py. Notable endpoints (all prefixed with /api):

• POST /api/upload/pdf — Accepts multipart/form-data file (PDF). Returns extracted text as a downloadable pdf_data.txt.
• POST /api/download/report — Accepts a plain text file and returns a generated summary as summary.txt.
• POST /api/sanitize — Accepts a plain text file and returns a sanitized text file as sanitized.txt.
• GET /api/connection — Checks Ollama connection (returns JSON connection status).
• GET /api/download/dictionary — Returns a small JSON dictionary file.
• GET /api/download/risk-report — Returns a JSON risk-report stub.

Curl examples (replace sample.pdf / sample.txt with your file):

# Upload a PDF and save extracted text
curl -F "file=@sample.pdf" http://localhost:8000/api/upload/pdf -o pdf_data.txt

# Upload a plain-text file to get a summary
curl -F "file=@sample.txt" http://localhost:8000/api/download/report -o summary.txt

# Sanitize a plain-text file
curl -F "file=@sample.txt" http://localhost:8000/api/sanitize -o sanitized.txt

# Health check for Ollama connection
curl http://localhost:8000/api/connection

Overview of the Redact AI's Working

SecureRedact Protocol is a multi-stage, user-controlled data anonymization framework designed to ensure sensitive information is handled with maximum transparency, privacy, and regulatory compliance. All critical screening and redaction steps occur locally before any optional cloud interaction.

---

Multi-Stage Trust Architecture

A transparent process where you control exactly how your sensitive data is handled.

Core Principles

• Zero-Knowledge Screening
• Local Redaction
• Optional Cloud Audit

Stage 1: Interactive Screening

Upon upload, the Privacy Companion performs a local scan that:

• Identifies document type
• Detects potential PII patterns (names, SSNs, identifiers)
• Operates entirely within the browser

No data leaves your device during this phase.

---

Stage 2: Local Redaction

Redaction is performed locally using Gemma-2-mini via Ollama.

• Sensitive values are replaced with deterministic placeholders such as:

[REDACTED_NAME_1]

• Original data never leaves the local environment
• Redaction logic is repeatable and auditable

---

Stage 3: Hybrid Verification

After local sanitization is complete, users may optionally send only anonymized content to Gemini-3-Pro for:

• Professional summarization
• Privacy leak detection
• Secondary risk analysis

Cloud interaction is strictly opt-in.

---

Zero-Knowledge Screening Engine

Conversational Intelligence

The Privacy Companion operates as a contextual, user-guided system:

• Automated detection of applicable privacy jurisdictions (GDPR, HIPAA, DPDP)
• Context-aware redaction (distinguishing headers, body text, identifiers)
• User-in-the-loop controls to adjust jurisdiction and redaction logic before processing

---

Processing Lifecycle

1. INITIAL_SCAN(local_buffer)
   → COMPANION_CHAT(jurisdiction_match)
2. OLLAMA_REDACTION(deterministic_masking)
3. CLOUD_CONSENT_CHECK()
   → GEMINI_AUDIT(anonymized_view)

---

Governance & Auditability

Secure Export Options

Each session produces two independent artifacts:

CLEAN DATA
Sanitized document suitable for sharing or storage in low-security environments.

AUDIT LOG
A local JSON key enabling reversible redaction, allowing original values to be restored locally when authorized.

---

Regulatory Compliance

By enforcing on-device risk assessment and redaction, SecureRedact supports strict data sovereignty and air-gapped compliance requirements.

Supported frameworks include:

• GDPR
• HIPAA
• DPDP
• PIPEDA
• CCPA
• LGPD

---

Security Model Acknowledgement

The Redact AI is designed to ensure:

• User accountability of sharing sensitive data
• Deterministic and auditable anonymization
• Explicit consent for any cloud-based processing

Understanding this model is required before proceeding with document processing.