From e7c7df2915addc779cfa4ea4ec9de2293fed1dc9 Mon Sep 17 00:00:00 2001
From: Yiorgis Gozadinos <ggozadinos@gmail.com>
Date: Fri, 29 May 2026 11:27:08 +0300
Subject: [PATCH] Always use the Store-owned embedder

---
 CHANGELOG.md                                  |  2 +
 docs/custom-pipelines.md                      | 12 ++--
 haiku_rag_slim/haiku/rag/client/__init__.py   | 26 ++++++--
 haiku_rag_slim/haiku/rag/client/documents.py  | 16 +++--
 haiku_rag_slim/haiku/rag/client/processing.py | 12 ++--
 haiku_rag_slim/haiku/rag/client/rebuild.py    | 15 +++--
 haiku_rag_slim/haiku/rag/client/search.py     |  4 +-
 .../haiku/rag/embeddings/__init__.py          |  4 +-
 haiku_rag_slim/haiku/rag/mcp.py               |  4 +-
 tests/test_chunk.py                           |  2 +-
 tests/test_client.py                          |  2 +-
 tests/test_context_enhancement.py             |  4 +-
 tests/test_embedder.py                        | 65 ++++++++++++++++---
 tests/test_picture_in_context.py              | 30 +++------
 tests/test_search.py                          |  4 +-
 tests/test_versioning.py                      | 24 +++++++
 16 files changed, 157 insertions(+), 69 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 29337ca3..e9f410ec 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,8 @@
 
 - A successful DELETE job auto-prunes dead jobs with the same `(source_id, uri)`. New `JobRepo.prune_dead(source_id, uri)`.
 - Reranker built once per `HaikuRAG` client instead of per `search()` call.
+- Embedder built once per `HaikuRAG` client instead of rebuilt per ingest/search operation. `embed_chunks` now takes it explicitly: `embed_chunks(chunks, embedder, config)` (was `embed_chunks(chunks, config)`). New `HaikuRAG.embedder` property exposes it.
+- `HaikuRAG` close runs a final vacuum after draining in-flight background vacuums when `storage.auto_vacuum` is enabled.
 
 ## [0.50.0] - 2026-05-27
 
diff --git a/docs/custom-pipelines.md b/docs/custom-pipelines.md
index 3321e4ed..a523b4af 100644
--- a/docs/custom-pipelines.md
+++ b/docs/custom-pipelines.md
@@ -26,7 +26,7 @@ The client exposes four primitives that can be composed into custom workflows:
 |-----------|-------|--------|---------|
 | `convert()` | file, URL, or text | `DoclingDocument` | Convert source to structured document |
 | `chunk()` | `DoclingDocument` | `list[Chunk]` | Split document into chunks |
-| `embed_chunks()` | `list[Chunk]` | `list[Chunk]` | Generate embeddings for chunks (includes contextualization) |
+| `embed_chunks()` | `list[Chunk]`, embedder | `list[Chunk]` | Generate embeddings for chunks (includes contextualization) |
 | `contextualize()` | `list[Chunk]` | `list[str]` | Get embedding-ready text (for custom embedders only) |
 
 ## Basic Pipeline
@@ -45,7 +45,7 @@ async with HaikuRAG("database.lancedb", create=True) as client:
     chunks = await client.chunk(docling_doc)
 
     # 3. Generate embeddings
-    embedded_chunks = await embed_chunks(chunks)
+    embedded_chunks = await embed_chunks(chunks, client.embedder)
 
     # 4. Store the document with chunks
     doc = await client.import_document(
@@ -117,13 +117,13 @@ Chunks are returned with:
 
 ## Embed
 
-`embed_chunks()` generates embeddings for chunks. It automatically contextualizes chunks (prepends section headings) before embedding for better semantic search, without modifying the stored content:
+`embed_chunks()` generates embeddings for chunks using the client's embedder. It automatically contextualizes chunks (prepends section headings) before embedding for better semantic search, without modifying the stored content:
 
 ```python
 from haiku.rag.embeddings import embed_chunks
 
 # Generate embeddings (returns new Chunk objects)
-embedded_chunks = await embed_chunks(chunks)
