Architecture Capabilities Solutions Platform Contact Request Demo
Autonomous Air Defense

Detect. Disrupt. Defend.

Autonomous air defense against drones, missiles and coordinated swarm attacks. Multi-spectral detection fuses into one track picture, rule-based evaluation decides, and soft-kill is sequenced before hard-kill. Detections are cryptographically signed and bound to a registered sensor, and every authorized engagement lands in a hash-chained, append-only audit log — if it can’t be recorded, it doesn’t happen.

Explore Architecture Launch Simulation →

Status — DARKNAIL runs today as a fully simulated system. Detection signing, provenance-gated engagement and the append-only audit log are live in production; the simulator is currently the only signed detection source. No hardware, no fielded deployment.

The Problem

The economics of defense are broken.

A $500 commercial drone can threaten a $100 million facility. Adversaries deploy autonomous swarms of hundreds. Traditional interceptors cost millions per engagement — an unsustainable equation that favors the attacker.

DARKNAIL inverts the equation with a layered engagement ladder: soft-kill is sequenced before anything expensive, and a track that no authenticated sensor has corroborated does not get to spend an interceptor at all. The figures below are illustrative category economics — DARKNAIL has measured none of them, and builds none of the effectors they price.

Attack Cost / Commercial Drone
$0
Traditional Defense / Interceptor Missile
$0
DARKNAIL changes the math
DARKNAIL Response / Soft-kill: decoy or jammer
$0
4,000:1 Interceptor-first cost ratio
vs
1:10 Soft-kill-first cost ratio
Architecture

Four Subsystems. One Continuous Loop.

A closed-loop autonomous defense system. Four integrated subsystems work in continuous coordination — detections in, decision, effector tasking, assessment back. All four are implemented and running today, against a simulator: no sensor or effector hardware exists behind them, and the backend runs as a single stack on a single host.

Sensors

Multi-spectral sensing — radar, EO/IR, acoustic and RF — designed for overlapping 360° coverage and sub-second track creation. Detections arrive over a vendor-neutral contract, are Ed25519-verified against a registered sensor identity, and are fused into one track picture. Unsigned, unregistered, tampered or stale detections are rejected at the boundary rather than de-prioritized inside it.

Decision Agents

Autonomous threat evaluation, response selection and effector coordination — weighted rule-based scoring, deterministic and inspectable, not a trained model. Engagement authorization is gated on the track’s provenance tier, so a decision the sensors cannot corroborate never becomes an effector command.

Effectors

Layered countermeasures from soft-kill (decoys, jammers) to hard-kill (directed energy, interceptors), tasked over gRPC contracts written against no vendor’s SDK. Cost-optimized escalation picks the cheapest response that fits the threat, and never goes above the ceiling the track’s trust tier allows. Every effector behind those contracts is simulated — DARKNAIL builds no hardware.

Assessors

Real-time engagement monitoring, with escalation when a response fails. Outcome reports must come from a service identity, must resolve to an effector command that actually exists, and have their key inputs re-derived server-side before a reward is stored. The reward signal is real and tamper-resistant. The policy behind it is not built — nothing is trained, and nothing yet feeds back into behavior.

Sensors

Multi-Spectral Detection

Four complementary sensor types, designed to provide overlapping coverage from 500 meters to 50 kilometers so no single sensor gap leaves a blind spot. The ranges below are the published envelopes for each class — DARKNAIL builds no sensors and has measured none of them. Today every detection in the system comes from the simulator, under a single registered identity.

Radar 5–50 km

All-weather primary detection and tracking. Micro-Doppler rotor signatures arrive as classification fields inside a signed detection — DARKNAIL verifies and fuses them; it does not compute them.

Primary Detection
EO/IR 1–10 km

Passive optical and infrared imaging for visual confirmation. A second, independently authenticated source is exactly what lifts a track from single-source to corroborated — and corroboration is what unlocks hard-kill.

Visual Confirmation
Acoustic 0.5–2 km

Passive sound-based detection optimized for low-altitude threats, keyed to rotor blade pass frequencies of 50–200 Hz. Passive means the target sees no emission from it.

Low-Altitude
RF Detection 2–20 km

Control-link and telemetry detection, including operator direction finding. Signal strength is one of the seven fields inside the signed payload, so it cannot be quietly altered between the sensor and the track picture.

Signal Intelligence
Decision Agents

From Detection to Action.

