Machina · the IDE for machines
Machina is an AI engineer for drones and robots. Describe the machine and its goal, and its agent picks real parts, models the frame, designs the electronics and checks them. We're closing the loop now: the real autopilot firmware flies every design in simulation, and the agent keeps improving it until the goal is met.
Live in your browser: a toy quadrotor flies 1,000 laps through seeded gusts and is redesigned every 250. Move the pointer to blow wind at it.
A recorded agent session, sped up. The agent picks motors, props and a battery from the parts library, builds the frame feature by feature in carbon-fibre sheet, checks each arm at full throttle with finite elements, and produces ISO drawings, DXF files and a bill of materials.
We built our first racing drone from scratch. We printed the frame and built the circuits, and thirty seconds into its first flight it fell out of the sky.
We had missed one capacitor. When the motors braked, the voltage spike on the battery bus reset the flight controller. That one mistake cost us three weeks and $600, and nothing we used could have caught it before we built.
So we're building the tool that would have.
Peak 36.9 V against a 36 V rating: the flight controller resets.
They hide where parts meet, so no single tool sees them. Each one takes minutes to fix on a screen, and a crashed prototype to find in the air.
A battery lead that rubs through its insulation on a carbon arm shorts the whole drone mid-flight.
An antenna tucked inside the frame loses the control link at range.
Carbon fibre bolted straight onto aluminium corrodes the aluminium.
Motor screws that reach the windings short the motor the first time it spins up.
At full throttle the motors pull more current than the pack can deliver. The voltage sags, the flight controller resets, and the drone falls.
On a 6S pack, braking motors push spikes back into the power lines that kill the flight controller, unless a capacitor sits in the right place.
If the frame's natural frequency falls inside the motors' RPM range, vibration floods the gyro and the drone shakes itself out of control.
At full thrust a thin arm bends far enough for a prop to strike the frame.
ESCs and regulators packed into a closed stack overheat halfway through a hard flight.
A robot joint that holds its load standing still but stalls the moment the arm accelerates.
Software engineers have agents that write, test and fix code in a loop. Hardware never got one, because its tools don't talk to each other. Machina keeps a machine's mechanics, electronics and physics in one model that the agent can query, change and test, so a change anywhere is checked everywhere.
Lap 4 · goal met
| Mass ≤ 650 g | 641 g | Pass |
|---|---|---|
| Thrust ÷ weight ≥ 10 | 10.1 | Pass |
| Endurance ≥ 4 min, 95% | 4.2 ± 0.1 | Pass |
| Bus surge ≤ 36 V | 26.1 V | Pass |
Goal met with 95% confidence. Export.
“A 5-inch racing drone, at most 650 g, with ten times its weight in thrust and four minutes of flight.”
Real parts from datasheets. Frame, electronics and power network in one parametric model.
The real autopilot firmware flies it, in lockstep with the physics: PX4, ArduPilot, Betaflight, INAV, or your own controller.
Every requirement checked with its uncertainty. Missing evidence is inconclusive, never a pass.
The agent changes geometry, parts or electronics, shows the result behind each change, and goes round again.
Drawings, CNC and 3D-print files, a bill of materials, and the firmware's own parameter file.
An illustrative run. In Machina, every change the agent makes shows the result behind it.
From a few sentences of requirements, the agent picks real parts from datasheets, models the machine in 3D, designs its electronics, checks stress, fatigue, dynamics and power, fixes what fails, and produces drawings, CNC and 3D-print files and a bill of materials.
The real autopilot firmware flying our physics, in lockstep with the simulation clock, so no sensor sample is late and nothing runs ahead.
Close the loop: power transients in flight, sensors and faults, and studies under uncertainty, joined to every autopilot and run until the goal is met.
CEO
Builds AI agents and infrastructure for a 64,000-GPU AI supercomputer. At Hooke he leads the agent, the electronics and the 3D modelling.
CTO
Over ten years writing software for hardware: embedded firmware, a GPU neural-network engine, computer vision. At Hooke he leads the architecture and the simulation.
Based in Dublin, Ireland.
ceiiinosssttuv
In 1676 Robert Hooke published a law of nature as an anagram, ceiiinosssttuv, to claim the discovery without giving it away. Two years later he gave the answer: ut tensio, sic vis. As the extension, so the force.
Every stress check Machina runs still starts there.
We're opening Machina to the first drone and robot builders in December 2026. Tell us what you're building.