Concept film: four firefighters advance on a burning block of flats at night while FireFly drones cool the fire from outside.

Drone swarm for fire crews · In development

The all-seeing swarm. So firefighters arrive knowing.

FireFly is a firefighting drone swarm in development for fire crews. It is designed to launch on the alarm, find the fire and the people it can see before the engine arrives, and start cooling the fire from outside on the officer's word. Firefighters arrive knowing, and put it out.

  • First on sceneFlies straight there on the alarm
  • The crew arrives knowingFire and people, sent ahead to the cab
  • Firefighters finish itThe swarm buys time, on their word
Play the simulation See how it works
  • Patent Pending
  • AWS Innovation Award10,000 AIdeas · 2026
  • FireFlyby Caligo Dynamics

01 · The problem

Every fire starts the same way. Smoke fills the room, the crew cannot see inside, and the first minutes are spent finding out. In a modern furnished room, flashover can arrive in 3 minutes 20 seconds.

Blind

Smoke is a blindfold

Crews go in without knowing where the fire is, which way it is running, or who is still inside. The first person through the door is the one who finds out.

Late

The first look costs minutes

Before anyone can attack, someone has to look: a walk round all four sides, a ladder, a radio relay. The fire keeps growing the whole time.

Unknown

A call rarely says enough

Callers report what they can see: smoke, an alarm, a smell. How big it is, and exactly where, is usually found out on arrival.

3:20
Time to flashover in a modern furnished room. Matching rooms with older natural furnishings took about half an hour
UL FSRI / Kerber, Fire Technology 2012
9:25
Average call to arrival in England: dispatch, turnout and real distance
UK Government (MHCLG), 2026
83
US firefighters who died on duty in 2025, 22 of them on the fireground
NFPA
1.08M
Hectares burned across the EU-27 in 2025, the highest in the EFFIS record
EU JRC / EFFIS

02 · Same building, two cameras

A camera sees smoke. Thermal shows the crew which flat is burning.

Long-wave infrared reads heat, not light, so smoke hides far less from it. Choose what is happening inside one flat, then drag across the building to compare the two views from the street.

Simulation: a block of flats at night in an ordinary camera. Smoke hangs across the upper floors and hides where it is coming from. Simulation: the same building in thermal. One window, four floors up, glows far hotter than every other, and it is marked. Camera Thermal
Drag · or use arrow keys

A simulation drawn in code, not sensor output. Temperatures are illustrative. Thermal does not see through window glass: it reads the glass, the frame and anything hot that escapes. Our demonstrator streams live infrared today; captures on request.

Inside the flat

Four floors up · third flat from the left

Thermal
One window is far hotter than every other and still climbing. Heat is leaking round its frame. That is the flat.
Camera
Smoke across the upper floors. From the street it is hard to say which flat it is coming from.
The officer gets
The floor and the flat are known before the engine stops, so the first hose line goes to the right door.
40%

of the incidents fire and rescue services attended in England in the year to March 2026 were fire false alarms: 255,047 of 641,227. Around two thirds of those came from automatic alarm systems, where a view from outside often shows nothing.

UK Government (MHCLG), 2026

Not every alarm is a fire. The ones that are need knowing early.

A call comes in with whatever the caller could see. FireFly is designed to be looking at the address within minutes. When heat or smoke is showing, the officer knows it is a working fire while still on the road, and can ask for more help before the first engine stops.

FireFly never cancels a call, and never downgrades one. An outside view cannot rule a fire out. When nothing is showing, someone still goes to confirm, and how many is the service's decision.

03 · What the crew sees on arrival

Three pictures, on the crew's phones, before the engine stops.

The officer's walk round the building takes minutes the fire does not give back. FireFly is designed to fly it first and send the answers ahead.

  1. Where the fire is

    All four sides in thermal: the seat of the fire and which way it is moving. The first line goes to the right door.

  2. Who is on the roof

    A person in the open stands out in thermal, through smoke and in the dark. The crew knows to go up before anyone has to shout.

