diff --git a/haiku_rag_slim/haiku/rag/client/__init__.py b/haiku_rag_slim/haiku/rag/client/__init__.py
index 1a65c714..a6d25ad2 100644
--- a/haiku_rag_slim/haiku/rag/client/__init__.py
+++ b/haiku_rag_slim/haiku/rag/client/__init__.py
@@ -16,6 +16,7 @@ from urllib.parse import urlparse
 import httpx
 
 from haiku.rag.client.documents import DocumentImport
+from haiku.rag.client.session import SingleDatabaseSession
 from haiku.rag.config import AppConfig, get_config
 from haiku.rag.converters import get_converter
 from haiku.rag.embeddings import get_embedder
@@ -50,17 +51,6 @@ if TYPE_CHECKING:
 
 logger = logging.getLogger(__name__)
 
-# Throttle for the background auto-vacuum: under sustained ingestion, scheduling
-# a compaction on every write degenerates into back-to-back optimize() passes
-# that churn the blob-bearing documents table. Fire at most one per interval; a
-# final vacuum on close collapses anything throttled here.
-_VACUUM_MIN_INTERVAL_S = 300.0
-
-
-# Failures whose message names the remedy and never the location, so the failing
-# database is named alongside it instead of in place of it.
-_NAMEABLE_FAILURES = (MigrationRequiredError, ConfigMismatchError, ReadOnlyError)
-
 
 async def first_found(
     clients: "list[HaikuRAG]",
@@ -154,14 +144,41 @@ class HaikuRAG:
         self._skip_validation = skip_validation
         self._create = create
         self._read_only = read_only
-        self._vacuum_tasks: set[asyncio.Task] = set()
-        self._last_vacuum_at: float | None = None
-        self._vacuum_dirty = False
         self._requested_sources = sources
         self._clients: dict[str, HaikuRAG] = {}
         self._federated: dict[str, str] = {}
         self._clients_lock = asyncio.Lock()
         self._source: str | None = None
+        self._session: SingleDatabaseSession | None = None
+
+    @property
+    def store(self) -> Store:
+        """The store of the database this client opened.
+
+        Absent while covering a set: a store has no unambiguous meaning across
+        several, and `clients_for` reaches the one holding a given database.
+        """
+        if self._session is None:
+            raise AttributeError("store")
+        return self._session.store
+
+    @property
+    def document_repository(self) -> DocumentRepository:
+        if self._session is None:
+            raise AttributeError("document_repository")
+        return self._session.document_repository
+
+    @property
+    def chunk_repository(self) -> ChunkRepository:
+        if self._session is None:
+            raise AttributeError("chunk_repository")
+        return self._session.chunk_repository
+
+    @property
+    def document_item_repository(self) -> DocumentItemRepository:
+        if self._session is None:
+            raise AttributeError("document_item_repository")
+        return self._session.document_item_repository
 
     @property
     def is_read_only(self) -> bool:
@@ -247,47 +264,14 @@ class HaikuRAG:
             if db_path is not None:
                 self._db_path = db_path
 
-        failure: str | None = None
-        try:
-            self.store = Store(
-                self._db_path,
-                config=self._config,
-                skip_validation=self._skip_validation,
-                create=self._create,
-                read_only=self._read_only,
-            )
-            # If _initialize fails mid-way (e.g. migration check raises after
-            # connect), close the store so we don't leak the LanceDB connection —
-            # __aexit__ won't run because the `async with` never entered.
-            try:
-                await self.store._initialize()
-            except BaseException:
-                self.store.close()
-                raise
-        except _NAMEABLE_FAILURES as error:
-            # These say what to run and never where the database is, so the name
-            # is added to the message rather than replacing it: the operator needs
-            # both which database failed and what to do about it.
-            if self._source is None:
-                raise
-            raise type(error)(f"database {self._source!r}: {error}") from error
-        except Exception as error:
-            # A legacy `uri` or `db_path` client has no name to report instead, so
-            # its error passes through as it always has.
-            if self._source is None:
-                raise
-            failure = type(error).__name__
-        if failure is not None:
-            # Raised outside the except block on purpose. A database named in
-            # config is reported by name, and the original spells out the path or
-            # the bucket: `from None` would only stop it being *printed*, leaving
-            # it on `__context__` for anything that walks the chain.
-            raise SourceUnavailableError(
-                f"database {self._source!r} could not be opened: {failure}"
-            )
-        self.document_repository = DocumentRepository(self.store)
-        self.chunk_repository = ChunkRepository(self.store)
-        self.document_item_repository = DocumentItemRepository(self.store)
+        self._session = await SingleDatabaseSession(
+            self._db_path,
+            self._config,
+            skip_validation=self._skip_validation,
+            create=self._create,
+            read_only=self._read_only,
+            source=self._source,
+        ).open()
         return self
 