Autonomous decision-making with human-on-the-loop oversight — the design point is a machine that can be overruled, not one that overrules. Every threat is evaluated, prioritized and engaged through the same four stages, and the provenance gate sits between the third and the fourth: a decision that cannot clear it never reaches an effector.

01

Perceive

  • Signature and identity verification
  • Fused multi-sensor track data
  • Trust tier from corroborating sensors
02

Evaluate

  • Weighted threat score (1–10)
  • Target value at risk
  • Confidence capped by trust tier
03

Decide

  • Response type selection
  • Engagement priority queue
  • Provenance gate — 403 if unverified
04

Execute

  • Audit write first — 503 if it fails
  • Effector tasking commands
  • Outcome bound back to the command
Effectors

Layered Response. Escalating Force.

Soft-kill first, hard-kill as last resort. Every engagement is optimized for cost-effectiveness, and if one layer fails the system escalates. How far up this ladder a track is allowed to go is capped by the trust tier it has earned from its sensors. DARKNAIL builds none of the four layers below — they are tasked over vendor-neutral contracts, and every one of them is simulated today.

Level 1

RF Decoys

Attract & Divert

Deploy RF and IR decoys to draw radar-guided and heat-seeking threats away from protected assets. Lowest cost per engagement. Soft-kill: reachable by a track with a single authenticated source. A track with no authenticated source at all gets monitoring and nothing else.

Level 2

GPS Jammers

Disrupt & Deny

Disrupt GPS navigation and sever command-and-control links, forcing a drone to land, return to base or drift off course. Still soft-kill — the highest action a single-source track is permitted to reach.

Level 3

Directed Energy

Precision Destroy

High-energy laser and HPM systems burn out electronics or structurally compromise small targets. Near-zero marginal cost per shot. Hard-kill: an autonomous command here requires a corroborated track — two or more distinct authenticated sensors — or it is rejected 403 and no effector is assigned.

Level 4

Interceptors

Kinetic Kill

Hard-kill interceptor missiles for threats that penetrate all prior layers. Highest certainty, highest cost, last resort only. Same corroboration requirement, and the same honest limit: an operator holding the engagement-approval permission can authorize it anyway. The system will not do this autonomously on an unverified track — and when a human overrides, the override carries their name into the record.

Design Targets

Targets. Not Measurements.

These are engineering goals for the production system. None of them has been benchmarked, because no real signal has ever been ingested — the only detection source in the system is the simulator. They are published here as the bar we expect to be held to, not as a capability. The honest counter-number sits underneath them: real-world detections processed to date, zero.

0 Detection Rate (Target)
0 Detection Latency (Target)
0 Decision Latency (Target)
0 Engagement Success (Target)
0 False Positive Rate (Target)
0 Simultaneous Threats (Target)
Capabilities

Built Different. By Design.

Not roadmap. Five feature flags currently enabled in the running stack, and the entire basis of everything claimed on this page. Sensors and effectors sit behind 21 gRPC service contracts with no vendor SDK compiled in — exercised against simulated equipment only, because no hardware exists yet.

01

Signed Detections

Detections on the signed ingest path are Ed25519-signed over a seven-field canonical payload and verified against a registered sensor identity before they can enter the track picture. Unsigned, unregistered, tampered or stale is rejected 422 at the boundary, and the signing timestamp sits inside the signature with a 30-second window, so a captured detection cannot be replayed later. Two older paths — gRPC track-share and manual track POST — are not signature-verified; detections arriving that way are marked unauthenticated and are constrained by the gate below. Stated plainly: in the deployed configuration the only registered identity is a development keypair that lives in the source tree, so this is an architectural control today, not yet a security guarantee.

02

Refuse-to-Fire

An autonomous hard-kill requires a track corroborated by two or more distinct authenticated sensors. Below that tier the effector command is rejected 403 and no effector is assigned. A human holding the engagement-approval permission can still authorize it — that is deliberate, the design point is human-on-the-loop rather than machine veto — so the exact claim is this: the system will not fire on an unverifiable track autonomously, and any override has a name attached to it.

03

Authenticity-Weighted Fusion

Unauthenticated detections are de-rated in fusion, and the confidence a weakly attested track can reach is capped by its trust tier. The consequence is the point: a low-provenance track cannot climb the engagement ladder by accumulating confidence. The ceiling moves with the provenance, not with the number.

