Getting Started¶
Installation¶
PyPI¶
With nmap integration¶
From source¶
Docker¶
docker build -t agentsniff .
# Web dashboard (host network for full visibility)
docker run -d --name agentsniff \
--network host \
--cap-add NET_RAW \
--cap-add NET_ADMIN \
agentsniff
# One-shot scan
docker run --rm --network host --cap-add NET_RAW \
agentsniff scan 192.168.1.0/24
Docker Compose¶
Requirements¶
- Python 3.11+
- Linux recommended (for
/proc/net/tcpanalysis) - Root/CAP_NET_RAW optional (enables passive DNS, TLS, and traffic monitoring)
Your First Scan¶
Scan a single host:
Scan a subnet:
Scan specific hosts:
Output Formats¶
# Table (default) — human-readable terminal output
agentsniff scan 192.168.1.0/24
# JSON — machine-readable
agentsniff scan 192.168.1.0/24 --format json --output results.json
# CSV — spreadsheet-friendly
agentsniff scan 192.168.1.0/24 --format csv --output results.csv
Web Dashboard¶
Start the dashboard and open it in your browser:
The dashboard provides real-time scan streaming, scan history, and settings management. See the Dashboard guide for details.
What Gets Detected¶
AgentSniff identifies:
- AI agent frameworks (LangChain, CrewAI, AutoGen, Dify, Flowise, n8n, and 50+ more)
- MCP (Model Context Protocol) servers
- LLM inference engines (Ollama, LM Studio, vLLM)
- Hosts querying LLM API domains (OpenAI, Anthropic, Google, Mistral, etc.)
- AgentPin identity documents
- Agent-like network traffic patterns
Each detection gets a confidence score from LOW to CONFIRMED, calculated using noisy-OR probability fusion across all detectors that produced signals for a given host.