     async def clients_for(self, names: list[str]) -> list["HaikuRAG"]:
@@ -410,44 +394,12 @@ class HaikuRAG:
             await _aclose_quietly(cached, name)
 
     async def _await_vacuum_tasks(self) -> None:
-        """Drain background vacuum work and run a final collapse before teardown.
-
-        Writes schedule a throttled background vacuum; many are debounced or skip
-        because another vacuum holds the lock. The final pass collapses the
-        versions those left behind. It runs whenever writes happened
-        (``_vacuum_dirty``) — not gated on in-flight tasks remaining, since a
-        debounced run may have scheduled none — but never when nothing was
-        written (so opening + closing a store still never writes).
-        """
-        if self._vacuum_tasks:
-            await asyncio.gather(*self._vacuum_tasks, return_exceptions=True)
-        if not self._vacuum_dirty:
-            return
-        self._vacuum_dirty = False
-        # __aexit__ runs during exception unwinding; a raising vacuum here would
-        # mask the original exception, so the drain stays best-effort.
-        try:
-            await self.store.vacuum()
-        except Exception:
-            logger.debug("Final vacuum on close failed", exc_info=True)
+        if self._session is not None:
+            await self._session.drain_vacuum()
 
     def _schedule_vacuum(self) -> None:
-        """Schedule a background vacuum, throttled to at most one per
-        ``_VACUUM_MIN_INTERVAL_S``. Sustained writes would otherwise trigger
-        back-to-back compaction of the blob-bearing documents table. The throttle
-        only skips the background task — ``_vacuum_dirty`` still marks that a
-        final vacuum on close is owed."""
-        self._vacuum_dirty = True
-        now = monotonic()
-        if (
-            self._last_vacuum_at is not None
-            and now - self._last_vacuum_at < _VACUUM_MIN_INTERVAL_S
-        ):
-            return
-        self._last_vacuum_at = now
-        task = asyncio.create_task(self.store.vacuum())
-        self._vacuum_tasks.add(task)
-        task.add_done_callback(self._vacuum_tasks.discard)
+        if self._session is not None:
+            self._session.schedule_vacuum()
 