  3. Who is at a window

    The side of the building nobody on the street can see. The ladder goes to the right window, first time.

What it cannot see, it says. These are the people FireFly can see from outside. Anyone behind a wall or a closed window does not show, so the search is still the crew's.

Concept screens. The building view is a simulation and the two pictures of people are AI-generated; none of it is sensor output.

Concept illustration: at the back of a fire engine at night, an officer holds out a phone with a thermal picture of the building and points up the street while two firefighters pull a hose line off the engine.

FireFly brings the picture.
Firefighters bring everything else.

Privacy, by design. FireFly is designed to record only during an incident. Pictures of people go to the responding crew and the incident record. Who else may see them, on which devices and for how long is set by the fire service and agreed in writing before a pilot flies. How we treat aircraft imagery

04 · How it works

One call, from the alarm to the last hose line.

Scroll through it. The swarm is first, and the crew is the ending.

  1. 01 / 06

    The alarm comes in.

    One tap on the address and the swarm lifts off its station nest as the crew turns out. Design target: airborne in under 90 seconds.

    Concept illustration: FireFly drones lined up on marked pads outside a fire station at night.
  2. 02 / 06

    First on scene.

    It flies the straight line while the engine takes the road, holds off, and starts looking.

    Concept illustration: a FireFly drone arriving at a burning apartment block while a fire engine is still far down the street.
  3. 03 / 06

    It finds the fire.

    One lap of all four sides in thermal: where the fire is, how hot, and which way it is moving.

    Simulation: a block of flats in thermal, one flat venting flame up the face of the building and the flat above marked as heating.
  4. 04 / 06

    It finds who it can see.

    Someone on the roof. Someone at a rear window. Each is marked and sent ahead. People it cannot see may still be inside.

    Concept illustration: seen from a drone, a person leans out of an open rear window and waves while smoke rolls out above them.
  5. 05 / 06

    The crew arrives knowing.

    The pictures reach the crew before the engine stops. On the officer's word, the swarm starts cooling the fire from outside.

    Concept illustration: in the cab of a moving fire engine, an officer turns to show two firefighters a thermal picture of the building on her phone.
  6. 06 / 06

    Firefighters put it out.

    The cooling stops before the crew goes in. The line goes to the right door, the ladder to the right window, and the swarm watches over them.

    Concept illustration: two firefighters in breathing apparatus drive a hose line through a doorway into a burning flat.
Still from the animated diagram: the fire engine has arrived, the drones ring the building, and the people seen on the roof and at a rear window are marked.
Concept illustration: a firefighter carries a resident wrapped in a blanket out of the building, a second firefighter beside them, two small drones holding position overhead.

FireFly finds it. Firefighters finish it.

Every decision stays with the firefighter.

Concept animation and AI-generated scenes of the system as designed. What’s inside FireFly →

05 · Meet FireFly

Built around one job: being the crew's eyes in the smoke.

  • Thermal + visible eyesSees heat, not lightTwo sensor eyes pair long-wave infrared with a visible camera. AI marks fire and the people it can see through the plume.
  • Directed nozzle armA cone, not a mistThe nozzle sits forward on a rigid arm, angled 45° down, and throws a directed cone at the hottest point from a safe standoff.
  • Twin suppressant podSwapped in minutesThe pod is the unit of resupply. An empty aircraft lands, the pod is exchanged, and it rejoins the rotation. Target: under three minutes.
  • Navigation stripsVisible to every crewRed arm lighting keeps each aircraft easy to track at night, in rain and in smoke.
  • Four-rotor airframeSmall, many, replaceableSeveral small aircraft beat one big one on cost, redundancy and safety. Lose one and the ring closes the gap.

Concept design · AI-generated render. Today's flying demonstrator is a Holybro X500 v2 on PX4.

06 · What changes for the crew

Today, and with FireFly.