+embedded_chunks = await embed_chunks(chunks, client.embedder)
 
 # Original chunks unchanged
 assert chunks[0].embedding is None
@@ -175,7 +175,7 @@ async with HaikuRAG("database.lancedb", create=True) as client:
 
     # Continue with standard pipeline
     chunks = await client.chunk(processed_doc)
-    embedded_chunks = await embed_chunks(chunks)
+    embedded_chunks = await embed_chunks(chunks, client.embedder)
 
     await client.import_document(
         chunks=embedded_chunks,
@@ -203,7 +203,7 @@ async with HaikuRAG("database.lancedb", create=True) as client:
     for i, chunk in enumerate(filtered):
         chunk.order = i
 
-    embedded_chunks = await embed_chunks(filtered)
+    embedded_chunks = await embed_chunks(filtered, client.embedder)
 
     await client.import_document(
         docling_document=docling_doc,
diff --git a/haiku_rag_slim/haiku/rag/client/__init__.py b/haiku_rag_slim/haiku/rag/client/__init__.py
index 6af61629..c4eac92d 100644
--- a/haiku_rag_slim/haiku/rag/client/__init__.py
+++ b/haiku_rag_slim/haiku/rag/client/__init__.py
@@ -31,6 +31,7 @@ if TYPE_CHECKING:
     from docling_core.types.doc.document import DoclingDocument
     from PIL import Image as PILImage
 
+    from haiku.rag.embeddings import EmbedderWrapper
     from haiku.rag.ingester.sources.base import Source
     from haiku.rag.reranking.base import RerankerBase
     from haiku.rag.sandbox import AnalysisResult
@@ -89,6 +90,11 @@ class HaikuRAG:
         """Whether the client is in read-only mode."""
         return self.store.is_read_only
 
+    @property
+    def embedder(self) -> "EmbedderWrapper":
+        """The embedder owned by the Store, reused across all operations."""
+        return self.store.embedder
+
     @cached_property
     def reranker(self) -> "RerankerBase | None":
         """The configured reranker, built once and reused across searches.
@@ -128,14 +134,23 @@ class HaikuRAG:
         return False
 
     async def _await_vacuum_tasks(self) -> None:
-        """Wait for all in-flight background vacuum tasks to complete.
+        """Drain background vacuum work before tearing down the connection.
 
         Each create_document / update_document can schedule its own vacuum task;
-        all must be awaited before tearing down the connection, not just the
-        most recently scheduled one.
+        all must be awaited, not just the most recently scheduled one. Vacuum
+        skips when another is already running, so the cleanup for the final
+        writes may have been a no-op. Run one more pass once the in-flight tasks
+        are done to collapse versions created after the last vacuum took the lock.
         """
-        if self._vacuum_tasks:
-            await asyncio.gather(*self._vacuum_tasks, return_exceptions=True)
+        if not self._vacuum_tasks:
+            return
+        await asyncio.gather(*self._vacuum_tasks, return_exceptions=True)
+        # __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)
 
     def _schedule_vacuum(self) -> None:
         """Schedule a background vacuum and track the task for later awaiting."""
@@ -180,6 +195,7 @@ class HaikuRAG:
         return await chunk(
             self._config,
             docling_document,
+            embedder=self.embedder,
             existing_picture_data=existing_picture_data,
             document_id=document_id,
         )
diff --git a/haiku_rag_slim/haiku/rag/client/documents.py b/haiku_rag_slim/haiku/rag/client/documents.py
index a667213d..bf395921 100644
--- a/haiku_rag_slim/haiku/rag/client/documents.py
+++ b/haiku_rag_slim/haiku/rag/client/documents.py
@@ -59,7 +59,7 @@ async def _store_document_with_chunks(
 
     Handles versioning/rollback on failure.
