A distributed hive mind that runs on the drones
How it works
The fleet decides on-board
Every drone hears every other drone's beacon and keeps a table of its neighbours. A control law on the drone turns that table into a position target for its own flight controller. No node is in charge, so no node is indispensable.
The model
Layered and local-first
The intelligence sits in layers on top of the swarm bus. It adds no radio traffic of its own: every input it needs is already on the air.
- Transport
- Swarm bus, 20-byte beacons at 2 Hz
- World model
- Neighbour table, stale after 3 s
- Roles
- None elected; the operator is the authority
- Tasking
- Consensus-based task auction
- Behaviors
- Formations and flocking
- Safety
- Repulsion from 8 m, climb and hold inside 4 m
- Human role
- Intent and oversight
Every drone runs the same logic; no leader holds the formation together.
Two invariants
Snappy and resilient by design
Local-first
No node waits on a central brain, so the fleet stays responsive. Each drone runs its own control loop ten times a second from what it hears.
No competing failsafe
When the swarm layer has nothing trustworthy to fly on, it sends nothing and lets the flight controller hold. The autopilot's own failsafes and geofence stay in charge.
Precedence
Five levels, safety first
Every drone arbitrates between five levels, highest first. The level that is actually active, not the one somebody asked for, rides in the beacon to the operator screen.
Command intent, not each drone
The swarm logic is open-source, ships switched off, and runs its flight scenarios as ordinary tests on a laptop. Follow the design as it comes together.
Read the docs