Using some sort of adaptive traffic light control system is becoming standard policy among metropolitan areas. However, controlling traffic lights efficiently on a city-wide scale is computationally intensive and theoretically complex. This paper aims to show a proof of concept of an efficient and modular traffic. light controller with comparatively little computational overhead. The proposed system features distributed agents, each representing an intersection, capable of making individual decisions. These agents base their decisions on short term traffic flow forecasting and received information from neighbouring agents on incoming traffic. Testing shows the proposed controller being more efficient than less adaptive systems in terms of reduced average vehicle time loss and reduced average vehicle stop time. This paper describes the attractive properties of the system in detail, shows the shortcomings of the design choices and gives suggestions on how to improve the system in the future.