Size-upThe officer walks all four sides on arrival.The walk round is flown before the engine stops, in thermal, and sent to the officer.
PeopleWho needs reaching, and where, is found out on arrival.People on the roof and at windows are marked before the crew arrives. The search inside is still the crew's.
First lineThe first line goes where the smoke suggests.The first line goes to the seat of the fire, marked floor by floor.
The waitThe fire grows unchecked until the engine arrives.On the officer's word the swarm cools the fire from outside until the crew is ready to go in, and never stops for a refill.
Early callHow big it is gets found out on arrival.A working fire is known while the engine is still on the road, so more help can be called early.
CommandInformation arrives piece by piece over the radio.One live picture the officer can share with the crew. The officer decides, and can stop the swarm with a word.
<90s

From dispatch to airborne, launching from the station nest

Design target
5+

Aircraft coordinated by a single operator

Design target
<3min

Pod swap on the nest, not counting the flight there and back

Design target
0

Pauses in suppression while aircraft rotate

In simulation

The "with FireFly" lines describe the system as designed. Arriving ahead of the engine depends on flight beyond the pilot's line of sight, which is in development and needs permissions we do not yet hold. How much fire a ring of aircraft can check has not been measured yet. The figures above are design targets and simulation results, not independent field measurements; we will publish measured values as field tests complete.

07 · Where it works

Wherever smoke takes the view away.

Concept illustration: five drones holding a line along a forest road, laying a containment line in front of a wildfire.
01 · Wildfire

Find the front before it jumps.

Find the front through smoke, map the perimeter and hit hot spots along roads and ridges, so crews are sent where the fire is going.

  • Perimeter mapping
  • Hot-spot suppression
  • Crew safety
Wildfire response
Concept illustration: top-down view of drones in a ring around fires on an industrial roof, spraying inward.
02 · Industrial sites

On the roof before the crew is through the gate.

Warehouses, plants and refineries: large roofs, fast smoke, dangerous interiors. The swarm gives the site team and the fire service the first picture, and starts work on the roof while they are on their way.

  • Instant response
  • Asset protection
  • Less downtime
Industrial response
Concept illustration: drones circling a rooftop terrace fire on a high-rise, spraying inward.
03 · High-rise & urban

Upper floors, rooftops, dense streets.

The first picture of the upper floors while ladders are still being set: which flat, which way it is moving, and who is on the roof or at a window.

  • Which flat
  • People located
  • Exterior cooling
Urban response
Concept illustration: a drone arriving over a highway crash with a burning car, directing a spray at the vehicle.
04 · Roads & emergencies

First eyes on a highway fire.

Vehicle fires and crashes: the scene assessed, people located and the crew briefed before the first engine arrives.

  • Vehicle fires
  • Scene assessment
  • Hazard zones
Road response

Scenes are AI-generated concept visuals.

08 · Safe by design

Autonomy that answers to a person.

These rules are enforced in the aircraft's configuration, not in a policy document. We will not trade them for a smoother demonstration.

The physical controller always wins

A pilot takes over from autonomy in one movement. The transmitter outranks the software at every instant of every flight.

A lost link means a safe return

If the link drops, failsafes bring the aircraft home. We never loosen failsafe timings to hide a weak link.

5G for reach, local radio for authority

The operator interface runs over 5G from anywhere, and every firefighter on scene can watch the live view on a phone. Underneath, the local radio and the physical transmitter stay in charge: if the 5G link drops, the aircraft's failsafes take over.

A human yes to start. A word to stop.

The swarm proposes. A firefighter confirms, every engagement, every time. The officer can stop it at any moment, and it always stops before a crew goes through the door.

09 · Inside FireFly

Everything the swarm needs to work inside the smoke.

Flight, perception, avoidance, suppression and the data behind every mission. One system, built by the team that flies it.

New

LiDAR + thermal fusion

A scanning LiDAR on board maps the geometry around the aircraft and feeds obstacle avoidance. When smoke thickens and laser returns fade, the thermal array measures how blind the LiDAR has become, and the aircraft slows down and lengthens its stopping distance on its own. It never mistakes smoke for open sky.