     """
-    chunks = await ensure_chunks_embedded(client._config, chunks)
+    chunks = await ensure_chunks_embedded(client._config, chunks, client.embedder)
 
     versions = await client.store.current_table_versions()
 
@@ -109,7 +109,7 @@ async def _update_document_with_chunks(
             await client.document_item_repository.get_all_picture_data(document.id)
         )
 
-    chunks = await ensure_chunks_embedded(client._config, chunks)
+    chunks = await ensure_chunks_embedded(client._config, chunks, client.embedder)
 
     versions = await client.store.current_table_versions()
 
@@ -160,7 +160,7 @@ async def create_document(
     converter = get_converter(client._config)
     docling_document = await converter.convert_text(content, format=format)
     chunks = await client.chunk(docling_document)
-    embedded_chunks = await embed_chunks(chunks, client._config)
+    embedded_chunks = await embed_chunks(chunks, client.embedder, client._config)
 
     stored_content = docling_document.export_to_markdown()
 
@@ -286,7 +286,9 @@ async def _ingest_fetch_result(
             chunks = await client.chunk(docling_document)
             chunk_span.set_attribute("chunks_created", len(chunks))
         with logfire.span("document.embed", uri=result.uri):
-            embedded_chunks = await embed_chunks(chunks, client._config)
+            embedded_chunks = await embed_chunks(
+                chunks, client.embedder, client._config
+            )
     finally:
         if cleanup_path is not None:
             cleanup_path.unlink(missing_ok=True)
@@ -635,7 +637,9 @@ async def update_document(
         existing_doc.set_docling(docling_document)
 
         new_chunks = await client.chunk(docling_document)
-        embedded_chunks = await embed_chunks(new_chunks, client._config)
+        embedded_chunks = await embed_chunks(
+            new_chunks, client.embedder, client._config
+        )
         return await _update_document_with_chunks(
             client, existing_doc, embedded_chunks, docling_document
         )
@@ -647,7 +651,7 @@ async def update_document(
     existing_doc.set_docling(converted_docling)
 
     new_chunks = await client.chunk(converted_docling)
-    embedded_chunks = await embed_chunks(new_chunks, client._config)
+    embedded_chunks = await embed_chunks(new_chunks, client.embedder, client._config)
     return await _update_document_with_chunks(
         client, existing_doc, embedded_chunks, converted_docling
     )
diff --git a/haiku_rag_slim/haiku/rag/client/processing.py b/haiku_rag_slim/haiku/rag/client/processing.py
index 224dbe6c..3300e55e 100644
--- a/haiku_rag_slim/haiku/rag/client/processing.py
+++ b/haiku_rag_slim/haiku/rag/client/processing.py
@@ -15,6 +15,8 @@ from haiku.rag.store.models.document_item import _picture_description_text
 if TYPE_CHECKING:
     from docling_core.types.doc.document import DoclingDocument
 
+    from haiku.rag.embeddings import EmbedderWrapper
+
 
 logger = logging.getLogger(__name__)
 
@@ -169,6 +171,7 @@ async def chunk(
     config: AppConfig,
     docling_document: "DoclingDocument",
     *,
+    embedder: "EmbedderWrapper",
     existing_picture_data: dict[str, bytes] | None = None,
     document_id: str | None = None,
 ) -> list[Chunk]:
@@ -184,12 +187,11 @@ async def chunk(
     path where the docling is loaded from the stored blob.
     """
     from haiku.rag.chunkers import get_chunker
-    from haiku.rag.embeddings import get_embedder
 
     chunker = get_chunker(config)
     text_chunks = await chunker.chunk(docling_document)
 
-    if not get_embedder(config).supports_images:
+    if not embedder.supports_images:
         for i, c in enumerate(text_chunks):
             c.order = i
         return text_chunks
@@ -277,7 +279,9 @@ def build_picture_chunks(
     return chunks
 
 
-async def ensure_chunks_embedded(config: AppConfig, chunks: list[Chunk]) -> list[Chunk]:
+async def ensure_chunks_embedded(
+    config: AppConfig, chunks: list[Chunk], embedder: "EmbedderWrapper"
+) -> list[Chunk]:
     """Ensure all chunks have embeddings, embedding any that don't.
