🔌 wiretaps

See what your AI agents are sending to LLMs.

A transparent MitM proxy for auditing AI agent traffic. Logs every prompt, response, and tool call. Auto-detects PII, credentials, and crypto wallet addresses.

---

Why?

AI agents have access to your emails, files, and credentials. But do you know what they're actually sending to OpenAI, Anthropic, or other LLM APIs?

wiretaps sits between your agent and the LLM, giving you complete visibility:

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│  AI Agent   │────▶│   wiretaps   │────▶│   LLM API   │
│ (any agent) │     │              │     │ (OpenAI,..) │
└─────────────┘     └──────┬───────┘     └─────────────┘
                          │
                          ▼
                   📋 Audit Logs
                   🚨 PII Alerts
                   📊 Dashboard

Live Dashboard

Features

• 🔍 Full Visibility — Log every prompt, response, and tool call
• 🚨 PII Detection — Auto-detect emails, phone numbers, SSNs, credit cards
• 🛡️ Redact Mode — Mask PII before it reaches the LLM
• ✅ Allowlist — Let specific values pass through (your email, company domain)
• ₿ Crypto Detection — Flag wallet addresses, private keys, seed phrases
• 📊 Live Dashboard — Terminal UI to monitor traffic in real-time
• 🔌 Zero Code Changes — Just set OPENAI_BASE_URL and go
• 🏠 Self-Hosted — Your data never leaves your machine
• 📦 SQLite Default — Zero dependencies, instant setup

Quick Start

# Install
pip install wiretaps

# Start the proxy
wiretaps start

# Point your agent to the proxy
export OPENAI_BASE_URL=http://localhost:8080/v1

# Run your agent as usual
python my_agent.py

# View logs
wiretaps logs

Dashboard

wiretaps dashboard

┌─────────────────────────────────────────────────────────────┐
│ wiretaps v0.1.0                          Requests: 1,234    │
├─────────────────────────────────────────────────────────────┤
│ 14:23:01 │ POST /chat/completions │ 1,200 tk │ ⚠️ PII: email │
│ 14:22:58 │ POST /chat/completions │ 856 tk   │ ✓ clean      │
│ 14:22:45 │ POST /chat/completions │ 2,100 tk │ ⚠️ PII: phone │
│ 14:22:30 │ POST /embeddings       │ 128 tk   │ ✓ clean      │
└─────────────────────────────────────────────────────────────┘

PII Detection

wiretaps automatically scans for sensitive data:

Pattern	Example
Email	user@example.com
Phone	+1 (555) 123-4567
SSN	123-45-6789
Credit Card	4111-1111-1111-1111
CPF (Brazil)	123.456.789-00
BTC Address	bc1qxy2kgdygjrsqtzq2n0yrf2493p83kkfjhx0wlh
ETH Address	0x71C7656EC7ab88b098defB751B7401B5f6d8976F
Private Key	0x... (64 hex chars)
Seed Phrase	12/24 BIP-39 words

Custom PII Patterns

Add your own patterns to detect company-specific sensitive data:

# Add a custom pattern
wiretaps patterns add --name "internal_id" --regex "INT-[0-9]{6}" --severity high

# Add another pattern
wiretaps patterns add -n "employee_id" -r "EMP[A-Z]{2}[0-9]{4}" -s critical

# List all custom patterns
wiretaps patterns list

# Remove a pattern
wiretaps patterns remove --name "internal_id"

Custom patterns are saved in ~/.wiretaps/config.yaml:

pii:
  custom:
    - name: internal_id
      regex: "INT-[0-9]{6}"
      severity: high
    - name: employee_id
      regex: "EMP[A-Z]{2}[0-9]{4}"
      severity: critical

Redact Mode

Automatically mask PII before it reaches the LLM:

wiretaps start --redact

Original:  "My email is john@example.com and SSN is 123-45-6789"
Redacted:  "My email is [EMAIL] and SSN is [US_SSN]"

The LLM never sees the real data. Your audit logs still capture the original for compliance.

Block Mode

Completely block requests that contain PII:

wiretaps start --block

When PII is detected, the request is rejected with HTTP 400:

{
  "error": "Request blocked: PII detected",
  "pii_types": ["email", "phone"]
}

Your agent receives an error, and the sensitive data never leaves your network. Perfect for strict compliance environments.

Export Logs

Export your audit logs to JSON or CSV for compliance reports:

# Export all logs to JSON
wiretaps export -f json -o logs.json

# Export to CSV
wiretaps export -f csv -o logs.csv

# Export only entries with PII detected
wiretaps export -f json -o pii-incidents.json --pii-only

# Export with date range
wiretaps export -f csv -o january.csv --since 2024-01-01 --until 2024-01-31

# Limit number of entries
wiretaps export -f json -o recent.json -n 100

Allowlist

Allow specific values to pass through without being flagged as PII:

