Autonomous drone swarms. No pilot. No GPS. No signal.
When drones stop being tools and start being teammates.
Watch the DemoWhen GPS drops, radios die, or pilots can't enter, most drones turn into expensive ceiling fans. Prometheus does the opposite.
We're building fully autonomous, LLM-driven drone swarms that fly and reason in GPS-denied environments with no pilot, no constant radio link, and no external compute. Each drone runs an onboard foundation model that fuses perception, mapping, planning, and natural communication.
You don't "program" missions – you state intent:
"Search the second floor."
"Map the west wing and find trapped people."
"Show me all heat sources and blocked exits."
- LLM reasoning and interface
- Full autonomy (no pilot / no GPS)
- Indoor & close-to-obstacles flight
- 3D mapping & point cloud
- GPS-denied environments
- Self-charging - 24/7 operations
LLM First
Intelligence is the architecture, not a bolt-on.
Prometheus is designed around onboard multi-modal LLMs that understand space, reason about missions, and communicate in natural language. The swarm plans, adapts, and explains what it's doing – instead of executing hard-coded scripts.
Swarm First
Designed from day one for multi-drone advantage.
Drones share maps, divide work, and coordinate without a central pilot. Complex tasks – search, mapping, relay, inspection – are handled as a team, not as a single flying camera.
Full-Stack Control
Custom airframes, autonomy stack, and AI – all in-house.
We control the whole stack: drone hardware, SLAM, pathfinding, docking, and the LLM agent. That lets us iterate quickly and avoid vendor lock-in with DJI or other closed ecosystems.
Open Stack, HW-Agnostic
Today we ship our own drones; tomorrow we can be the "swarm OS" for others.
Our long-term plan is to license the AI/OS to other drone manufacturers and robot platforms – aerial, legged, wheeled – while still selling our own reference hardware where it makes sense.
Cost-Optimized, Reliability-Obsessed
Intelligence replaces heavy, expensive sensor suites.
We design for low cost per unit and high reliability: robust SLAM, safe indoor flight with prop guards, self-charging docks, and aggressive stress-testing in smoke, glass, low light, and narrow corridors.
Built in Europe, for Europe
Autonomy at this level is strategically important.
We want GoodAI Swarm Robotics to become Europe's leader in intelligent drone swarms – a sovereign capability for SAR, defense logistics, and critical infrastructure. This tech will exist; the question is who builds it and where .
An example of an LLM-controlled drone in the real world.
Swarm example in simulated hospital.
The briefcase dock is deployed to an unknown site.
The operator uses a web-based UI to send a drone inside the building. After it returns and auto-docks, they speak to the onboard LLM to review what the drone saw – people, hazards, structural details – directly on the 3D map.
This shows open-domain reasoning in a GPS-denied environment: the LLM can flag likely people, highlight potential fire hazards, and explain its reasoning in plain language.
Concept video, AI Generated
Context
Firefighters arrive at a multi-story building with smoke, unknown hazards, and potential victims. Time and safety are everything.
Deployment
- Set the briefcase-style docks on the floor near the entrance, open the lids.
- Speak: "Drones, take off. Map the interior and search for people. Identify all fire sources and blocked exits."
- The swarm launches, establishes a mesh, and begins coordinated exploration.
What Happens Next
Autonomous SAR
Drones navigate smoke and tight corridors using vision-based SLAM and onboard reasoning. They split up intelligently, avoid each other and obstacles, and cover more ground faster than any single pilot could manage.
Situational Awareness
A live 3D point cloud builds on the tablet as they fly, annotated with detected people, heat sources, hazards, and structural features. The LLM narrates findings in plain language.
Natural Interaction
Firefighters update intent without touching a joystick. "Prioritize the third floor stairwell and confirm safe egress routes." The swarm replans instantly.
Power Autonomy
Low-battery units return to dock and auto-charge while others continue the mission. No human juggling of batteries or flight rotations.
Mission Close-out
On command, "Drones, return home," the swarm lands and docks itself, charges, and saves the full map and mission report. No pilot ever entered harm's way.
Outcome
Faster victim discovery, clearer decisions, and fewer humans in danger – all with zero manual piloting, no GPS, and no live radio link required.
Phase I
Where We ArePrometheus already flies.
We have a working prototype where a cloud LLM controls a real drone and coordinates multiple drones in simulation. The stack interprets natural-language goals, reasons about the mission, and converts decisions into robust low-level flight behavior.
Current focus:
- Robust indoor autonomy (narrow corridors, low light, mild smoke)
- Reliable self-docking and self-charging
- Stress-testing and improving resilience and user experience
Phase II
What Comes NextFrom one drone to real-world swarms.
Drones will collaborate, share partial maps, divide tasks, and adapt on the fly without centralized micromanagement. Think real-time search patterns, formation maneuvers, and mission reprioritization driven by high-level instructions.
Goals for this phase:
- Coordinated multi-drone operation in real buildings
- Better flight time (~2×) and faster / cheaper LLM loop
- Pilot projects with fire brigades, inspection, and construction partners
Phase III
The HorizonComplete autonomy with onboard LLMs.
