diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/test_gateway_integration.py b/tests/test_gateway_integration.py
new file mode 100644
index 0000000..85ad685
--- /dev/null
+++ b/tests/test_gateway_integration.py
@@ -0,0 +1,643 @@
+"""Integration tests for semantic object pipeline wiring in the gateway.
+
+Tests the Session ↔ Segmenter ↔ ObjectStore ↔ ContextAssembler integration:
+  - Session.__init__ creates segmenter, object_store, _segmented_objects
+  - _preprocess segments messages and stores objects in ObjectStore
+  - memory_query phantom tool resolves through ContextAssembler
+  - Graceful degradation when embedder is unavailable (DummyEmbedder)
+"""
+
+from __future__ import annotations
+
+import asyncio
+import copy
+from pathlib import Path
+from tempfile import TemporaryDirectory
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from mnemosyne.context_assembler import ContextAssembler, MicroFaultResult
+from mnemosyne.gateway import Session, _run_async
+from mnemosyne.object_store import DummyEmbedder, InMemoryBackend, ObjectStore
+from mnemosyne.phantom import PhantomCall, _handle_phantom_call, inject_phantom_results
+from mnemosyne.segmenter import SegmentedObject, Segmenter
+
+
+# ── Fixtures ─────────────────────────────────────────────────────────────
+
+
+@pytest.fixture
+def tmp_log_dir():
+    with TemporaryDirectory() as d:
+        yield Path(d)
+
+
+@pytest.fixture
+def session(tmp_log_dir):
+    return Session("integ01", tmp_log_dir)
+
+
+def _user_msg(text: str) -> dict:
+    return {"role": "user", "content": text}
+
+
+def _assistant_msg(text: str) -> dict:
+    return {"role": "assistant", "content": text}
+
+
+def _tool_use_block(tool_id: str, name: str, input_data: dict) -> dict:
+    return {"type": "tool_use", "id": tool_id, "name": name, "input": input_data}
+
+
+def _tool_result_block(tool_id: str, content: str) -> dict:
+    return {"type": "tool_result", "tool_use_id": tool_id, "content": content}
+
+
+# ── Session initialization ───────────────────────────────────────────────
+
+
+class TestSessionSemanticAttributes:
+    """Session.__init__ must create segmenter, object_store, and related attrs."""
+
+    def test_session_has_segmenter(self, session):
+        assert hasattr(session, "segmenter")
+        assert isinstance(session.segmenter, Segmenter)
+
+    def test_session_has_object_store(self, session):
+        assert hasattr(session, "object_store")
+        assert isinstance(session.object_store, ObjectStore)
+
+    def test_session_has_segmented_objects_list(self, session):
+        assert hasattr(session, "_segmented_objects")
+        assert isinstance(session._segmented_objects, list)
+        assert len(session._segmented_objects) == 0
+
+    def test_session_has_context_assembler_slot(self, session):
+        assert hasattr(session, "context_assembler")
+        # Initially None — set by create_app after helper_llm is resolved
+        assert session.context_assembler is None
+
+    def test_object_store_uses_in_memory_backend(self, session):
+        assert isinstance(session.object_store._backend, InMemoryBackend)
+
+    def test_object_store_has_embedder(self, session):
+        """ObjectStore should have an embedder (DummyEmbedder fallback if no sentence-transformers)."""
+        assert session.object_store._embedder is not None
+
+    def test_segmenter_per_session(self, tmp_log_dir):
+        s1 = Session("sess_a", tmp_log_dir)
+        s2 = Session("sess_b", tmp_log_dir)
+        assert s1.segmenter is not s2.segmenter
+
+    def test_object_store_per_session(self, tmp_log_dir):
+        s1 = Session("sess_a", tmp_log_dir)
+        s2 = Session("sess_b", tmp_log_dir)
+        assert s1.object_store is not s2.object_store
+
+
+# ── Segmenter integration ───────────────────────────────────────────────
+
+
+class TestSegmenterIntegration:
+    """Test that the segmenter correctly processes messages into objects."""
+
+    def test_segment_simple_conversation(self, session):
+        messages = [
+            _user_msg("Please read the auth module"),
+            _assistant_msg(
+                "I'll read the auth module for you. The authentication system uses "
+                "JWT tokens with RS256 signing. The main entry point is auth.py which "
+                "handles token validation, refresh, and revocation."
