diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
new file mode 100644
index 0000000..a37d889
--- /dev/null
+++ b/.github/workflows/test.yml
@@ -0,0 +1,51 @@
+name: test
+
+on:
+  push:
+    branches: [main, master]
+  pull_request:
+  workflow_dispatch:
+
+concurrency:
+  group: test-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    name: build + pytest (Julia ${{ matrix.julia }}, py${{ matrix.python }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        julia: ["1.12.6"]
+        python: ["3.12"]
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: ${{ matrix.julia }}
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+
+      - name: Resolve Julia bindir
+        run: echo "JULIA_BINDIR=$(julia -e 'print(Sys.BINDIR)')" >> "$GITHUB_ENV"
+
+      - name: Build libjl2py
+        run: make -C out/jl2py/csrc
+
+      - name: Install Python deps
+        run: python -m pip install --upgrade pip "numpy>=2.0" pytest
+
+      # The `jl` fixture (tests/conftest.py) provides a session-scoped Julia.
+      # Memory-wrap tests that need the experimental libjulia-internal build
+      # skip themselves in this default build.
+      - name: Run tests
+        env:
+          JL2PY_LIB_PATH: ${{ github.workspace }}/out/jl2py/csrc/build/libjl2py.so
+        run: python -m pytest out/jl2py/tests -v
diff --git a/out/jl2py/tests/conftest.py b/out/jl2py/tests/conftest.py
new file mode 100644
index 0000000..5b2b3b1
--- /dev/null
+++ b/out/jl2py/tests/conftest.py
@@ -0,0 +1,18 @@
+"""
+pytest configuration for jl2py tests.
+"""
+import os
+import sys
+
+# Ensure src/ is on path
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "src"))
+
+import pytest
+import jl2py
+
+
+@pytest.fixture(scope="session")
+def jl():
+    """Session-scoped jl2py runtime instance."""
+    julia_bindir = os.environ.get("JULIA_BINDIR") or None
+    return jl2py.init(julia_bindir=julia_bindir)
diff --git a/out/jl2py/tests/test_array.py b/out/jl2py/tests/test_array.py
index be8d69c..fc3e8fc 100644
--- a/out/jl2py/tests/test_array.py
+++ b/out/jl2py/tests/test_array.py
@@ -10,17 +10,13 @@
 import sys, os
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "src"))
 
+import pytest
 import numpy as np
 import jl2py
 from jl2py.array import JuliaArray, JuliaMemory
 from jl2py.types import JuliaInt64, JuliaFloat64
 
 
-def setup():
-    julia_bindir = os.environ.get("JULIA_BINDIR", os.path.expanduser("~/content/workspace/usr/bin"))
-    return jl2py.init(julia_bindir=julia_bindir)
-
-
 def has_internal():
     """Detect whether libjl2py was built with JL2PY_HAS_INTERNAL (libjulia-internal).
     Memory wrap/alloc functions require internal symbols; they return NULL without it."""
@@ -163,6 +159,7 @@ def test_int_array(jl):
 
 # ── Memory tests ─────────────────────────────────────────────
 
+@pytest.mark.skipif(not has_internal(), reason="requires JL2PY_INTERNAL build")
 def test_memory_from_numpy(jl):
     """Wrap NumPy as Julia Memory{T}."""
     np_arr = np.array([1.0, 2.0, 3.0, 4.0], dtype=np.float64)
@@ -173,6 +170,7 @@ def test_memory_from_numpy(jl):
     print(f"[PASS] numpy -> Memory: len={len(mem)}")
 
 
+@pytest.mark.skipif(not has_internal(), reason="requires JL2PY_INTERNAL build")
 def test_memory_numpy_view(jl):
     """NumPy view of Julia Memory."""
     np_arr = np.array([5.0, 6.0, 7.0], dtype=np.float64)
@@ -183,6 +181,7 @@ def test_memory_numpy_view(jl):
     print(f"[PASS] Memory -> numpy: {list(np_view)}")
 
 
+@pytest.mark.skipif(not has_internal(), reason="requires JL2PY_INTERNAL build")
 def test_memory_shared_write(jl):
     """Memory shares buffer with NumPy."""
     np_arr = np.array([1.0, 2.0, 3.0], dtype=np.float64)
@@ -222,42 +221,3 @@ def test_array_gc_safety(jl):
 