04

Verified Reward Inputs

Outcome reports must come from a service identity, must resolve to an effector command that actually exists, and have their key inputs re-derived server-side rather than taken on trust. The reward signal is real and hard to forge. It is also, today, only a signal — there is no trained policy and no update loop, so nothing here learns from an engagement. We will not say otherwise until something does.

05

No Audit, No Fire

Engagement decisions, approvals, overrides and effector commands are written to an append-only log. UPDATE, DELETE and TRUNCATE are revoked from the application role and PostgreSQL rewrite rules turn them into no-ops — enforced below the application, so a future commit cannot quietly weaken it — and a SHA-256 chain links every entry to the one before it, with an endpoint that re-walks the chain to prove nothing has been edited. That makes the record tamper-evident, not tamper-proof: a database superuser can still rewrite rows, and the chain is what makes it show. And the property that gives the whole thing teeth — if the audit write fails, the engagement is refused 503. The system will not act unless it can record that it acted.

Solutions

Scalable Protection

From single-facility point defense to theater-wide coverage, tailored to the threat environment. These are planned system configurations, not shipping products — the node counts and coverage figures are design targets, no configuration below has ever been built in hardware or deployed anywhere, and none of them has been quoted to a buyer. There are no customers and no pilots.

Sentinel
5 km Coverage Radius (Target)
50 Simultaneous Threats (Target)

Point defense for individual facilities. Intended for power substations, water treatment plants, data centers and communication towers.

  • 4 sensor nodes
  • 2 compute nodes
  • 4 effector systems
  • Autonomous operation, human-on-the-loop
Contact Us
Reference Configuration Guardian
15 km Coverage Radius (Target)
200 Simultaneous Threats (Target)

Area defense for medium infrastructure. Designed for airports, military bases, oil refineries and industrial complexes.

  • 12 sensor nodes
  • 6 compute nodes
  • 12 effector systems
  • Full engagement ladder
Contact Us
Fortress
40 km Coverage Radius (Target)
1000+ Simultaneous Threats (Target)

Theater-wide defense for critical national infrastructure and forward operating bases. Full-spectrum protection against mass coordinated attacks.

  • 50+ sensor nodes
  • 20+ compute nodes
  • 50+ effector systems
  • Multi-site coordination
Contact Us
Threat Landscape

The Threat is Evolving

From lone reconnaissance drones to coordinated swarm attacks with thousands of autonomous platforms. The threat grows faster than traditional defense can adapt — and it now includes the sensor feed itself, because a fabricated detection is indistinguishable from a real one and can induce a real engagement.

Reconnaissance Drones Low-cost ISR platforms gathering intelligence
Attack Drones Weaponized UAVs carrying precision munitions
Kamikaze UAS One-way loitering munitions with terminal guidance
Drone Swarms Hundreds to thousands of coordinated autonomous platforms
Cruise Missiles Precision-guided subsonic and supersonic threats
Spoofed Detections Fabricated or replayed tracks that induce a real engagement
Counter-UAS Market $2.1B → $7.8B by 2030 24.3% CAGR — industry projection, unverified, not DARKNAIL data
TRACKING
Platform

Experience the System

A live operational platform: tactical simulation, system dashboard, and a dedicated control station for every sensor and effector. Thirteen web apps and eight backend services, deployed and reachable right now — click any of them. Every station below drives simulated equipment, because there is none behind it: the simulator is the only detection source in the system.

WebXR

RTS Simulation

Command Blue Forces in a real-time 3D tactical simulation. Defend against waves of aerial threats using the full DARKNAIL system.

Launch Simulation → Dashboard

System Overview

System-wide monitoring, status and control.

Open → Sensor

Radar Station

Primary detection and tracking control interface.

Open → Sensor

EO/IR Station

Optical and infrared imaging control interface.

Open → Effector

Decoy Station

RF decoy deployment and management.

Open → Effector

Jammer Station

GPS and C2 link jamming control interface.

Open → Effector

Laser Station

Directed energy weapon control interface.

Open → Effector

Interceptor Station

Interceptor launch sequencing and guidance — simulated.

Open → Sensor

Acoustic Station

Sound-based detection and analysis.

Open → Sensor

RF Station

RF signal detection and direction finding.

Open →
Contact

Ready to Defend?

Contact us for demonstrations, partnership opportunities, and technical assessments of your defense requirements. To be direct about where we are: DARKNAIL is simulated end to end today, with no hardware and no deployments, and what we most want is a first design partner willing to run it against at least one non-simulated sensor feed — a recorded dataset counts.

contact@darknail.com LinkedIn