 
     Chunks that already have embeddings are passed through unchanged; missing
@@ -290,7 +294,7 @@ async def ensure_chunks_embedded(config: AppConfig, chunks: list[Chunk]) -> list
     if not chunks_to_embed:
         return chunks
 
-    embedded = await embed_chunks(chunks_to_embed, config)
+    embedded = await embed_chunks(chunks_to_embed, embedder, config)
 
     # Build result maintaining original order
     embedded_map = {(c.content, c.order): c for c in embedded}
diff --git a/haiku_rag_slim/haiku/rag/client/rebuild.py b/haiku_rag_slim/haiku/rag/client/rebuild.py
index 030717fb..6144dae4 100644
--- a/haiku_rag_slim/haiku/rag/client/rebuild.py
+++ b/haiku_rag_slim/haiku/rag/client/rebuild.py
@@ -503,11 +503,11 @@ async def _rebuild_rechunk(
     client: "HaikuRAG", documents: list[Document]
 ) -> AsyncGenerator[str, None]:
     """Re-chunk and re-embed each document from its stored docling blob."""
-    from haiku.rag.embeddings import embed_chunks, get_embedder
+    from haiku.rag.embeddings import embed_chunks
 
     pending_chunks: list[Chunk] = []
     pending_docs: list[Document] = []
-    embedder = get_embedder(client._config)
+    embedder = client.embedder
 
     async for doc in _hydrate(client, documents):
         assert doc.id is not None
@@ -530,7 +530,7 @@ async def _rebuild_rechunk(
             existing_picture_data=existing_picture_data,
             document_id=doc.id,
         )
-        embedded_chunks = await embed_chunks(chunks, client._config)
+        embedded_chunks = await embed_chunks(chunks, embedder, client._config)
 
         # Prepare chunks with document_id and order
         for order, chunk in enumerate(embedded_chunks):
@@ -638,7 +638,7 @@ async def _rebuild_descriptions(
     cost remains. Idempotent: pictures whose ``meta.description.text`` is
     already populated are not re-described.
     """
-    from haiku.rag.embeddings import embed_chunks, get_embedder
+    from haiku.rag.embeddings import embed_chunks
 
     if client._config.processing.pictures != "description":
         raise ValueError(
@@ -648,7 +648,7 @@ async def _rebuild_descriptions(
 
     pending_chunks: list[Chunk] = []
     pending_docs: list[Document] = []
-    embedder = get_embedder(client._config)
+    embedder = client.embedder
 
     described_total = 0
     async for doc in _hydrate(client, documents):
@@ -676,7 +676,7 @@ async def _rebuild_descriptions(
             existing_picture_data=existing_picture_data,
             document_id=doc.id,
         )
-        embedded_chunks = await embed_chunks(chunks, client._config)
+        embedded_chunks = await embed_chunks(chunks, embedder, client._config)
 
         for order, chunk in enumerate(embedded_chunks):
             chunk.document_id = doc.id
@@ -710,6 +710,7 @@ async def _rebuild_full(
     pending_chunks: list[Chunk] = []
     pending_docs: list[Document] = []
     converter = get_converter(client._config)
+    embedder = client.embedder
 
     for light_doc in documents:
         assert light_doc.id is not None
@@ -751,7 +752,7 @@ async def _rebuild_full(
 
         docling_document = await converter.convert_text(doc.content, format="md")
         chunks = await client.chunk(docling_document)
-        embedded_chunks = await embed_chunks(chunks, client._config)
+        embedded_chunks = await embed_chunks(chunks, embedder, client._config)
 
         doc.set_docling(docling_document)
 
diff --git a/haiku_rag_slim/haiku/rag/client/search.py b/haiku_rag_slim/haiku/rag/client/search.py