# Allow your own email
wiretaps allowlist add -t email -v "me@mycompany.com"

# Allow all emails from your company domain
wiretaps allowlist add -t email -p ".*@mycompany\.com"

# Allow a specific phone number
wiretaps allowlist add -t phone -v "+5511999999999"

# Allow all phone numbers (use with caution)
wiretaps allowlist add -t phone

# List all rules
wiretaps allowlist list

# Remove a rule
wiretaps allowlist remove -t email -v "me@mycompany.com"

# Clear all rules
wiretaps allowlist clear

Or configure in ~/.wiretaps/config.yaml:

pii:
  allowlist:
    # Exact value
    - type: email
      value: "ceo@company.com"
    
    # Regex pattern
    - type: email
      pattern: ".*@company\\.com"
    
    # Allow entire type
    - type: phone

Usage Statistics

View usage statistics and PII detection metrics:

# Show overall stats
wiretaps stats

# Output as JSON (for scripts)
wiretaps stats --json

# Stats grouped by day
wiretaps stats --by-day

# Stats grouped by hour
wiretaps stats --by-hour

# Filter by API key (multi-tenant)
wiretaps stats --api-key sk-xxx

Example output:

📊 wiretaps Statistics

  Total Requests: 1,234
  Total Tokens: 456,789

  PII Detections: 23 (1.86%)
  Blocked: 5
  Redacted: 18
  Errors: 0

Top PII Types:
  - email: 15
  - phone: 8
  - us_ssn: 3

REST API

Serve logs and stats via a REST API for dashboards and integrations:

# Start the API server
wiretaps api start --port 8081

Endpoints:

Endpoint	Description
GET /health	Health check
GET /logs	List logs with pagination
GET /logs/:id	Get log details
GET /stats	Usage statistics

Examples:

# Health check
curl http://localhost:8081/health

# List logs (with pagination)
curl "http://localhost:8081/logs?limit=10&offset=0"

# Filter logs with PII only
curl "http://localhost:8081/logs?pii_only=true"

# Get specific log
curl http://localhost:8081/logs/123

# Get stats
curl http://localhost:8081/stats

# Get stats by day
curl "http://localhost:8081/stats?by_day=true"

Webhook Alerts

Get notified when PII is detected:

# ~/.wiretaps/config.yaml
alerts:
  webhook: https://hooks.slack.com/services/xxx/yyy/zzz
  on:
    - pii_detected
    - blocked

When PII is detected, wiretaps sends a POST request:

{
  "timestamp": "2024-01-15T10:30:00.000000",
  "endpoint": "/v1/chat/completions",
  "pii_types": ["email", "phone"],
  "redacted": true,
  "blocked": false
}

Works with Slack, Discord, custom webhooks, or any HTTP endpoint.

Multi-tenant Support

Track usage across multiple API keys (multi-tenant environments):

# View logs filtered by API key
wiretaps logs --api-key sk-xxx

# View stats filtered by API key
wiretaps stats --api-key sk-xxx

# API also supports filtering
curl "http://localhost:8081/logs?api_key=sk-xxx"

The dashboard shows masked API keys for each request (e.g., sk-t...cdef).

API keys are automatically extracted from the Authorization: Bearer <key> header.

Supported LLM APIs

• ✅ OpenAI (api.openai.com)
• ✅ Anthropic (api.anthropic.com)
• ✅ Azure OpenAI
• ✅ Google AI (Gemini)
• ✅ Local models (Ollama, vLLM, etc.)
• ✅ Any OpenAI-compatible API

Configuration

# Generate default config
wiretaps init

# Edit config
vim ~/.wiretaps/config.yaml

# ~/.wiretaps/config.yaml
proxy:
  host: 127.0.0.1
  port: 8080

storage:
  type: sqlite  # or postgresql
  path: ~/.wiretaps/logs.db

pii:
  enabled: true
  patterns:
    - email
    - phone
    - credit_card
    - crypto_address
  
  # Auto-redact before sending to LLM (enterprise)
  redact: false

alerts:
  # Webhook on PII detection
  webhook: https://your-server.com/alerts

Use Cases

Compliance & Audit

Keep audit logs of all AI interactions for GDPR, LGPD, SOC2 compliance.

Security Review

Ensure your agents aren't leaking credentials, API keys, or sensitive data.

Crypto/Fintech

Detect wallet addresses and private keys before they reach external APIs.

Development

Debug exactly what your agent is sending without modifying code.

Integrations

Clawdbot

# clawdbot.yaml
llm:
  baseUrl: http://localhost:8080/v1

LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="http://localhost:8080/v1"
)

Any OpenAI SDK

import openai

client = openai.OpenAI(
    base_url="http://localhost:8080/v1"
)

Enterprise

Self-hosted open source not enough? wiretaps Enterprise adds:

• 🌐 Web dashboard with auth
• 🔐 SSO (SAML, OIDC)
• 👥 RBAC (role-based access)
• 📊 Compliance reports (LGPD, GDPR, SOC2)
• 🔔 Alerts (Slack, email, webhook)
• ☁️ Managed hosting option

Contact: hello@wiretaps.ai

Contributing

Contributions welcome! See CONTRIBUTING.md.

# Clone
git clone https://github.com/marcosgabbardo/wiretaps
cd wiretaps

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linter
ruff check .

License

MIT — use it however you want.

---

Built with 🔌 by @marcosgabbardo

wiretaps.ai