 # ── Main ─────────────────────────────────────────────────────
 
-def main():
-    print("=" * 60)
-    print("jl2py Phase 4 tests: Array + Memory interop")
-    print("=" * 60)
-
-    jl = setup()
-
-    test_array_from_eval(jl)
-    test_array_getitem_0based(jl)
-    test_array_setitem(jl)
-    test_array_jl_getindex(jl)
-    test_array_iter(jl)
-    test_array_contains(jl)
-    test_numpy_to_julia(jl)
-    test_julia_to_numpy(jl)
-    test_zero_copy_shared_memory(jl)
-    test_2d_array(jl)
-    test_int_array(jl)
-    # Memory-from-array works in all modes (struct field access only)
-    test_array_backing_memory(jl)
-
-    # Memory wrap/alloc require libjulia-internal (jl_alloc_genericmemory, jl_ptr_to_genericmemory)
-    if has_internal():
-        test_memory_from_numpy(jl)
-        test_memory_numpy_view(jl)
-        test_memory_shared_write(jl)
-        print(f"[INFO] Memory wrap/alloc tests: PASSED (experimental build)")
-    else:
-        print(f"[SKIP] Memory wrap/alloc tests: libjl2py built without JL2PY_HAS_INTERNAL")
-
-    test_array_gc_safety(jl)
-
-    print()
-    print("All Phase 4 tests passed!")
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()
diff --git a/out/jl2py/tests/test_exceptions.py b/out/jl2py/tests/test_exceptions.py
new file mode 100644
index 0000000..2a643c7
--- /dev/null
+++ b/out/jl2py/tests/test_exceptions.py
@@ -0,0 +1,60 @@
+"""Typed Julia-error mapping: a Julia exception surfaces as the mapped Python type
+(see jl2py.exceptions.EXCEPTION_MAP) and carries its message.
+
+Run:
+    JULIA_BINDIR=<julia>/bin python tests/test_exceptions.py
+"""
+import os
+import sys
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "src"))
+
+import jl2py
+from jl2py.exceptions import (
+    JuliaBoundsError,
+    JuliaDomainError,
+    JuliaError,
+    JuliaMethodError,
+    JuliaUndefVarError,
+)
+
+
+def _expect(jl, expr, exc_type, py_base):
+    """Evaluating `expr` must raise `exc_type`, which is also a `py_base`."""
+    try:
+        jl.eval(expr)
+    except JuliaError as e:
+        assert isinstance(e, exc_type), f"{expr!r}: expected {exc_type.__name__}, got {type(e).__name__}: {e}"
+        assert isinstance(e, py_base), f"{exc_type.__name__} should also be a {py_base.__name__}"
+        assert str(e)
+        print(f"[PASS] {expr!r} -> {type(e).__name__} ({py_base.__name__})")
+        return
+    raise AssertionError(f"{expr!r}: expected an exception, none raised")
+
+
+def test_boundserror(jl):
+    _expect(jl, "[1, 2, 3][10]", JuliaBoundsError, IndexError)
+
+
+def test_domainerror(jl):
+    _expect(jl, "sqrt(-1.0)", JuliaDomainError, ValueError)
+
+
+def test_methoderror(jl):
+    _expect(jl, '1 + "a"', JuliaMethodError, TypeError)
+
+
+def test_undefvarerror(jl):
+    _expect(jl, "this_global_does_not_exist_zzz", JuliaUndefVarError, NameError)
+
+
+def test_base_error_carries_message(jl):
+    try:
+        jl.eval('error("boom-xyz")')
+    except JuliaError as e:
+        assert "boom-xyz" in str(e)
+        print("[PASS] base JuliaError carries its message")
+        return
+    raise AssertionError("expected an exception, none raised")
+
+
diff --git a/out/jl2py/tests/test_identity.py b/out/jl2py/tests/test_identity.py
new file mode 100644
index 0000000..3ee5d72
--- /dev/null
+++ b/out/jl2py/tests/test_identity.py
@@ -0,0 +1,32 @@
+"""Round-trip identity: a Julia value crossing to Python and back keeps its
+`jl_value_t*` identity (PRD §4.3 — the ref system must not re-box an existing
+Julia value).
+
+Run:
+    JULIA_BINDIR=<julia>/bin python tests/test_identity.py
+"""
+import os
+import sys
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "src"))
+
+import jl2py
+
+
+def test_pointer_identity_roundtrip(jl):
+    """A heap value passed Py->Jl->Py comes back as the same jl_value_t*, not a copy."""
+    arr = jl.eval("[1, 2, 3]")  # heap-allocated mutable Julia Array
+    identity = jl.eval("x -> x")
+    same = identity(arr)
+    assert same.ptr == arr.ptr
+    print("[PASS] pointer identity preserved across Py->Jl->Py")
+
+
+def test_objectid_stable(jl):
+    """objectid of a wrapped value is stable across repeated boundary crossings."""
+    x = jl.eval("Ref(42)")
+    objectid = jl.eval("objectid")
+    assert int(objectid(x)) == int(objectid(x))
+    print("[PASS] objectid stable for a wrapped value")
+
+
diff --git a/out/jl2py/tests/test_juliaany.py b/out/jl2py/tests/test_juliaany.py
index 18be74e..40e5661 100644
--- a/out/jl2py/tests/test_juliaany.py
+++ b/out/jl2py/tests/test_juliaany.py
@@ -14,11 +14,6 @@
 from jl2py.types import JuliaAny, LayoutFlags, register_wrapper, unregister_wrapper
 