index e441f51a..20a867a2 100644
--- a/haiku_rag_slim/haiku/rag/client/search.py
+++ b/haiku_rag_slim/haiku/rag/client/search.py
@@ -55,9 +55,7 @@ async def search(
             chunks = [chunk for chunk, _ in raw_results]
             chunk_results = await reranker.rerank(query, chunks, top_n=limit)
     else:
-        from haiku.rag.embeddings import get_embedder
-
-        embedder = get_embedder(client._config)
+        embedder = client.embedder
         if not embedder.supports_images:
             raise ValueError(
                 "Image queries require a multimodal embedder. Configure "
diff --git a/haiku_rag_slim/haiku/rag/embeddings/__init__.py b/haiku_rag_slim/haiku/rag/embeddings/__init__.py
index 8b8cb563..ce3e9e4a 100644
--- a/haiku_rag_slim/haiku/rag/embeddings/__init__.py
+++ b/haiku_rag_slim/haiku/rag/embeddings/__init__.py
@@ -83,7 +83,7 @@ def contextualize(chunks: list["Chunk"]) -> list[str]:
 
 
 async def embed_chunks(
-    chunks: list["Chunk"], config: AppConfig = Config
+    chunks: list["Chunk"], embedder: "EmbedderWrapper", config: AppConfig = Config
 ) -> list["Chunk"]:
     """Generate embeddings for chunks, dispatching text vs picture variants.
 
@@ -97,8 +97,6 @@ async def embed_chunks(
 
     from haiku.rag.store.models.chunk import Chunk
 
-    embedder = get_embedder(config)
-
     text_chunks: list[Chunk] = []
     picture_chunks: list[Chunk] = []
     for chunk in chunks:
diff --git a/haiku_rag_slim/haiku/rag/mcp.py b/haiku_rag_slim/haiku/rag/mcp.py
index 43aba551..e2e9d1bd 100644
--- a/haiku_rag_slim/haiku/rag/mcp.py
+++ b/haiku_rag_slim/haiku/rag/mcp.py
@@ -108,7 +108,9 @@ def create_mcp_server(
             return []
 
     # Image-as-query tool, only registered when the configured embedder
-    # supports image embeddings.
+    # supports image embeddings. Probed at server-build time when no Store is
+    # open, so there is no cached embedder to read; this is the one place
+    # outside Store that builds one.
     from haiku.rag.embeddings import get_embedder
 
     if get_embedder(config).supports_images:
diff --git a/tests/test_chunk.py b/tests/test_chunk.py
index f0679d1f..8d9e2d01 100644
--- a/tests/test_chunk.py
+++ b/tests/test_chunk.py
@@ -108,7 +108,7 @@ async def test_chunking_pipeline(qa_corpus: Dataset, temp_db_path):
         # Use client primitives: convert → chunk → embed
         docling_document = await client.convert(document_text)
         chunks = await client.chunk(docling_document)
-        embedded_chunks = await embed_chunks(chunks)
+        embedded_chunks = await embed_chunks(chunks, client.embedder)
 
         # Verify chunks were created with embeddings
         assert len(chunks) > 0
diff --git a/tests/test_client.py b/tests/test_client.py
index e4cfce2e..f0f7e8e7 100644
--- a/tests/test_client.py
+++ b/tests/test_client.py
@@ -1580,7 +1580,7 @@ async def test_sql_injection_is_blocked_with_escaping(temp_db_path):
 
 
 def _patch_embed_chunks(monkeypatch):
-    async def fake_embed_chunks(chunks, config):
+    async def fake_embed_chunks(chunks, embedder, config):
         for chunk in chunks:
             chunk.embedding = [0.0] * 2560
         return chunks
diff --git a/tests/test_context_enhancement.py b/tests/test_context_enhancement.py
index 977df96c..eb4db2b0 100644
--- a/tests/test_context_enhancement.py
+++ b/tests/test_context_enhancement.py
@@ -14,7 +14,9 @@ async def create_document_with_docling(
 ):
     """Helper to create a document from a DoclingDocument using import_document."""