     # =========================================================================
     # Processing Primitives
@@ -935,4 +887,5 @@ class HaikuRAG:
     def close(self):
         """Close the underlying store connection."""
         self._require_one_database("close")
-        self.store.close()
+        assert self._session is not None
+        self._session.close()
diff --git a/haiku_rag_slim/haiku/rag/client/session.py b/haiku_rag_slim/haiku/rag/client/session.py
new file mode 100644
index 00000000..ce10fcfd
--- /dev/null
+++ b/haiku_rag_slim/haiku/rag/client/session.py
@@ -0,0 +1,147 @@
+import asyncio
+import logging
+from pathlib import Path
+from time import monotonic
+
+from haiku.rag.config import AppConfig
+from haiku.rag.store.engine import Store
+from haiku.rag.store.exceptions import (
+    ConfigMismatchError,
+    MigrationRequiredError,
+    ReadOnlyError,
+    SourceUnavailableError,
+)
+from haiku.rag.store.repositories.chunk import ChunkRepository
+from haiku.rag.store.repositories.document import DocumentRepository
+from haiku.rag.store.repositories.document_item import DocumentItemRepository
+
+logger = logging.getLogger(__name__)
+
+# Throttle for the background auto-vacuum: under sustained ingestion, scheduling
+# a compaction on every write degenerates into back-to-back optimize() passes
+# that churn the blob-bearing documents table. Fire at most one per interval; a
+# final vacuum on close collapses anything throttled here.
+_VACUUM_MIN_INTERVAL_S = 300.0
+
+
+# Failures whose message names the remedy and never the location, so the failing
+# database is named alongside it instead of in place of it.
+_NAMEABLE_FAILURES = (MigrationRequiredError, ConfigMismatchError, ReadOnlyError)
+
+
+class SingleDatabaseSession:
+    """One database: its store, its repositories, and their lifecycle.
+
+    Everything that needs a store lives here, so nothing above has to ask whether
+    it has one. ``source`` is the configured name this database answers to, or
+    None where nothing names it.
+    """
+
+    def __init__(
+        self,
+        db_path: Path | str,
+        config: AppConfig,
+        *,
+        skip_validation: bool = False,
+        create: bool = False,
+        read_only: bool = False,
+        source: str | None = None,
+    ) -> None:
+        self._db_path = db_path
+        self._config = config
+        self._skip_validation = skip_validation
+        self._create = create
+        self._read_only = read_only
+        self.source = source
+        self._vacuum_tasks: set[asyncio.Task] = set()
+        self._last_vacuum_at: float | None = None
+        self._vacuum_dirty = False
+
+    async def open(self) -> "SingleDatabaseSession":
+        """Connect, validate, and build the repositories."""
+        failure: str | None = None
+        try:
+            self.store = Store(
+                self._db_path,
+                config=self._config,
+                skip_validation=self._skip_validation,
+                create=self._create,
+                read_only=self._read_only,
+            )
+            # If _initialize fails mid-way (e.g. migration check raises after
+            # connect), close the store so we don't leak the LanceDB connection —
+            # the caller's `async with` never entered, so its exit won't run.
+            try:
+                await self.store._initialize()
+            except BaseException:
+                self.store.close()
+                raise
+        except _NAMEABLE_FAILURES as error:
+            # These say what to run and never where the database is, so the name
+            # is added to the message rather than replacing it: the operator needs
+            # both which database failed and what to do about it.
+            if self.source is None:
+                raise
+            raise type(error)(f"database {self.source!r}: {error}") from error
+        except Exception as error:
+            # A legacy `uri` or `db_path` session has no name to report instead,
+            # so its error passes through as it always has.
+            if self.source is None:
+                raise
+            failure = type(error).__name__
+        if failure is not None:
+            # Raised outside the except block on purpose. A database named in
+            # config is reported by name, and the original spells out the path or
+            # the bucket: `from None` would only stop it being *printed*, leaving
+            # it on `__context__` for anything that walks the chain.
+            raise SourceUnavailableError(
+                f"database {self.source!r} could not be opened: {failure}"
+            )
+        self.document_repository = DocumentRepository(self.store)
+        self.chunk_repository = ChunkRepository(self.store)
+        self.document_item_repository = DocumentItemRepository(self.store)
+        return self
+
+    async def drain_vacuum(self) -> None:
+        """Drain background vacuum work and run a final collapse before teardown.
+
+        Writes schedule a throttled background vacuum; many are debounced or skip
+        because another vacuum holds the lock. The final pass collapses the
+        versions those left behind. It runs whenever writes happened
+        (``_vacuum_dirty``) — not gated on in-flight tasks remaining, since a
+        debounced run may have scheduled none — but never when nothing was
+        written (so opening + closing a store still never writes).
+        """
+        if self._vacuum_tasks:
+            await asyncio.gather(*self._vacuum_tasks, return_exceptions=True)
+        if not self._vacuum_dirty:
+            return
+        self._vacuum_dirty = False
+        # Teardown runs during exception unwinding; a raising vacuum here would
+        # mask the original exception, so the drain stays best-effort.
+        try:
+            await self.store.vacuum()
+        except Exception:
+            logger.debug("Final vacuum on close failed", exc_info=True)
+
+    def schedule_vacuum(self) -> None:
+        """Schedule a background vacuum, throttled to at most one per
+        ``_VACUUM_MIN_INTERVAL_S``. Sustained writes would otherwise trigger
+        back-to-back compaction of the blob-bearing documents table. The throttle
+        only skips the background task — ``_vacuum_dirty`` still marks that a
+        final vacuum on close is owed."""
+        self._vacuum_dirty = True
+        now = monotonic()
+        if (
+            self._last_vacuum_at is not None
+            and now - self._last_vacuum_at < _VACUUM_MIN_INTERVAL_S
+        ):
+            return
+        self._last_vacuum_at = now
+        task = asyncio.create_task(self.store.vacuum())
+        self._vacuum_tasks.add(task)
+        task.add_done_callback(self._vacuum_tasks.discard)
+
+    def close(self) -> None:
+        """Close the underlying store connection."""
+        self.store.close()
diff --git a/tests/test_client.py b/tests/test_client.py
index 8636a99a..7e473fab 100644
--- a/tests/test_client.py
+++ b/tests/test_client.py
@@ -1196,10 +1196,10 @@ async def test_delete_marks_vacuum_dirty(temp_db_path):
             uri="mem://del",
         )
         assert doc.id is not None
-        client._vacuum_dirty = False  # isolate the delete
+        client._session._vacuum_dirty = False  # isolate the delete
 