 
-def setup():
-    julia_bindir = os.environ.get("JULIA_BINDIR", os.path.expanduser("~/content/workspace/usr/bin"))
-    return jl2py.init(julia_bindir=julia_bindir)
-
-
 def test_juliaany_fallback(jl):
     jl.eval("struct TestImmBits; x::Float64; y::Int64; end")
     v = jl.eval("TestImmBits(3.14, 42)")
@@ -156,27 +151,3 @@ def span(self):
     unregister_wrapper("UnitRange")
 
 
-def main():
-    print("=" * 60)
-    print("jl2py JuliaAny + registry + layout tests")
-    print("=" * 60)
-
-    jl = setup()
-
-    test_juliaany_fallback(jl)
-    test_layout_isbits(jl)
-    test_layout_mutable_pointers(jl)
-    test_layout_singleton(jl)
-    test_jl_getfield(jl)
-    test_register_wrapper(jl)
-    test_register_parametric(jl)
-    test_juliaany_repr(jl)
-    test_julia_type_delegation(jl)
-
-    print()
-    print("All JuliaAny tests passed!")
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()
diff --git a/out/jl2py/tests/test_kwargs_slice.py b/out/jl2py/tests/test_kwargs_slice.py
index 71df238..a58ba82 100644
--- a/out/jl2py/tests/test_kwargs_slice.py
+++ b/out/jl2py/tests/test_kwargs_slice.py
@@ -21,11 +21,6 @@
 from jl2py.array import JuliaArray
 
 
-def setup():
-    julia_bindir = os.environ.get("JULIA_BINDIR", os.path.expanduser("~/content/workspace/usr/bin"))
-    return jl2py.init(julia_bindir=julia_bindir)
-
-
 # ── kwargs round-trip ────────────────────────────────────────
 
 def test_kwargs_user_function(jl):
@@ -96,26 +91,3 @@ def test_slice_matches_jl_getindex(jl):
 
 # ── Main ─────────────────────────────────────────────────────
 
-def main():
-    print("=" * 60)
-    print("jl2py regression tests: kwargs round-trip + slice->Range")
-    print("=" * 60)
-
-    jl = setup()
-
-    test_kwargs_user_function(jl)
-    test_kwargs_multiple(jl)
-    test_kwargs_base_round(jl)
-
-    test_slice_unit_range(jl)
-    test_slice_full(jl)
-    test_slice_step(jl)
-    test_slice_matches_jl_getindex(jl)
-
-    print()
-    print("All kwargs + slice regression tests passed!")
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()
diff --git a/out/jl2py/tests/test_smoke.py b/out/jl2py/tests/test_smoke.py
index 6cd878c..d1bcd7a 100644
--- a/out/jl2py/tests/test_smoke.py
+++ b/out/jl2py/tests/test_smoke.py
@@ -110,25 +110,3 @@ def test_exception_handling(jl):
         print(f"[PASS] Exception handling: caught {e!r}")
 
 
-def main():
-    print("=" * 60)
-    print("jl2py smoke test")
-    print("=" * 60)
-
-    jl = test_init()
-    test_eval(jl)
-    test_module_access(jl)
-    test_function_call(jl)
-    test_repr_str(jl)
-    test_bool(jl)
-    test_gc_rooting(jl)
-    test_exception_handling(jl)
-
-    print()
-    print("All smoke tests passed!")
-
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()
diff --git a/out/jl2py/tests/test_struct.py b/out/jl2py/tests/test_struct.py
index d7c9614..5fd0923 100644
--- a/out/jl2py/tests/test_struct.py
+++ b/out/jl2py/tests/test_struct.py
@@ -11,11 +11,27 @@
 import sys, os, gc as python_gc, ctypes
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "src"))
 
+import pytest
 import jl2py
 from jl2py.types import JuliaFloat64, JuliaInt64, JuliaString, wrap_julia_value
 from jl2py._lib import get_lib
 