Every drone carries its own reasoning engine, launches from self-charging docks, executes missions end-to-end, and syncs knowledge on return. No external compute. No tether. Indoor & outdoor.
The system becomes a distributed robotic intelligence platform that generalizes across missions and environments: SAR, logistics, construction, inspection – and eventually off-planet and deep-ocean deployments.
Technical Snapshot
- Autonomy stack: vision-based SLAM, voxel map / point cloud, path planning, collision avoidance, swarm coordination
- Compute: Jetson / mobile-class edge compute with on-board LLM agent
- Environment: Indoor + close-to-obstacles, GPS-denied, low-light, mild smoke
Prometheus is the starting point for GoodAI Swarm Robotics becoming Europe's leader in intelligent autonomous swarms. We're building the aerial equivalent of humanoid robots: fast, aware, and self-reliant.
Near-term, our focus is indoor search & rescue, inspection, and mapping – places where GPS doesn't work and pilots shouldn't go. Over time, the same cognitive swarm platform scales to:
- Defense & medical logistics in contested environments
- Industrial inspection & construction, including actuator-equipped drones
- Autonomous security and monitoring for critical infrastructure
- Deep-ocean and space operations – in-orbit inspection, debris removal, off-planet construction
Long term, we see Prometheus as a hardware-agnostic "swarm OS":
- We can sell our own drones
- License the AI/OS to other drone manufacturers
- And support other platforms (legged, wheeled, VTOL) where onboard intelligence and swarm coordination are needed.
We plan to open this up via dev kits and APIs, so partners and the community can build specialized modules and applications on top of the swarm.
Prometheus: when drones stop being tools and start being teammates.
We're now moving from prototype to real-world pilots and company-building.
We're looking for:
- Founding CEO / Operator – Proven fundraising + scaling experience in robotics/AI or deep tech, ready to secure the Seed round and build Prometheus into a global business.
- Pilot Partners – Fire brigades, disaster response teams, industrial inspection, construction, security. If you want to test autonomous swarms in your operations, we'd like to talk.
- Engineers & Researchers – Robotics, SLAM, LLMs, embedded systems. We're not hiring at scale yet, but we want to keep a warm pipeline of people who can build this with us.
- Investors – Seed-stage and dual-use/defense-focused funds who believe Europe should lead in intelligent autonomous systems.
Contact & newsletter:
- For general inquiries and pilots: info@goodai.com
- For CEO / careers: see goodai.com/careers
- For launch updates and tech progress: subscribe to the GoodAI newsletter.
FAQ
Q1: Why put an LLM on board instead of in the cloud?
Because in real missions you can't rely on signal. Fires, collapsed buildings, basements, or contested environments are all GPS-denied and connectivity-denied. Onboard LLMs let the swarm reason, adapt, and communicate even with no uplink, then sync data when it returns.
Q2: Why no pilot? Isn't it safer to keep a human in the loop?
We still keep a human in command, but not on the sticks. Pilots are a scarce resource and become a bottleneck in complex missions. Our goal is:
"Send three drones into a building, get a 3D map and a briefing back – without anyone learning FPV."
Humans define goals and constraints; the swarm handles low-level flight, collision avoidance, and coordination. That's safer, faster, and more scalable.
Q3: How is this different from DJI / Elios / other inspection drones?
- They are typically single drones with a pilot, optimized for video capture.
- Prometheus is a swarm with an onboard LLM, optimized for autonomous mission completion in GPS-denied spaces.
- We focus on indoor mapping + semantic understanding + autonomy + self-charging, not just getting a camera into a hard-to-reach spot.
Q4: Will you sell hardware, license software, or both?
Both options stay open on purpose.
- Short term: we sell a full stack (drones + docking + software) so pilot partners can test end-to-end.
- Mid/long term: we also license the AI/OS "brain" to other drone manufacturers and robot platforms, so they can make their hardware intelligent and swarm-capable.
Q5: What are the current limitations?
We're still in alpha. Today, limitations include:
- Short mission time per drone (we extend this with auto-recharge cycles)
- Edge cases: glass, reflective surfaces, heavy smoke (we're adding sensors and training for this)
- Mostly indoor and close-to-obstacle operation. Outdoor & all-weather support comes later.
We're deliberately honest about limits and treat reliability as a brand: publish real KPIs, share failure modes, and avoid over-promising.
Q6: Do you have a simulator?
Yes. We use simulation for end-to-end testing of intelligent swarms: mapping, pathfinding, LLM-driven decision-making, and multi-drone coordination. This lets us test edge cases that would be too risky or expensive to reproduce in the real world and speeds up iteration.
Q7: Is this a dual-use / defense technology?
Yes. Prometheus is dual-use by design: the same core system can save lives in civilian SAR and provide critical capability in defense and national security. Our stance is pragmatic:
- We start with civilian SAR and industrial use-cases,
- But we are open to defense applications that align with European security and democratic values.