We develop a new perturbative method for studying any steady states of quantum impurities, in or out of equilibrium. We show that steady-state averages are completely fixed by basic properties of the steady-state (Hershfield's) density matrix along with dynamical "impurity conditions". This gives the full perturbative expansion without Feynman diagrams (matrix products instead are used), and "re-sums" into an equilibrium average that may lend itself to numerical procedures. We calculate the universal current in the interacting resonant level model (IRLM) at finite bias V to first order in Coulomb repulsion U for all V and temperatures. We find that the bias, like the temperature, cuts off low-energy processes. In the IRLM, this implies a power-law decay of the current at large V (also recently observed by Boulat and Saleur at some finite value of U ).PACS numbers: 73.63. Kv, 72.15.Qm, 72.10.Fk Impurity models describe mesoscopic quantum objects in contact with large conducting leads. Quantum dots are much-studied examples, and experiments in a bias voltage [1] lead to accurate descriptions of their nonequilibrium steady-state properties. In impurity models, such states are of high interest as they pose the theoretical challenge of capturing the effect of non-equilibrium in a truly quantum system, yet they constitute the simplest non-equilibrium situation, where properties are time independent. At low energies, interactions between the leads' Landau quasi-particles and the impurity occur in the s-wave channel, and the spectrum may be linearised around the two Fermi points. Then, the universal behaviour is described by free massless fermions on the half line (bulk conformal field theory) with a non-conformal boundary interaction at the end-point. Many methods are known for studying equilibrium behaviors in such models, but understanding and accessing properties of non-equilibrium steady states is still a much harder task. Wilson's picture does not apply, and, for instance, the effect of a bias on low-energy processes (which determines the large-bias current) is still under study [2,3].Two calculational schemes exist:the real-time Schwinger-Keldysh formulation and the scattering-state Lippman-Schwinger formulation. The former relies on the infinite extent of the half line in order to absorb the energy released by relaxation from an appropriate "uninteracting" density matrix to the steady state. It requires relaxation mechanisms (see, e.g. [3]), whose absence may lead to pathologies in perturbative expansions [4]. The latter describes directly the expected end result just from "how the state looks" asymptotically far from the impurity. Hershfield's Y operator [5] (see also the studies [3,6,7]) gives a "steady-state density matrix" that encodes these scattering states. This is interesting, since a non-equilibrium steady state is not described by the usual density matrix, but it is still hard to apply to interacting systems. Exact results for integable models occur when the exact quasi-particles do not mix the bias...