Where It Started
The goal was programmatic flight control — a proper API layer over drone hardware instead of manual controllers and disconnected tools. Once that foundation was solid, the next question was obvious: what if an AI agent drove it? Fleet dispatch, mission planning, incident response — all the things operators do manually, automated through the same API.
What We Built
Two layers. DroneOS is the open infrastructure — the control stack every drone application sits on top of. AI Fleet Command is the first application built on it: an agent-driven ops layer that handles incident intake through mission execution, with the operator supervising rather than driving.
- DroneOS — the open-source infrastructure layer: PX4, ROS 2, simulation, deployment tooling, and operator-facing drone control APIs
- AI Fleet Command — the application layer on top: incident intake, drone selection, autonomous dispatch logic, and operator-supervised mission execution
- Agent-driven control loop: an AI agent calls the DroneOS API to select a drone, plan the mission, and execute dispatch — operator stays in the loop for oversight
- Emergency-response demo: ~10 seconds from incident to dispatch in simulation
The Facts
- Two-layer architecture: DroneOS (open infrastructure) + AI Fleet Command (agent ops layer)
- Agent-operated: AI calls the control API — no manual dispatch required
- Operator-centered: built for real oversight, not black-box autonomy claims
- Open source: DroneOS is publicly available — audit it, extend it, build on it
- Proven in simulation, path to private-site deployments (industrial, logistics, construction)
Why It Matters
DroneOS proves the same thing FlixML does from a different angle: an AI agent can operate real, physical-world infrastructure end to end. The agent isn't writing emails or summarizing documents — it's dispatching hardware. That's the kind of system we build.