+            ),
+        ]
+        result = session.segmenter.segment_incremental(
+            messages, session._segmented_objects, start_turn=0
+        )
+        assert len(result) > 0
+        assert all(isinstance(obj, SegmentedObject) for obj in result)
+
+    def test_segment_incremental_extends(self, session):
+        msgs1 = [
+            _user_msg("What is the project structure?"),
+            _assistant_msg("The project has src/, tests/, and docs/ directories."),
+        ]
+        result1 = session.segmenter.segment_incremental(
+            msgs1, session._segmented_objects, start_turn=0
+        )
+        session._segmented_objects = result1
+
+        msgs2 = msgs1 + [
+            _user_msg("Tell me about the src/ directory"),
+            _assistant_msg("The src/ directory contains the main application code."),
+        ]
+        result2 = session.segmenter.segment_incremental(
+            msgs2, session._segmented_objects, start_turn=0
+        )
+        assert len(result2) >= len(result1)
+
+    def test_segment_with_tool_results(self, session):
+        messages = [
+            _user_msg("Read the config file"),
+            {
+                "role": "assistant",
+                "content": [
+                    {"type": "text", "text": "I'll read the config file."},
+                    _tool_use_block("tu_1", "Read", {"file_path": "/app/config.py"}),
+                ],
+            },
+            {
+                "role": "user",
+                "content": [
+                    _tool_result_block(
+                        "tu_1",
+                        "# config.py\nDATABASE_URL = 'postgres://localhost/mydb'\nDEBUG = True",
+                    ),
+                ],
+            },
+        ]
+        result = session.segmenter.segment_incremental(
+            messages, session._segmented_objects, start_turn=0
+        )
+        assert len(result) > 0
+        # At least one object should be file_context or tool_result
+        types = {obj.object_type for obj in result}
+        assert types & {"file_context", "tool_result", "conversation_phase"}
+
+
+# ── ObjectStore integration ──────────────────────────────────────────────
+
+
+class TestObjectStoreIntegration:
+    """Test that segmented objects are stored in ObjectStore via _run_async."""
+
+    def test_store_segmented_object(self, session):
+        """Manually store a segmented object and verify it's in the store."""
+        seg_obj = SegmentedObject(
+            content="The auth module uses JWT tokens for authentication.",
+            object_type="file_context",
+            source_tool="Read",
+            source_key="/app/auth.py",
+            stub="[file_context: auth.py — JWT authentication]",
+            turn_start=1,
+            turn_end=1,
+            token_estimate=12,
+            key_entities=["auth.py", "JWT"],
+            tags=["auth"],
+        )
+
+        stored = _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content=seg_obj.content,
+                object_type=seg_obj.object_type,
+                source_tool=seg_obj.source_tool,
+                source_key=seg_obj.source_key,
+                stub=seg_obj.stub,
+                tags=seg_obj.tags,
+                key_entities=seg_obj.key_entities,
+                turn=seg_obj.turn_start,
+            )
+        )
+
+        assert stored is not None
+        assert stored.session_id == session.id
+        assert stored.object_type == "file_context"
+        assert stored.content_full == seg_obj.content
+        assert stored.source_tool == "Read"
+        assert stored.source_key == "/app/auth.py"
+        assert len(stored.embedding) == 384  # DummyEmbedder produces 384-dim
+
+    def test_store_multiple_objects_and_search(self, session):
+        """Store multiple objects and verify semantic search works."""
+        contents = [
+            ("The database uses PostgreSQL with connection pooling.", "file_context"),
+            ("Authentication is handled by JWT tokens with RS256.", "file_context"),
+            ("The CI pipeline runs pytest and mypy on every PR.", "conversation_phase"),
+        ]
+
+        for content, obj_type in contents:
+            _run_async(
+                session.object_store.store_object(
+                    session_id=session.id,
+                    content=content,
+                    object_type=obj_type,
+                )
+            )
+
+        # Search for auth-related content
+        results = _run_async(
+            session.object_store.semantic_search(session.id, "authentication JWT", limit=3)
+        )
+        assert len(results) > 0
+        # The auth-related object should be in results
+        found_auth = any("JWT" in obj.content_full for obj, _score in results)
+        assert found_auth
+
+    def test_run_async_from_sync_context(self):
+        """_run_async should work from a plain sync context."""