     chunks = await client.chunk(docling_doc)
-    embedded_chunks = await ensure_chunks_embedded(client._config, chunks)
+    embedded_chunks = await ensure_chunks_embedded(
+        client._config, chunks, client.embedder
+    )
     return await client.import_document(
         docling_document=docling_doc,
         chunks=embedded_chunks,
diff --git a/tests/test_embedder.py b/tests/test_embedder.py
index 49552e98..2f20db9c 100644
--- a/tests/test_embedder.py
+++ b/tests/test_embedder.py
@@ -3,8 +3,18 @@ from pathlib import Path
 import numpy as np
 import pytest
 
-from haiku.rag.config import AppConfig, EmbeddingModelConfig, EmbeddingsConfig
-from haiku.rag.embeddings import contextualize, embed_chunks, get_embedder
+from haiku.rag.config import (
+    AppConfig,
+    Config,
+    EmbeddingModelConfig,
+    EmbeddingsConfig,
+)
+from haiku.rag.embeddings import (
+    EmbedderWrapper,
+    contextualize,
+    embed_chunks,
+    get_embedder,
+)
 from haiku.rag.store.models.chunk import Chunk
 
 
@@ -105,6 +115,45 @@ def test_contextualize_empty_list():
     assert texts == []
 
 
+class _StubEmbedder(EmbedderWrapper):
+    def __init__(self):
+        super().__init__(embedder=None, vector_dim=8)
+        self.doc_batches = 0
+
+    async def embed_documents(self, texts):
+        self.doc_batches += 1
+        return [[0.1] * 8 for _ in texts]
+
+
+async def test_embed_chunks_uses_provided_embedder(monkeypatch):
+    """embed_chunks embeds via the embedder it is given and never builds one."""
+    import haiku.rag.embeddings as embeddings_mod
+
+    def fail(*args, **kwargs):
+        raise AssertionError("embed_chunks must not build its own embedder")
+
+    monkeypatch.setattr(embeddings_mod, "get_embedder", fail)
+
+    embedder = _StubEmbedder()
+    chunks = [
+        Chunk(id="a", content="alpha", order=0),
+        Chunk(id="b", content="beta", order=1),
+    ]
+
+    embedded = await embed_chunks(chunks, embedder, AppConfig())
+
+    assert [c.embedding for c in embedded] == [[0.1] * 8, [0.1] * 8]
+    assert embedder.doc_batches == 1
+
+
+async def test_client_embedder_is_store_embedder(temp_db_path):
+    """The client exposes the Store's cached embedder rather than its own."""
+    from haiku.rag.client import HaikuRAG
+
+    async with HaikuRAG(temp_db_path, create=True) as client:
+        assert client.embedder is client.store.embedder
+
+
 @pytest.mark.vcr()
 async def test_embed_chunks_basic(allow_model_requests):
     """Test that embed_chunks generates embeddings for chunks."""
@@ -125,7 +174,7 @@ async def test_embed_chunks_basic(allow_model_requests):
         ),
     ]
 
-    embedded_chunks = await embed_chunks(chunks)
+    embedded_chunks = await embed_chunks(chunks, get_embedder(Config))
 
     assert len(embedded_chunks) == 2
     # Check that all original fields are preserved
@@ -144,7 +193,7 @@ async def test_embed_chunks_basic(allow_model_requests):
 async def test_embed_chunks_returns_new_objects(allow_model_requests):
     """Test that embed_chunks returns new Chunk objects (immutable pattern)."""
     original = Chunk(id="orig", content="Test content.")
-    embedded = await embed_chunks([original])
+    embedded = await embed_chunks([original], get_embedder(Config))
 
     # Original should be unchanged
     assert original.embedding is None
@@ -156,7 +205,7 @@ async def test_embed_chunks_returns_new_objects(allow_model_requests):
 
 async def test_embed_chunks_empty_list():
     """Test that embed_chunks handles empty list."""