 
+@pytest.fixture(scope="module", autouse=True)
+def _define_structs(jl):
+    """Define the 8 test structs once per module."""
+    jl.eval("""
+        struct ImmPoint; x::Float64; y::Float64; end
+        mutable struct MutPoint; x::Float64; y::Float64; end
+        struct Parametric{T}; val::T; label::String; end
+        mutable struct MutParametric{T}; val::T; label::String; end
+        struct Nested{A,B}; a::A; b::B; end
+        struct Empty; end
+        struct SingleField; x::Int64; end
+        mutable struct Counter; n::Int64; end
+    """)
+
+
 def apply_type(tc, *params):
     """
     Python-side parametric type application using jl2py C API.
@@ -44,24 +60,6 @@ def apply_type(tc, *params):
     return wrap_julia_value(result)
 
 
-def setup():
-    julia_bindir = os.environ.get("JULIA_BINDIR", os.path.expanduser("~/content/workspace/usr/bin"))
-    jl = jl2py.init(julia_bindir=julia_bindir)
-
-    # Define test structs once
-    jl.eval("""
-        struct ImmPoint; x::Float64; y::Float64; end
-        mutable struct MutPoint; x::Float64; y::Float64; end
-        struct Parametric{T}; val::T; label::String; end
-        mutable struct MutParametric{T}; val::T; label::String; end
-        struct Nested{A,B}; a::A; b::B; end
-        struct Empty; end
-        struct SingleField; x::Int64; end
-        mutable struct Counter; n::Int64; end
-    """)
-    return jl
-
-
 # ══════════════════════════════════════════════════════════════
 # Immutable structs
 # ══════════════════════════════════════════════════════════════
@@ -484,48 +482,3 @@ def test_gc_stress_mutation_under_pressure(jl):
 # Main
 # ══════════════════════════════════════════════════════════════
 
-def main():
-    print("=" * 60)
-    print("jl2py struct tests: immutable, mutable, parametric, GC stress")
-    print("=" * 60)
-
-    jl = setup()
-
-    # Immutable
-    test_immutable_construct_and_fields(jl)
-    test_immutable_via_constructor_dispatch(jl)
-    test_immutable_cannot_mutate(jl)
-    test_immutable_repr(jl)
-    test_empty_struct(jl)
-    test_single_field_struct(jl)
-
-    # Mutable
-    test_mutable_construct_and_fields(jl)
-    test_mutable_setfield(jl)
-    test_mutable_via_constructor(jl)
-    test_mutable_counter(jl)
-
-    # Parametric
-    test_parametric_float64(jl)
-    test_parametric_int64(jl)
-    test_parametric_string(jl)
-    test_parametric_constructor_chain(jl)
-    test_parametric_apply_type2(jl)
-    test_parametric_apply_type_generic(jl)
-    test_mutable_parametric(jl)
-    test_nested_parametric(jl)
-
-    # GC stress
-    test_gc_stress_explicit_collection(jl)
-    test_gc_stress_interleaved_dispatch(jl)
-    test_gc_stress_create_and_drop(jl)
-    test_gc_stress_nested_parametric(jl)
-    test_gc_stress_mutation_under_pressure(jl)
-
-    print()
-    print("All struct tests passed!")
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()
diff --git a/out/jl2py/tests/test_types_containers.py b/out/jl2py/tests/test_types_containers.py
index daf8168..32ee642 100644
--- a/out/jl2py/tests/test_types_containers.py
+++ b/out/jl2py/tests/test_types_containers.py
@@ -18,11 +18,6 @@
 from jl2py.containers import JuliaTuple, JuliaDict
 
 
-def setup():
-    julia_bindir = os.environ.get("JULIA_BINDIR", os.path.expanduser("~/content/workspace/usr/bin"))
-    return jl2py.init(julia_bindir=julia_bindir)
-
-
 # ── Typed wrapping ───────────────────────────────────────────
 
 def test_typed_wrapping(jl):
@@ -362,27 +357,3 @@ def test_constructor_gc_safety(jl):
 
 # ── Main ─────────────────────────────────────────────────────
 
-def main():
-    print("=" * 60)
-    print("jl2py Phase 3 tests: type system + containers")
-    print("=" * 60)
-
-    jl = setup()
-
-    test_typed_wrapping(jl)
-    test_numeric_ops(jl)
-    test_string_ops(jl)
-    test_tuple_ops(jl)
-    test_dict_ops(jl)
-    test_struct_fields(jl)
-    test_explicit_constructors(jl)
-    test_type_constructors_via_dispatch(jl)
-    test_constructor_gc_safety(jl)
-
-    print()
-    print("All Phase 3 tests passed!")
-    jl2py.shutdown()
-
-
-if __name__ == "__main__":
-    main()