+
+        async def simple_coro():
+            return 42
+
+        result = _run_async(simple_coro())
+        assert result == 42
+
+
+# ── memory_query phantom tool resolution ─────────────────────────────────
+
+
+class TestMemoryQueryResolution:
+    """Test that memory_query resolves through ContextAssembler."""
+
+    def test_memory_query_with_context_assembler(self, session):
+        """memory_query should resolve through ContextAssembler when available."""
+        # Store some content first
+        _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content="The server runs on port 8080 with TLS enabled.",
+                object_type="file_context",
+                source_tool="Read",
+                source_key="/app/server.py",
+                tags=["server", "config"],
+                key_entities=["port 8080", "TLS"],
+            )
+        )
+
+        # Create ContextAssembler without HelperLLM (fallback mode)
+        assembler = ContextAssembler(session.object_store, helper_llm=None)
+
+        call = PhantomCall(
+            name="memory_query",
+            tool_use_id="toolu_mq_001",
+            input={"question": "What port does the server run on?"},
+        )
+
+        result = _handle_phantom_call(
+            call,
+            page_store=None,
+            context_assembler=assembler,
+            session_id=session.id,
+        )
+
+        # Should NOT contain the pending placeholder
+        assert "[memory_query:pending]" not in result
+        # Should contain the resolved answer
+        assert "[memory_query resolved:" in result
+
+    def test_memory_query_with_scope(self, session):
+        """memory_query with scope should narrow the search."""
+        _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content="Database config: host=localhost, port=5432, name=mydb",
+                object_type="file_context",
+                source_key="/app/db.py",
+                tags=["database"],
+            )
+        )
+        _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content="Server config: host=0.0.0.0, port=8080",
+                object_type="file_context",
+                source_key="/app/server.py",
+                tags=["server"],
+            )
+        )
+
+        assembler = ContextAssembler(session.object_store, helper_llm=None)
+
+        call = PhantomCall(
+            name="memory_query",
+            tool_use_id="toolu_mq_002",
+            input={
+                "question": "What is the database port?",
+                "scope": "database config",
+            },
+        )
+
+        result = _handle_phantom_call(
+            call,
+            page_store=None,
+            context_assembler=assembler,
+            session_id=session.id,
+        )
+
+        assert "[memory_query:pending]" not in result
+        assert "[memory_query resolved:" in result
+
+    def test_memory_query_fallback_without_assembler(self):
+        """memory_query without context_assembler should return pending placeholder."""
+        call = PhantomCall(
+            name="memory_query",
+            tool_use_id="toolu_mq_003",
+            input={"question": "What auth library is used?"},
+        )
+
+        result = _handle_phantom_call(call, page_store=None)
+        assert "[memory_query:pending]" in result
+        assert "What auth library is used?" in result
+
+    def test_memory_query_no_results(self, session):
+        """memory_query with no matching content should still resolve gracefully."""
+        assembler = ContextAssembler(session.object_store, helper_llm=None)
+
+        call = PhantomCall(
+            name="memory_query",
+            tool_use_id="toolu_mq_004",
+            input={"question": "What is the meaning of life?"},
+        )
+
+        result = _handle_phantom_call(
+            call,
+            page_store=None,
+            context_assembler=assembler,
+            session_id=session.id,
+        )
+
+        # Should resolve (not pending) even with no results
+        assert "[memory_query:pending]" not in result
+        assert "[memory_query resolved:" in result
+
+    def test_memory_query_with_mock_helper_llm(self, session):
+        """memory_query with a mocked HelperLLM should return the LLM's answer."""
+        _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content="The application uses bcrypt for password hashing with cost factor 12.",
+                object_type="file_context",
+                source_key="/app/auth.py",
+                key_entities=["bcrypt", "password hashing"],
+            )
+        )
+
+        # Mock HelperLLM
+        mock_helper = MagicMock()
+        mock_helper.answer_micro_fault = AsyncMock(
+            return_value="The application uses bcrypt with cost factor 12."
