diff --git a/Sources/DeepLearning/Layer.swift b/Sources/DeepLearning/Layer.swift
index 7c4cf1d23..28116fe5f 100644
--- a/Sources/DeepLearning/Layer.swift
+++ b/Sources/DeepLearning/Layer.swift
@@ -1314,3 +1314,44 @@ public struct Reshape<Scalar: TensorFlowFloatingPoint>: Layer {
         return input.reshaped(toShape: shape)
     }
 }
+
+/// An input to a recurrent neural network.
+public struct RNNInput<TimeStepInput: Differentiable, State: Differentiable>: Differentiable {
+    /// The input at the current time step.
+    public var timeStepInput: TimeStepInput
+    /// The previous state.
+    public var previousState: State
+
+    @differentiable
+    public init(timeStepInput: TimeStepInput, previousState: State) {
+        self.timeStepInput = timeStepInput
+        self.previousState = previousState
+    }
+}
+
+/// A recurrent neural network cell.
+public protocol RNNCell: Layer where Input == RNNInput<TimeStepInput, State> {
+    /// The input at a time step.
+    associatedtype TimeStepInput: Differentiable
+    /// The state that may be preserved across time steps.
+    typealias State = Output
+    /// The zero state.
+    var zeroState: State { get }
+}
+
+public extension RNNCell {
+    /// Returns the new state obtained from applying the RNN cell to the input at the current time
+    /// step and the previous state.
+    ///
+    /// - Parameters:
+    ///   - timeStepInput: The input at the current time step.
+    ///   - previousState: The previous state of the RNN cell.
+    ///   - context: The contextual information for the layer application, e.g. the current learning
+    ///     phase.
+    /// - Returns: The output.
+    @differentiable
+    func applied(to timeStepInput: TimeStepInput, previous: State, in context: Context) -> State {
+        return applied(to: Input(timeStepInput: timeStepInput, previousState: previous),
+                       in: context)
+    }
+}