-    result = await embed_chunks([])
+    result = await embed_chunks([], _StubEmbedder())
     assert result == []
 
 
@@ -176,7 +225,7 @@ async def test_embed_chunks_picture_with_text_only_embedder_raises():
     )
 
     with pytest.raises(ValueError, match="multimodal embedder"):
-        await embed_chunks([chunk], config)
+        await embed_chunks([chunk], get_embedder(config), config)
 
 
 async def test_embed_chunks_respects_configured_batch_size(monkeypatch):
@@ -206,7 +255,7 @@ async def test_embed_chunks_respects_configured_batch_size(monkeypatch):
         for i in range(num_chunks)
     ]
 
-    result = await embed_chunks(chunks, config)
+    result = await embed_chunks(chunks, get_embedder(config), config)
 
     assert len(result) == num_chunks
     # 20 chunks / 7 per batch -> 7, 7, 6
@@ -230,7 +279,7 @@ async def test_embed_chunks_preserves_all_fields(allow_model_requests):
         document_meta={"author": "Test"},
     )
 
-    embedded = await embed_chunks([chunk])
+    embedded = await embed_chunks([chunk], get_embedder(Config))
 
     assert embedded[0].id == "test-id"
     assert embedded[0].document_id == "doc-id"
diff --git a/tests/test_picture_in_context.py b/tests/test_picture_in_context.py
index 7c959b77..431e75e8 100644
--- a/tests/test_picture_in_context.py
+++ b/tests/test_picture_in_context.py
@@ -440,10 +440,6 @@ async def test_chunk_interleaves_picture_in_structural_order(monkeypatch):
                 ),
             ]
 
-    monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder",
-        lambda *a, **kw: StubMultimodalEmbedder(),
-    )
     monkeypatch.setattr(
         "haiku.rag.chunkers.get_chunker", lambda *a, **kw: StubChunker()
     )
@@ -459,7 +455,7 @@ async def test_chunk_interleaves_picture_in_structural_order(monkeypatch):
     doc.add_picture(image=ImageRef.from_pil(img, dpi=72))
     doc.add_text(label=DocItemLabel.PARAGRAPH, text="C")
 
-    chunks = await chunk(AppConfig(), doc)
+    chunks = await chunk(AppConfig(), doc, embedder=StubMultimodalEmbedder())
 
     contents = [c.content for c in chunks]
     assert contents == ["before", "", "after"], (
@@ -474,7 +470,7 @@ async def test_chunk_interleaves_picture_in_structural_order(monkeypatch):
 
 
 @pytest.mark.asyncio
-async def test_embed_chunks_dispatches_text_vs_picture(monkeypatch):
+async def test_embed_chunks_dispatches_text_vs_picture():
     """embed_chunks routes text chunks through embed_documents (batched) and
     picture chunks through embed_image (one at a time), reassembling
     in original order."""
