API Reference

Chat Completions

The chat completions endpoint is the primary way to send LLM requests through NeuronEdge. It automatically protects PII before forwarding to your chosen provider.

POST/v1/{provider}/chat/completions

Create a chat completion with automatic PII protection.

Headers

Header	Type	Description
Authorization	stringrequired	Bearer token with NeuronEdge API key (ne_...)
X-Provider-API-Key	stringrequired	Your LLM provider API key
Content-Type	stringrequired	application/json
X-NeuronEdge-Policy	string	Policy ID to use for this request
X-NeuronEdge-Format	string	Redaction format: token (default), hash, synthetic
X-NeuronEdge-Mode	string	Detection mode: real-time, balanced, thorough

Parameters

Parameter	Type	Description
provider	stringrequired	LLM provider: openai, anthropic, google, azure, etc.

Request Body

json

{
  "model": "gpt-5.2",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "My name is John Smith and my email is john@example.com."
    }
  ],
  "temperature": 0.7,
  "max_tokens": 1000,
  "stream": false
}

Response

json

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1702000000,
  "model": "gpt-5.2",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello John! I see you've shared your email..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 25,
    "completion_tokens": 15,
    "total_tokens": 40
  }
}

How PII Protection Works

Request Received

Your API request arrives at NeuronEdge with PII in the messages.

Prompt Injection Scan

Inbound messages are scanned for prompt injection patterns targeting PII extraction (Professional+).

PII Detection

Dual-engine detection (regex + NER) identifies sensitive entities in <1ms.

Redaction

PII is replaced with tokens, hashes, or synthetic data based on your policy.

Forward to Provider

The redacted request is sent to OpenAI, Anthropic, etc.

Response Detection

Outbound LLM response is scanned for hallucinated or echoed PII using a sliding buffer (Professional+).

Restore & Return

Original PII is restored in the response before returning to you.

Redaction Formats

Control how PII is replaced using the X-NeuronEdge-Format header:

Token (Default)

Replaces PII with type-based placeholders. Simple and fast.

text

Input:  "Contact John Smith at john@example.com"
Output: "Contact [PERSON] at [EMAIL]"

Hash (Professional+)

Deterministic hashes for consistent replacement across requests.

text

Input:  "Contact John Smith at john@example.com"
Output: "Contact [HASH:a1b2c3d4] at [HASH:e5f6g7h8]"

Synthetic (Professional+)

Realistic fake data that maintains semantic meaning.

text

Input:  "Contact John Smith at john@example.com"
Output: "Contact Sarah Johnson at sarah.j@demo.org"

Tool Call Redaction

NewAll tiers

NeuronEdge automatically detects and redacts PII in tool/function call arguments and tool results during agentic AI workflows.

json

// Before redaction
{
  "tool_calls": [
    {
      "id": "call_abc123",
      "type": "function",
      "function": {
        "name": "get_patient_record",
        "arguments": "{\"patient_name\": \"John Smith\", \"ssn\": \"123-45-6789\"}"
      }
    }
  ]
}

// After redaction
{
  "tool_calls": [
    {
      "id": "call_abc123",
      "type": "function",
      "function": {
        "name": "get_patient_record",
        "arguments": "{\"patient_name\": \"[PERSON]\", \"ssn\": \"[SSN]\"}"
      }
    }
  ]
}

Tool results (role: "tool" and role: "function" messages) are automatically scanned on inbound requests. No configuration needed.

Response-Side PII Detection

NewProfessional+

LLMs can hallucinate or echo PII in their responses. NeuronEdge scans the outbound response stream using a sliding buffer approach to catch and redact hallucinated data.

json

{
  "response_redaction": {
    "enabled": true,
    "method": "regex",
    "action": "redact",
    "buffer_size": 256
  }
}

Example:

text

LLM generates: "Based on the records, John Smith's appointment is..."
Client receives: "Based on the records, [PERSON]'s appointment is..."

Prompt Injection Detection

NewProfessional+

Prompt injection attacks attempt to manipulate the LLM into extracting or leaking PII. NeuronEdge detects and blocks these attacks using three modes:

log (default) — Record detection, forward request
warn — Forward request with detection header
block — Return 400 error, don't forward to provider

json

{
  "prompt_injection_detection": {
    "enabled": true,
    "mode": "block",
    "sensitivity": "medium"
  }
}

Blocked request response:

json

{
  "error": {
    "code": "PROMPT_INJECTION_DETECTED",
    "message": "Request blocked: potential PII extraction attempt detected",
    "status": 400,
    "details": {
      "category": "extraction_command",
      "action": "block"
    }
  }
}

Streaming

NeuronEdge fully supports streaming responses. Set stream: true in your request body to receive Server-Sent Events (SSE).

json

{
  "model": "gpt-5.2",
  "messages": [{"role": "user", "content": "Hello"}],
  "stream": true
}

PII detection runs on each streamed token to ensure protection even in real-time responses.

Complete Examples

OpenAI

bash

curl -X POST https://api.neuronedge.ai/v1/openai/chat/completions \
  -H "Authorization: Bearer ne_live_your_api_key" \
  -H "X-Provider-API-Key: sk-your-openai-key" \
  -H "X-NeuronEdge-Format: token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.2",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "My SSN is 123-45-6789. Is that secure?"}
    ],
    "temperature": 0.7
  }'

Anthropic

bash

curl -X POST https://api.neuronedge.ai/v1/anthropic/messages \
  -H "Authorization: Bearer ne_live_your_api_key" \
  -H "X-Provider-API-Key: sk-ant-your-anthropic-key" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "My SSN is 123-45-6789. Is that secure?"}
    ]
  }'

Detection Metrics

Every response includes headers with detection metrics:

http

X-Request-ID: 01HXYZ123ABC456DEF
X-NeuronEdge-Entities-Detected: 1
X-NeuronEdge-Detection-Time-Ms: 0.42
X-NeuronEdge-Response-Entities-Detected: 0
X-NeuronEdge-Prompt-Injection-Score: 0.0
X-RateLimit-Remaining: 999