+        )
+
+        assembler = ContextAssembler(session.object_store, helper_llm=mock_helper)
+
+        call = PhantomCall(
+            name="memory_query",
+            tool_use_id="toolu_mq_005",
+            input={"question": "What password hashing is used?"},
+        )
+
+        result = _handle_phantom_call(
+            call,
+            page_store=None,
+            context_assembler=assembler,
+            session_id=session.id,
+        )
+
+        assert "[memory_query:pending]" not in result
+        assert "bcrypt" in result
+        assert "[memory_query resolved:" in result
+        mock_helper.answer_micro_fault.assert_called_once()
+
+    def test_inject_phantom_results_passes_assembler(self, session):
+        """inject_phantom_results should pass context_assembler to handler."""
+        _run_async(
+            session.object_store.store_object(
+                session_id=session.id,
+                content="The API rate limit is 100 requests per minute.",
+                object_type="conversation_phase",
+                key_entities=["rate limit", "100 rpm"],
+            )
+        )
+
+        assembler = ContextAssembler(session.object_store, helper_llm=None)
+
+        phantom_calls = [
+            PhantomCall(
+                name="memory_query",
+                tool_use_id="toolu_inject_001",
+                input={"question": "What is the API rate limit?"},
+            )
+        ]
+
+        messages = [
+            _user_msg("Hello"),
+            {
+                "role": "assistant",
+                "content": [
+                    {"type": "text", "text": "Let me check."},
+                ],
+            },
+        ]
+
+        result_messages = inject_phantom_results(
+            messages,
+            phantom_calls,
+            page_store=None,
+            observe_only=False,
+            context_assembler=assembler,
+            session_id=session.id,
+        )
+
+        # The result should have injected a tool_result
+        # Find the user message with tool_result
+        found_result = False
+        for msg in result_messages:
+            content = msg.get("content", [])
+            if isinstance(content, list):
+                for block in content:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        result_text = block.get("content", "")
+                        if "[memory_query resolved:" in result_text:
+                            found_result = True
+        assert found_result, "Expected resolved memory_query in injected results"
+
+
+# ── DummyEmbedder fallback ───────────────────────────────────────────────
+
+
+class TestDummyEmbedderFallback:
+    """Test graceful degradation when real embedder is unavailable."""
+
+    def test_dummy_embedder_produces_384_dim(self):
+        embedder = DummyEmbedder()
+        vec = embedder.embed("test text")
+        assert len(vec) == 384
+
+    def test_dummy_embedder_deterministic(self):
+        embedder = DummyEmbedder()
+        v1 = embedder.embed("same text")
+        v2 = embedder.embed("same text")
+        assert v1 == v2
+
+    def test_dummy_embedder_different_for_different_text(self):
+        embedder = DummyEmbedder()
+        v1 = embedder.embed("text one")
+        v2 = embedder.embed("text two")
+        assert v1 != v2
+
+    def test_object_store_works_with_dummy_embedder(self):
+        """ObjectStore should function correctly with DummyEmbedder."""
+        backend = InMemoryBackend()
+        embedder = DummyEmbedder()
+        store = ObjectStore(backend, embedder=embedder)
+
+        stored = _run_async(
+            store.store_object(
+                session_id="test_session",
+                content="Test content for embedding",
+                object_type="conversation_phase",
+            )
+        )
+
+        assert len(stored.embedding) == 384
+
+        # Search should work
+        results = _run_async(store.semantic_search("test_session", "test content", limit=5))
+        assert len(results) == 1
+        assert results[0][0].id == stored.id
+
+    @patch("mnemosyne.embedder.try_get_embedder", return_value=None)
+    def test_session_falls_back_to_dummy_embedder(self, mock_try, tmp_log_dir):
+        """When try_get_embedder returns None, Session should use DummyEmbedder."""
+        s = Session("fallback_test", tmp_log_dir)
+        assert s.object_store._embedder is not None
+        assert isinstance(s.object_store._embedder, DummyEmbedder)
+
+
+# ── End-to-end segmentation + storage ────────────────────────────────────
+
+
+class TestEndToEndSegmentationStorage:
+    """Test the full pipeline: messages → segmenter → object_store."""
