Honeyverse 🔐🍯🤖

An LLM powered honeypot simulating virtually any system or platform, and designed to trap attackers inside realistic deception scenarios.

How It Works

SCENARIO.md  ──▶  LLM (system prompt)
                       │
attacker types command  ──▶  LLM generates terminal output  ──▶  streamed back line by line
                       │
                  session log (.jsonl)

1. You write SCENARIO.md describing any system you want to simulate
2. The honeypot starts a service (e.g. SSH)
3. When an attacker connects, the LLM validates their credentials against the scenario
4. Once in, every command they type is sent to the LLM with the full conversation history
5. The LLM responds as the real system would — maintaining filesystem state, tracking the working directory, generating realistic file contents
6. Everything is logged: credentials, commands, outputs

The conversation history is the state. The LLM stays consistent because it remembers everything it already said.

SSH Honeypot

ssh/
├── SCENARIO.md          # Describe your target system here
├── main.go              # Entry point
└── internal/
    ├── scenario/        # Loads SCENARIO.md
    ├── llm/             # Provider interface + Anthropic & Ollama backends
    ├── server/          # gliderlabs/ssh server, host key management
    ├── shell/           # PTY session: echo, backspace, line-buffered streaming
    └── logger/          # Per-session JSONL logs

Quick Start

cd ssh
export ANTHROPIC_API_KEY=sk-ant-...
go run . --scenario SCENARIO.md --port 2222

# In another terminal — attack your own honeypot
ssh admin@localhost -p 2222
# password: admin123 (as defined in SCENARIO.md)

Flags

Flag	Default	Description
--scenario	SCENARIO.md	Path to scenario file
--port	2222	SSH listen port
--log-dir	sessions/	Directory for session logs
--host-key	host_key	Path to persist the RSA host key
--provider	anthropic	LLM backend: anthropic or ollama
--api-key	$ANTHROPIC_API_KEY	Anthropic API key
--ollama-url	http://localhost:11434	Ollama base URL
--ollama-model	qwen2.5:0.5b	Ollama model name

Writing a Scenario

SCENARIO.md is free-form markdown. Claude reads it as the system prompt. Describe whatever you want:

# prod-db-01 — PostgreSQL Database Server

## System Identity
- Hostname: prod-db-01
- OS: Debian 11 (Bullseye)
- Uptime: ~120 days

## Users & Credentials
- postgres / postgres (weak default, should have been changed)
- backup / backup2023

## Installed Software
- PostgreSQL 14 running on port 5432
- ...

## Sensitive Files
- /etc/postgresql/14/main/pg_hba.conf — auth config
- /home/postgres/.pgpass — plaintext credentials
- ...

## Vulnerabilities
- Default postgres password never changed
- ...

The more detail you provide, the more convincing the simulation. Planted secrets, realistic bash history, misconfigured services — the LLM will honour all of it.

Session Logs

Every session is logged to sessions/<id>.jsonl, one JSON object per line:

{"session_id":"20240115-143022-a1b2c3d4","timestamp":"...","type":"auth_attempt","remote_ip":"203.0.113.45","username":"root","password":"toor"}
{"session_id":"20240115-143022-a1b2c3d4","timestamp":"...","type":"auth_accept","username":"root","password":"toor"}
{"session_id":"20240115-143022-a1b2c3d4","timestamp":"...","type":"command","username":"root","command":"cat /etc/shadow"}
{"session_id":"20240115-143022-a1b2c3d4","timestamp":"...","type":"output","username":"root","command":"cat /etc/shadow","output":"root:$6$..."}

Event types: connect, auth_attempt, auth_accept, auth_reject, command, output, disconnect

Roadmap

• Wrong-password delay — slow down brute-force attempts
• Trap commands — deeper engagement when attacker runs sensitive commands
• Session replay — replay captured sessions from logs
• Webhook alerts — Slack/Discord notification on new session
• Multiple scenarios — run several honeypots on different ports simultaneously
• HTTP honeypot — fake web admin panel (WordPress, phpMyAdmin, etc.)
• DNS honeypot
• YAML scenario format with stricter structure

LLM Providers

Provider	Flag	Notes
Anthropic Claude	--provider anthropic	Best quality. Requires API key.
Ollama (local)	--provider ollama	Fully offline. Recommend 7B+ models.

# Anthropic (default)
go run . --provider anthropic

# Ollama — local, no API key needed
go run . --provider ollama --ollama-model llama3.1:8b
go run . --provider ollama --ollama-model qwen2.5:7b

Note: Models below 7B parameters typically cannot follow strict terminal-emulation instructions reliably.

Requirements

• Go 1.22+
• Anthropic API key (for Anthropic provider) or Ollama running locally
• For port 22 redirect: iptables -t nat -A PREROUTING -p tcp --dport 22 -j REDIRECT --to-port 2222