         assert await client.delete_document(doc.id) is True
-        assert client._vacuum_dirty is True
+        assert client._session._vacuum_dirty is True
 
 
 async def test_delete_rolls_back_on_partial_failure(temp_db_path, monkeypatch):
@@ -1284,9 +1284,9 @@ async def test_delete_missing_id_returns_false_without_vacuum(temp_db_path):
     """Deleting an id that doesn't exist returns False and owes no vacuum (the
     existence check is inside the lock, so a no-op delete stays a no-op)."""
     async with HaikuRAG(temp_db_path, create=True) as client:
-        client._vacuum_dirty = False
+        client._session._vacuum_dirty = False
         assert await client.delete_document("does-not-exist") is False
-        assert client._vacuum_dirty is False
+        assert client._session._vacuum_dirty is False
 
 
 @pytest.mark.vcr()
@@ -2843,7 +2843,7 @@ async def test_import_documents_schedules_vacuum_per_config(temp_db_path, auto_v
                     )
                 ]
             )
-            await asyncio.gather(*client._vacuum_tasks)
+            await asyncio.gather(*client._session._vacuum_tasks)
 
             assert vacuum.await_count == (1 if auto_vacuum else 0)
 
diff --git a/tests/test_multi_db.py b/tests/test_multi_db.py
index f0741747..63946667 100644
--- a/tests/test_multi_db.py
+++ b/tests/test_multi_db.py
@@ -485,6 +485,23 @@ class TestOperationsThatNeedOneDatabase:
             with pytest.raises(AmbiguousDatabaseError, match="clients_for"):
                 await rag.vacuum()
 
+    @pytest.mark.asyncio
+    async def test_a_set_has_no_store_of_its_own(self, tmp_path):
+        """A store and its repositories belong to one database. `clients_for`
+        reaches the one holding a given database."""
+        config = _config(tmp_path, ["alpha", "beta"])
+        await _seed(config, "alpha", ["alpha one"])
+
+        async with HaikuRAG(config=config) as rag:
+            for name in (
+                "store",
+                "document_repository",
+                "chunk_repository",
+                "document_item_repository",
+            ):
+                with pytest.raises(AttributeError, match=name):
+                    getattr(rag, name)
+
     @pytest.mark.asyncio
     async def test_a_selected_database_is_still_writable(self, tmp_path):
         """Naming one of the set is how a write picks its database."""
diff --git a/tests/test_vacuum_debounce.py b/tests/test_vacuum_debounce.py
index 6e6de526..57c2f3b4 100644
--- a/tests/test_vacuum_debounce.py
+++ b/tests/test_vacuum_debounce.py
@@ -2,7 +2,7 @@ import asyncio
 
 import pytest
 
-import haiku.rag.client as client_mod
+import haiku.rag.client.session as session_mod
 from haiku.rag.client import HaikuRAG
 from haiku.rag.client.documents import _refresh_doc_metadata
 from haiku.rag.config import get_config
@@ -23,7 +23,7 @@ async def test_schedule_vacuum_is_debounced(temp_db_path, monkeypatch):
     """Rapid writes within the throttle window schedule only one background
     vacuum; once the interval elapses, a new one is scheduled."""
     t = {"now": 1000.0}
-    monkeypatch.setattr(client_mod, "monotonic", lambda: t["now"])
+    monkeypatch.setattr(session_mod, "monotonic", lambda: t["now"])
 