@@ -498,10 +494,6 @@ async def test_embed_chunks_dispatches_text_vs_picture(monkeypatch):
             image_calls.append(image)
             return [0.9, 0.8, 0.7, 0.6]
 
-    monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder", lambda *a, **kw: StubEmbedder()
-    )
-
     text_chunk = Chunk(content="hello", order=0)
     pic_chunk = Chunk(
         content="figure 1",
@@ -510,7 +502,9 @@ async def test_embed_chunks_dispatches_text_vs_picture(monkeypatch):
     )
     pic_chunk._picture_data = b"PNGBYTES"
 
-    embedded = await embed_chunks([text_chunk, pic_chunk, text_chunk.model_copy()])
+    embedded = await embed_chunks(
+        [text_chunk, pic_chunk, text_chunk.model_copy()], StubEmbedder()
+    )
 
     assert len(embedded) == 3
     assert embedded[0].embedding == [0.1, 0.2, 0.3, 0.4]
@@ -521,9 +515,7 @@ async def test_embed_chunks_dispatches_text_vs_picture(monkeypatch):
 
 
 @pytest.mark.asyncio
-async def test_embed_chunks_raises_on_picture_chunks_with_text_only_embedder(
-    monkeypatch,
-):
+async def test_embed_chunks_raises_on_picture_chunks_with_text_only_embedder():
     from haiku.rag.embeddings import EmbedderWrapper, embed_chunks
     from haiku.rag.store.models.chunk import Chunk
 
@@ -534,14 +526,10 @@ async def test_embed_chunks_raises_on_picture_chunks_with_text_only_embedder(
         async def embed_documents(self, texts):
             return [[0.0] * 4 for _ in texts]
 
-    monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder", lambda *a, **kw: TextOnlyEmbedder()
-    )
-
     pic_chunk = Chunk(content="x", metadata={"labels": ["picture"]}, order=0)
     pic_chunk._picture_data = b"PNG"
     with pytest.raises(ValueError, match="multimodal embedder"):
-        await embed_chunks([pic_chunk])
+        await embed_chunks([pic_chunk], TextOnlyEmbedder())
 
 
 @pytest.mark.asyncio
@@ -569,7 +557,7 @@ async def test_ingest_emits_picture_chunks_with_multimodal_embedder(
             return [0.9] * 4
 
     monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder",
+        "haiku.rag.store.engine.get_embedder",
         lambda *a, **kw: StubMultimodalEmbedder(),
     )
 
@@ -585,7 +573,7 @@ async def test_ingest_emits_picture_chunks_with_multimodal_embedder(
 
     async with HaikuRAG(temp_db_path, config=config, create=True) as rag:
         chunks = await rag.chunk(docling_doc)
-        embedded = await embed_chunks(chunks, rag._config)
+        embedded = await embed_chunks(chunks, rag.embedder, rag._config)
 
         document = Document(content="x", uri="test://doc")
         document.set_docling(docling_doc)
diff --git a/tests/test_search.py b/tests/test_search.py
index a8e109af..b15199aa 100644
--- a/tests/test_search.py
+++ b/tests/test_search.py
@@ -354,7 +354,7 @@ async def test_search_with_bytes_query_uses_multimodal_embedder(
             return [0.5, 0.5, 0.5, 0.5]
 
     monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder",
+        "haiku.rag.store.engine.get_embedder",
         lambda *a, **kw: StubMultimodal(),
     )
 
@@ -446,7 +446,7 @@ async def test_search_with_pil_image_works_like_bytes(temp_db_path, monkeypatch)
             return [0.1] * 4
 
     monkeypatch.setattr(
-        "haiku.rag.embeddings.get_embedder",
+        "haiku.rag.store.engine.get_embedder",
         lambda *a, **kw: StubMultimodal(),
     )
 
diff --git a/tests/test_versioning.py b/tests/test_versioning.py
index 5d0a0198..a23520ac 100644
--- a/tests/test_versioning.py
+++ b/tests/test_versioning.py
@@ -320,3 +320,27 @@ async def test_auto_vacuum_enabled_triggers_vacuum(temp_db_path, monkeypatch):
         assert final_versions <= 2, (
             f"With auto-vacuum enabled, should have minimal versions, got {final_versions}"
         )
+
+
+async def test_close_suppresses_failing_drain_vacuum(temp_db_path, monkeypatch):
+    """A failing final vacuum on close must not raise out of __aexit__,
+    where it would mask an in-flight exception from the context body."""
+    client = HaikuRAG(db_path=temp_db_path, create=True)
+    await client.__aenter__()
+
+    calls: list[int] = []
+
+    async def boom(*args, **kwargs):
+        calls.append(1)
+        raise RuntimeError("vacuum boom")
+
+    # A finished task in the set forces the drain branch to run.
+    task = asyncio.create_task(asyncio.sleep(0))
+    await task
+    client._vacuum_tasks.add(task)
+    monkeypatch.setattr(client.store, "vacuum", boom)
+
+    # Must not raise despite the drain vacuum erroring.
+    await client.__aexit__(None, None, None)
+
+    assert calls, "drain vacuum should have been attempted"