diff --git a/src/TriangularSolve.jl b/src/TriangularSolve.jl
index 153f940..ba8e3e9 100644
--- a/src/TriangularSolve.jl
+++ b/src/TriangularSolve.jl
@@ -932,6 +932,58 @@ function ldiv!(
   return C
 end
 
+# Vector right-hand sides reuse the matrix kernels: the left-division paths
+# already operate on `transpose(A)`, and for a vector `b` the required 1×n
+# transposed form is just `transpose(b)` — no reshape, no allocation. A single
+# right-hand side runs the kernels' scalar row-remainder, which beats BLAS
+# trsv up to the cutoff below (1.2-2.4x measured) but loses to trsv's blocked
+# sweep once the triangle falls out of L2, so larger solves keep the
+# LinearAlgebra path.
+const VECTOR_RHS_CUTOFF = 128
+
+for (wrap, dispatch, UNIT) in (
+  (:LowerTriangular, :div_dispatch!, false),
+  (:UnitLowerTriangular, :div_dispatch!, true),
+  (:UpperTriangular, :div_dispatch_L!, false),
+  (:UnitUpperTriangular, :div_dispatch_L!, true)
+)
+  @eval begin
+    function ldiv!(
+      U::$wrap{T,<:StridedMatrix{T}},
+      b::StridedVector{T},
+      ::Val{thread} = Val(true)
+    ) where {T<:Union{Float32,Float64},thread}
+      length(b) > VECTOR_RHS_CUTOFF && return LinearAlgebra.ldiv!(U, b)
+      nt = thread ? _nthreads() : static(1)
+      $dispatch(
+        transpose(b),
+        transpose(b),
+        transpose(parent(U)),
+        nt,
+        Val($UNIT)
+      )
+      return b
+    end
+    function ldiv!(
+      c::StridedVector{T},
+      U::$wrap{T,<:StridedMatrix{T}},
+      b::StridedVector{T},
+      ::Val{thread} = Val(true)
+    ) where {T<:Union{Float32,Float64},thread}
+      length(b) > VECTOR_RHS_CUTOFF && return LinearAlgebra.ldiv!(c, U, b)
+      nt = thread ? _nthreads() : static(1)
+      $dispatch(
+        transpose(c),
+        transpose(b),
+        transpose(parent(U)),
+        nt,
+        Val($UNIT)
+      )
+      return c
+    end
+  end
+end
+
 ldiv!(A, B, ::Val = Val(true)) = LinearAlgebra.ldiv!(A, B)
 ldiv!(Y, A, B, ::Val = Val(true)) = LinearAlgebra.ldiv!(Y, A, B)
 rdiv!(A, B, ::Val = Val(true)) = LinearAlgebra.rdiv!(A, B)
diff --git a/test/runtests.jl b/test/runtests.jl
index 0897499..0a492ba 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -133,6 +133,40 @@ end
       }
     end
   end
+  @testset "vector right-hand sides" begin
+    # sizes straddling VECTOR_RHS_CUTOFF cover both the kernel path and the
+    # LinearAlgebra deferral
+    for T ∈ (Float64, Float32),
+      n ∈ (1, 2, 5, 8, 16, 33, 64, 127, 128, 129, 200, 500)
+
+      P = rand(T, n, n) + T(n) * I
+      b = rand(T, n)
+      for wrap ∈ (
+          UpperTriangular,
+          UnitUpperTriangular,
+          LowerTriangular,
+          UnitLowerTriangular
+        ),
+        thread ∈ (Val(false), Val(true))
+
+        U = wrap(P)
+        x = TriangularSolve.ldiv!(U, copy(b), thread)
+        @test x ≈ Matrix(U) \ b rtol = sqrt(eps(T)) * n
+        c = similar(b)
+        @test TriangularSolve.ldiv!(c, U, copy(b), thread) ≈ x
+      end
+    end
+    F = lu!(rand(100, 100) + 100I).factors
+    bf = rand(100)
+    Uf = UpperTriangular(F)
+    @test TriangularSolve.ldiv!(Uf, copy(bf), Val(false)) ≈ Matrix(Uf) \ bf
+    xf = copy(bf)
+    for thread ∈ (Val(false), Val(true))
+      TriangularSolve.ldiv!(Uf, xf, thread)
+      xf .= bf
+      @test iszero(@allocated TriangularSolve.ldiv!(Uf, xf, thread))
+    end
+  end
   @testset "dimension mismatch throws" begin
     @test_throws DimensionMismatch TriangularSolve.rdiv!(
       rand(4, 8), LowerTriangular(rand(6, 6) + 6I), Val(false)