     async with HaikuRAG(temp_db_path, create=True) as client:
         calls: list[int] = []
@@ -34,13 +34,13 @@ async def test_schedule_vacuum_is_debounced(temp_db_path, monkeypatch):
         monkeypatch.setattr(client.store, "vacuum", fake_vacuum)
 
         for _ in range(3):
-            client._schedule_vacuum()
-        await asyncio.gather(*client._vacuum_tasks)
+            client._session.schedule_vacuum()
+        await asyncio.gather(*client._session._vacuum_tasks)
         assert len(calls) == 1  # debounced within the interval
 
-        t["now"] += client_mod._VACUUM_MIN_INTERVAL_S + 1
-        client._schedule_vacuum()
-        await asyncio.gather(*client._vacuum_tasks)
+        t["now"] += session_mod._VACUUM_MIN_INTERVAL_S + 1
+        client._session.schedule_vacuum()
+        await asyncio.gather(*client._session._vacuum_tasks)
         assert len(calls) == 2  # interval elapsed -> a new vacuum scheduled
 
 
@@ -49,7 +49,7 @@ async def test_debounced_writes_still_collapse_on_close(temp_db_path, monkeypatc
     """Even when scheduled vacuums after the first are debounced, the writes are
     marked dirty so the close-time drain runs a final collapse."""
     t = {"now": 1000.0}
-    monkeypatch.setattr(client_mod, "monotonic", lambda: t["now"])
+    monkeypatch.setattr(session_mod, "monotonic", lambda: t["now"])
     calls: list[int] = []
 
     async with HaikuRAG(temp_db_path, create=True) as client:
@@ -59,13 +59,13 @@ async def test_debounced_writes_still_collapse_on_close(temp_db_path, monkeypatc
 
         monkeypatch.setattr(client.store, "vacuum", fake_vacuum)
 
-        client._schedule_vacuum()  # schedules the first background pass
-        client._schedule_vacuum()  # debounced (no task)
+        client._session.schedule_vacuum()  # schedules the first background pass
+        client._session.schedule_vacuum()  # debounced (no task)
 
-        await client._await_vacuum_tasks()
+        await client._session.drain_vacuum()
         # one scheduled background pass + one final collapse on drain
         assert len(calls) == 2
-        assert client._vacuum_dirty is False
+        assert client._session._vacuum_dirty is False
 
 
 @pytest.mark.asyncio
@@ -82,7 +82,7 @@ async def test_metadata_refresh_sweep_schedules_vacuum(temp_db_path):
             metadata={"source_revision": "r1"},
         )
         # Isolate the refresh: the import already scheduled a vacuum.
-        client._vacuum_dirty = False
+        client._session._vacuum_dirty = False
 
         await _refresh_doc_metadata(
             client,
@@ -91,7 +91,7 @@ async def test_metadata_refresh_sweep_schedules_vacuum(temp_db_path):
             user_metadata={},
             source_metadata={"source_revision": "r2", "md5": "same"},
         )
-        assert client._vacuum_dirty is True
+        assert client._session._vacuum_dirty is True
 
 
 @pytest.mark.asyncio
diff --git a/tests/test_versioning.py b/tests/test_versioning.py
index 9e8e7953..fe5a0c18 100644
--- a/tests/test_versioning.py
+++ b/tests/test_versioning.py
@@ -205,7 +205,7 @@ async def test_existing_database_checks_migrations(monkeypatch, temp_db_path):
 
 async def _wait_for_background_vacuum(client):
     """Wait for any in-flight background vacuum tasks to complete."""
-    await client._await_vacuum_tasks()
+    await client._session.drain_vacuum()
 
 
 @pytest.mark.vcr()
@@ -438,7 +438,8 @@ async def test_close_suppresses_failing_drain_vacuum(temp_db_path, monkeypatch):
         raise RuntimeError("vacuum boom")
 
     # Writes happened, so close owes a final vacuum — force that drain branch.
-    client._vacuum_dirty = True
+    assert client._session is not None
+    client._session._vacuum_dirty = True
     monkeypatch.setattr(client.store, "vacuum", boom)
 
     # Must not raise despite the drain vacuum erroring.