Thermal perception

A long-wave infrared array reads heat instead of light and streams live to the ground station alongside visible video.

Autonomous flight

PX4 on a Pixhawk 6C with a dedicated companion computer: verified offboard control, takeoff referenced to the ground it left, map-point navigation at held altitude and an orbital scan on arrival.

AI fire & smoke recognition

A GPU-accelerated detector watches the live feed and raises hotspot alerts, each stamped with position, heading and altitude.

Obstacle avoidance

A potential-field planner fed by the LiDAR scan steers around obstacles, with a tangential bias so the aircraft escapes dead ends instead of hovering in them.

Directed suppression

A servo-actuated valve on a pressurised canister, commanded over MAVLink, releases agent as a directed cone once the operator confirms.

Self-healing swarm

Adaptive control rescales thrust as the payload empties, and a standby aircraft matches orbital speed to take a vacated slot, so suppression never stops.

5G link, live view on any phone

Over 5G the operator interface opens from anywhere: map, live video and mission commands. Every firefighter on scene can open the drone's live view on their phone, and the aircraft sizes its video to the network it is on.

Automatic flight records

Every flight is segmented, logged and closed automatically: telemetry, command latency, link health and every detection event. Nobody has to remember to press record.

Request the flight records

This section describes the full system as designed. What is flying on the demonstrator today, and what is still simulation, is set out under Is FireFly flying today?

10 · Play it

Run the swarm yourself, in your browser.

Our playable simulation puts you in the operator's chair. The drones fly themselves; you make the calls.

  • No installClick the link and it opens in your browser.
  • First mission, about 6 minA rooftop fire in Brooklyn's DUMBO.
  • The full loopLaunch, thermal scan, suppression, hand-off.
  • CampaignMake New York safe, district by district.
Play the simulation

“The drones fly themselves. You make the calls.”

Concept simulation. Flight times, spray range and hand-off timings are game values, not measured performance.

Intellectual propertyPatent Pending
AWS 10,000 AIdeasInnovation Award 2026Recognised in Amazon Web Services' global programme.
Delaware C-CorpCaligo Dynamics Inc.Investable on standard instruments.
Flying prototypeX500 v2 · Pixhawk 6C · PX4Thermal and visible video, Raspberry Pi companion.

11 · Next

Bring FireFly to your service.

Fire services, municipalities, industrial sites

See it for yourself

We are looking for the first fire services to design a controlled field pilot with. If you run a fire service, a forestry operation or an industrial site, we would rather hear your objections now than discover them in a field.

Request a demo

Investors

Read the case, then ask for the brief

Why now, the wedge, the objections answered, the roadmap and what the money buys, with flight data behind every claim.

For investors

Sources

  1. Kerber, S. “Analysis of Changing Residential Fire Dynamics and Its Implications on Firefighter Operational Timeframes.” Fire Technology 48 (2012). Companion experiments: UL Fire Safety Research Institute, Comparison of Natural and Synthetic Home Furnishings. springer.com · fsri.org
  2. UK Ministry of Housing, Communities and Local Government. Detailed analysis of fires and response times, England, year ending March 2026, published 19 August 2026. gov.uk
  3. Montagna et al. “Effect of fire department response time on fire damage in the Netherlands.” Fire Safety Journal 157 (November 2025). pure.eur.nl
  4. EU Joint Research Centre / Copernicus EFFIS. Advance Report on Forest Fires in Europe, Middle East and North Africa 2025, 31 March 2026. ec.europa.eu
  5. NFPA. Fatal Firefighter Injuries in the US in 2025 (released July 2026). nfpa.org
  6. UK Ministry of Housing, Communities and Local Government. Fire and rescue incident statistics, England, year ending March 2026, published 22 July 2026. Of 641,227 incidents attended, 255,047 were fire false alarms: 172,247 due to apparatus, 77,181 made with good intent and 5,619 malicious. gov.uk