● ACTIVE — built, deployed, field-tested on real hardware

Warlock OS

A self-contained field cyberdeck on a ClockworkPi uConsole (Raspberry Pi Compute Module 5). One platform — web UI, terminal UI, and an on-device AI operator — over a single API, behind a hard engagement-scope gate and a signed audit trail. Designed to run dark: offline, on battery, over LoRa mesh.

Warlock OS terminal dashboard running live on the deck Warlock OS shield emblem

01 // not "a Pi with Kali on it"

ONE PLATFORM

Three front-ends, one truth

A web UI, an Ink terminal UI, and an AI chat operator all speak to the same FastAPI backend. One source of truth for state, scope, and audit.

HARD GATE

Authorization-first

Offense is inert until an operator arms a scoped engagement. Scope is enforced at execution time; out-of-scope is refused + logged; a kill switch reaches every queue.

PROVABLE

Signed attestation

Every audit record is emitted as an Ed25519 / did:web attestation a third party can verify offline — proof of conduct without having to trust the deck.

02 // what it can do

Capabilities

17 backend modules across red, blue, radio, and platform domains — offensive operations are engagement-gated; defensive and recon tooling runs for own-network monitoring.

DomainCapability
Wireless reconAP/client discovery, channel survey, handshake & PMKID capture (aircrack-ng, hcxdumptool) on a MediaTek MT7961 in monitor mode
Wireless offensegated — deauth, evil-twin & karma (airbase-ng), WPS (reaver / bully)
Wireless IDSKismet-driven rogue-AP / evil-twin / deauth-flood detection → live alert pager
NetworkARP & port scanning, baseline + diff change-detection (ARP-spoof / new-host alerts)
SDRRTL-SDR receive; ADS-B aircraft map (43-field intel) live over the bus; RF capture / replay / analyze — transmit hard-gated
Mesh / commsMeshtastic / LoRa SX1262 off-grid messaging; u-blox GPS with PPS time discipline
Server auditnmap-vuln, nikto, lynis, SSH-config review (remote targets gated; secrets env-passed, never logged)
Crackingmanaged hashcat queue; auto-converts captures to .hc22000 (hcxpcapngtool)
AI operatorWaRL0c — on-device assistant that guides an engagement and can drive in-scope ops

03 // the build — build-your-own

Hardware

Documented to the bill-of-materials and the wiring. You assemble it, you understand every layer. This is not a product you buy.

ComponentDetail
ChassisClockworkPi uConsole (handheld — keyboard, LCD, battery)
ComputeRaspberry Pi Compute Module 5, 8 GB RAM
Storage4 TB NVMe (Samsung 990 EVO Plus) — boots from NVMe
ExpansionHacker Gadgets AIO V2 — RTL-SDR + LoRa SX1262 + GPS/PPS + PCF85063A RTC + USB 3.0 + RJ45
Wi-Fi (attack)MediaTek MT7961 USB — monitor mode + injection
SDRRTL2838 (RX) on the AIO; HackRF for TX / replay
OSDebian 13 (trixie), aarch64, kernel 6.12

AIO V2 power rails are enabled at boot via GPIO (SDR 7, GPS 27, LoRa 16, internal-USB 23, PPS 6); the staged build playbook brings up each subsystem with a health probe, so a build is verified, not assumed.

04 // software on the deck

The stack

Security & radio

ToolRole
kismetwireless IDS / capture
aircrack-ngwifi recon / attack suite
hcxdumptool / hcxtoolsPMKID / handshake + conversion
hashcatoffline cracking
reaver / bullyWPS
nmap / nikto / lynisscanning + hardening audit
bettercapnetwork recon / MITM
rtl-sdr / hackrf / soapysdrSDR rx / tx
meshtasticd / gpsd / chronymesh + GPS + time

Application

LayerBuilt with
BackendPython 3.11+ · FastAPI · Uvicorn · Pydantic v2 · SQLAlchemy
Crypto / auditcryptography (Ed25519) · rfc8785 (JCS) for signed AAR records
Web UIReact · Vite · Leaflet (dark theme)
Terminal UIInk (TypeScript) — 16 screens, console or SSH
AI operatorPI agent libs + an OpenAI-compatible provider, key local to the deck

05 // the safety model

The Charter, rendered as code.

SAFE by default
Offensive modules are inert until a scoped engagement is armed.
Scope enforced at run time
Host / subnet / SSID / BSSID checked on every action — for humans and the AI.
Kill switch everywhere
One action cancels every in-flight job across all queues and restores safe state.
RF gets the strictest gate
Transmitting requires an active engagement and a named in-scope target.
The AI can't self-authorize
It acts only inside an engagement a human armed; never arms one itself.
Signed & verifiable
Ed25519 / JCS / did:web attestation — proven against a reference verifier at L1.

06 // see it running

The deck, live.

Real screens from a running Warlock OS deck — one web UI, one terminal UI, and an on-device AI operator, all over the same gated API.

Warlock OS command dashboard — SAFE mode banner, module rail, and live CPU / temperature / memory / disk telemetry tiles with engagement status and mesh node count
Command dashboard — SAFE mode, live telemetry
Warlock OS Offensive SDR page — RF capture and replay refused with HTTP 403 until an engagement is active and the target is in scope
Offensive SDR — RF capture / replay, hard-gated
Warlock OS wireless guided-flow page (ARM, recon, act, loot)
Wireless — guided ARM → RECON → ACT → LOOT
Warlock OS wireless intrusion-detection page — rogue-AP, evil-twin, and deauth-flood detection
Wireless IDS — rogue-AP / deauth detection