+
+    def test_segment_and_store_conversation(self, session):
+        """Simulate what _preprocess does: segment messages and store objects."""
+        messages = [
+            _user_msg("Can you read the database configuration?"),
+            _assistant_msg(
+                "I'll read the database configuration file. It contains "
+                "the connection string, pool size, and timeout settings. "
+                "The database is PostgreSQL running on port 5432 with "
+                "a connection pool of 20 connections."
+            ),
+            _user_msg("What about the authentication setup?"),
+            _assistant_msg(
+                "The authentication system uses OAuth 2.0 with JWT tokens. "
+                "Tokens are signed with RS256 and have a 1-hour expiry. "
+                "Refresh tokens are stored in the database with a 30-day TTL."
+            ),
+        ]
+
+        segmented = session.segmenter.segment_incremental(
+            messages, session._segmented_objects, start_turn=0
+        )
+
+        new_count = len(segmented) - len(session._segmented_objects)
+        assert new_count > 0
+
+        new_objects = segmented[len(session._segmented_objects) :]
+        for seg_obj in new_objects:
+            _run_async(
+                session.object_store.store_object(
+                    session_id=session.id,
+                    content=seg_obj.content,
+                    object_type=seg_obj.object_type,
+                    source_tool=seg_obj.source_tool,
+                    source_key=seg_obj.source_key,
+                    stub=seg_obj.stub,
+                    tags=seg_obj.tags,
+                    key_entities=seg_obj.key_entities,
+                    turn=seg_obj.turn_start,
+                )
+            )
+        session._segmented_objects = segmented
+
+        # Verify objects are in the store
+        stored_objects = _run_async(session.object_store.get_session_objects(session.id))
+        assert len(stored_objects) == new_count
+
+        # Verify search works on stored objects
+        results = _run_async(
+            session.object_store.semantic_search(session.id, "database PostgreSQL", limit=5)
+        )
+        assert len(results) > 0
+
+    def test_incremental_segmentation_stores_only_new(self, session):
+        """Incremental segmentation should only store new objects."""
+        msgs1 = [
+            _user_msg("Hello"),
+            _assistant_msg("Hi there! How can I help you today?"),
+        ]
+
+        seg1 = session.segmenter.segment_incremental(
+            msgs1, session._segmented_objects, start_turn=0
+        )
+        for seg_obj in seg1:
+            _run_async(
+                session.object_store.store_object(
+                    session_id=session.id,
+                    content=seg_obj.content,
+                    object_type=seg_obj.object_type,
+                )
+            )
+        session._segmented_objects = seg1
+        count1 = len(seg1)
+
+        # Add more messages
+        msgs2 = msgs1 + [
+            _user_msg("Tell me about the project architecture"),
+            _assistant_msg(
+                "The project follows a layered architecture with "
+                "gateway, service, and repository layers."
+            ),
+        ]
+
+        seg2 = session.segmenter.segment_incremental(
+            msgs2, session._segmented_objects, start_turn=0
+        )
+        new_count = len(seg2) - len(session._segmented_objects)
+
+        # Only store new objects
+        if new_count > 0:
+            new_objects = seg2[len(session._segmented_objects) :]
+            for seg_obj in new_objects:
+                _run_async(
+                    session.object_store.store_object(
+                        session_id=session.id,
+                        content=seg_obj.content,
+                        object_type=seg_obj.object_type,
+                    )
+                )
+        session._segmented_objects = seg2
+
+        # Total stored should be count1 + new_count
+        stored = _run_async(session.object_store.get_session_objects(session.id))
+        assert len(stored) == count1 + new_count
+
+
+# ── _run_async helper ────────────────────────────────────────────────────
+
+
+class TestRunAsync:
+    """Test the _run_async helper for sync/async bridging."""
+
+    def test_run_async_simple(self):
+        async def add(a, b):
+            return a + b
+
+        assert _run_async(add(2, 3)) == 5
+
+    def test_run_async_with_await(self):
+        async def delayed():
+            await asyncio.sleep(0.01)
+            return "done"
+
+        assert _run_async(delayed()) == "done"
+
+    def test_run_async_exception_propagates(self):
+        async def failing():
+            raise ValueError("test error")
+
+        with pytest.raises(ValueError, match="test error"):
+            _run_async(failing())