AegisPrompt OpenClaw Proxy

Privacy proxy that sits between OpenClaw and your LLM provider. Scans every prompt for sensitive data (PII, credentials, API keys, medical records) and redacts or blocks it before it leaves your machine.

Quick Start

# From the project root
cd packages/openclaw-proxy
npm install
npm start

The proxy starts on http://127.0.0.1:18790 by default.

Connect to OpenClaw

In your OpenClaw config (~/.openclaw/config.yaml or via the CLI), point your LLM provider through the proxy:

For Anthropic (Claude)

# Instead of pointing directly to api.anthropic.com:
providers:
  anthropic:
    baseUrl: http://127.0.0.1:18790/v1/messages

For OpenAI

providers:
  openai:
    baseUrl: http://127.0.0.1:18790/v1

That's it. OpenClaw sends requests to the proxy, the proxy scans them, and forwards clean requests to the real API.

Scanning Modes

| Mode | Behavior | |----------|---------------------------------------------------------| | audit | Logs detections but passes everything through | | warn | Logs + flags detections, passes through | | redact | Replaces sensitive data with [REDACTED] (default) | | block | Rejects requests containing critical sensitive data |

Set the mode via:

# CLI flag
npm start -- --mode block

# Environment variable
AEGIS_MODE=redact npm start

# Config file
# Edit aegisprompt.config.json (run `npm start init` to generate)

What It Detects

46 detection patterns across 6 categories:

• PII — Email, phone, SSN, addresses, passport, driver's license, UK NINO
• Credentials — AWS, GitHub, Stripe, OpenAI, Anthropic, database URLs, private keys, JWTs, env secrets, passwords, Bearer tokens, SendGrid, Twilio, Azure, GCP, OpenClaw gateway tokens
• Financial — Credit cards (Luhn validated), bank accounts, routing numbers, IBAN, SWIFT/BIC, EIN/TIN
• HIPAA — Medical record numbers, NPI, DEA (with checksum), health plan IDs
• Crypto — Bitcoin, Ethereum addresses, seed/mnemonic phrases
• System — Sensitive file paths (SSH keys, .env files, OpenClaw credentials)

Configuration

Generate a config file:

node src/index.js init

This creates aegisprompt.config.json:

{
  "proxy": {
    "host": "127.0.0.1",
    "port": 18790,
    "routes": {
      "/v1/messages": "https://api.anthropic.com",
      "/v1/": "https://api.openai.com"
    }
  },
  "scanning": {
    "mode": "redact",
    "enabledTiers": ["core", "tier1", "tier2"],
    "minSeverity": "medium",
    "scanResponses": true
  },
  "audit": {
    "enabled": true,
    "filePath": "./aegisprompt-audit.log"
  }
}

Industry Templates

For healthcare / HIPAA:

{
  "scanning": {
    "enabledCategories": ["pii", "hipaa", "credential"],
    "mode": "block",
    "blockSeverities": ["critical", "high"]
  }
}

For financial services:

{
  "scanning": {
    "enabledCategories": ["pii", "financial", "credential"],
    "mode": "redact"
  }
}

Audit Log

When enabled, every detection is written to aegisprompt-audit.log in JSON-lines format:

{"event":"detection","timestamp":"2026-02-16T14:30:00.000Z","method":"POST","url":"/v1/messages","mode":"redact","detections":[{"type":"ssn","label":"Social Security Numbers","severity":"critical","count":1}]}

View stats:

node src/index.js stats

CLI Options

--port <number>     Proxy port (default: 18790)
--host <string>     Bind address (default: 127.0.0.1)
--mode <mode>       Scanning mode: audit|warn|redact|block
--upstream <url>    Default upstream LLM URL
--verbose           Enable debug logging
--audit-off         Disable audit logging
--no-responses      Don't scan LLM responses

Environment Variables

AEGIS_PORT=18790
AEGIS_HOST=127.0.0.1
AEGIS_MODE=redact
AEGIS_UPSTREAM=https://api.openai.com
AEGIS_LOG_LEVEL=info
AEGIS_SCAN_RESPONSES=true