Faasera Documentation

🔁 Faasera LLM Privacy Proxy – End-to-End Workflow

This document explains how the Faasera Proxy intercepts, redacts, and routes LLM (Large Language Model) requests securely using Faasera Profiler.

---

1. 📨 Client Sends Request

A user or app sends a standard OpenAI-style request to the proxy instead of directly to OpenAI or another LLM provider.

POST http://localhost:8080/v1/chat/completions
Authorization: Bearer <API_KEY>
Content-Type: application/json

{
  "model": "gpt-4",
  "messages": [
    {"role": "user", "content": "What is Jane Doe's SSN 123-45-6789?"}
  ]
}

---

2. 🔒 Faasera Proxy Intercepts the Request

• The proxy extracts the messages array from the payload.
• It prepares to redact each message.

---

3. ✂️ Faasera Profiler Redacts Each Message

For each message:

POST /api/profile-mask/default
Content-Type: application/json
{
  "value": "What is Jane Doe's SSN 123-45-6789?"
}

The Faasera Profiler returns:

{
  "replacement": "What is [[NAME]]'s SSN [[SSN]]?"
}

---

4. 🔁 Proxy Rebuilds Redacted Payload

• Redacted content replaces the original message.content.
• Entire payload is now safe.

---

5. 🎯 Proxy Routes to Correct Target

Based on the environment config (DEFAULT_TARGET=openai|claude|gemini|local), the proxy determines where to send the request.

Example:

https://api.openai.com/v1/chat/completions

---

6. 🚀 Proxy Forwards Redacted Payload to LLM

• Uses client-supplied Authorization header.
• Sends cleaned messages to selected LLM endpoint.

---

7. 🧠 LLM Processes Request

Example LLM response:

{
  "id": "chatcmpl-xyz",
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "I'm sorry, but I cannot access personal information such as SSNs."
    }
  }]
}

---

8. 📤 Proxy Sends Response to Client

• Response is returned as-is from the LLM to the client.

---

🧭 Text-Based Diagram

┌────────────────────────┐
│  1. Client Request     │
└────────────┬───────────┘
             │
             ▼
┌────────────────────────┐
│  2. Proxy Receives     │
└────────────┬───────────┘
             │
             ▼
┌────────────────────────────┐
│  3. Call Faasera Profiler. │
│  for Redaction             │
└────────────┬───────────────┘
             │
             ▼
┌────────────────────────┐
│  4. Replace Message    │
└────────────┬───────────┘
             │
             ▼
┌──────────────────────────────────────────┐
│  5. Route to LLM (OpenAI, Claude, etc.)  │
└────────────┬─────────────────────────────┘
             │
             ▼
┌────────────────────────┐
│  6. Send to LLM        │
└────────────┬───────────┘
             │
             ▼
┌────────────────────────┐
│  7. Receive Response   │
└────────────┬───────────┘
             │
             ▼
┌────────────────────────┐
│  8. Return to Client   │
└────────────────────────┘

---

🔐 Security Highlights

• Redaction always occurs before reaching any LLM.
• Supports HTTPS (TLS), tokens, and masking policies.
• Customizable backend and provider switching.

---

🔗 Learn More

Visit https://www.faasera.ai for additional SDKs, plugins, and compliance features.