We introduce a L-infinity notion of approximate Nash equilibrium which, unlike the standard one, guarantees a uniform small departure of agent's performance from their maximum value. Our result covers situations where agents criteria depend on the conditional law and allow for controlled agent interactions.