ATC simulator with AI-driven virtual pilots. No cloud costs. Built to democratize aviation training on consumer hardware.
How It Works
Load any ICAO airport and start controlling. Each AI agent acts as an independent pilot — listening to your instructions, reasoning over them, and transmitting a readback through synthesised voice in real time. Three standard positions cover the full flight cycle, from engine startup to takeoff and from landing to stand.
Clearance Delivery — Flight plan clearance.
CTLR “Iberia 3142, cleared to Palma via LARPA6Q, runway 24L, initial climb 10,000 feet, QNH 1018, squawk 2512.”
IB3142 · AI — Agent generated response “Cleared to Palma via LARPA6Q, runway 24L, initial climb 10,000 feet, QNH 1018, squawk 2512, Iberia 3142.”
Ground Control — Taxi management.
CTLR “Iberia 3142, taxi to holding point runway 25L via taxiways Alpha, Bravo.”
IB3142 · AI — Agent generated response “Taxi to holding point runway 25L via Alpha, Bravo, Iberia 3142.”
Tower Control — Takeoff and landing authorization, and separation in the traffic circuit.
CTLR “Iberia 3142, wind 260 degrees, 12 knots, runway 25L, cleared for takeoff.”
IB3142 · AI — Agent generated response “Cleared for takeoff runway 25L, Iberia 3142.”
Why AIrport
Run the full ATC stack on your own machine with local LLMs (Ollama). No internet, no accounts — fully private and free on consumer hardware.
Rather skip local GPUs? Offload the AI pilots and speech pipeline to Google Cloud Run services. Same ICAO phraseology, scaled in the cloud.
Local mode is free and offline; cloud mode is optional and pay-as-you-go. Either way — no subscriptions, no lock-in.
Watch aircraft move across the airport in real time — taxiing, lining up, and taking off — all driven by X-Plane 12 integration responding directly to your ATC instructions.
Tech Stack
Each piece chosen for a concrete reason: local performance, minimal latency, and zero cloud dependencies.
Core LLM
Local LLM inference without cloud, without quotas.
Orchestration
Agent Development Kit for building virtual pilots with memory and persistent state.
Speech Input
Offline speech recognition. The controller speaks — Whisper transcribes in real time.
Speech Output
Voice synthesis with aeronautical normalization: numbers, callsigns, and ICAO phraseology.
Infrastructure
Isolated microservices. Each subsystem is an independent, restartable container.
Real Time
Message bus and shared state between microservices. Sub-millisecond latency on local hardware.
Persistence
Storage for flights, traffic, and simulation sessions for post-analysis.
Simulator
Direct integration via XPPython3 for aircraft spawning at the simulated airport.
Timeline
Real project progress, no marketing.
Docker infrastructure, PostgreSQL schema, Redis event bus, and first HMI prototype with electronic flight strips (EFS).
XPPython3 integration for spawning aircraft at any ICAO airport within X-Plane 12.
Google ADK orchestration of virtual pilots with memory and full voice pipeline (Whisper → LLM → TTS) deployed in the simulation.
Aircraft responding in real time to controller instructions — taxi routing, lineup, takeoff clearance, and landing sequences.
Post-session analysis, performance metrics, and public demo at LEBL (Barcelona-El Prat). Open release of the project.