diff --git a/pyproject.toml b/pyproject.toml
index 7058106..4591a9e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -516,6 +516,13 @@ module = [
     "fastapi.*",
     "uvicorn.*",
     "whisper.*",
+    # Optional deps for ``locus.rag.multimodal`` — only loaded inside
+    # try/except for users who want OCR / PDF / image / audio extraction.
+    "pytesseract.*",
+    "PIL.*",
+    "pypdf.*",
+    "PyPDF2.*",
+    "pdf2image.*",
 ]
 ignore_missing_imports = true
 
@@ -528,7 +535,6 @@ ignore_missing_imports = true
 # the strict block above.
 [[tool.mypy.overrides]]
 module = [
-    "locus.rag.*",
     # ``locus.hooks.builtin.*`` is held back: the module-level migration
     # surfaces a real runtime bug — the LoggingHook / GuardrailsHook /
     # TelemetryHook signatures pre-date the event-based ``HookProvider``
diff --git a/src/locus/rag/__init__.py b/src/locus/rag/__init__.py
index 63c9a79..de46285 100644
--- a/src/locus/rag/__init__.py
+++ b/src/locus/rag/__init__.py
@@ -62,6 +62,8 @@
     ... )
 """
 
+from typing import Any
+
 # Embeddings
 from locus.rag.embeddings.base import (
     BaseEmbedding,
@@ -128,7 +130,7 @@
 ]
 
 
-def __getattr__(name: str):
+def __getattr__(name: str) -> Any:
     """Lazy import providers and stores."""
     # Embedding providers
     if name == "OCIEmbeddings":
diff --git a/src/locus/rag/embeddings/__init__.py b/src/locus/rag/embeddings/__init__.py
index dea830f..07e48f8 100644
--- a/src/locus/rag/embeddings/__init__.py
+++ b/src/locus/rag/embeddings/__init__.py
@@ -9,6 +9,8 @@
 - OpenAIEmbeddings: OpenAI text-embedding models
 """
 
+from typing import Any
+
 from locus.rag.embeddings.base import (
     BaseEmbedding,
     EmbeddingConfig,
@@ -29,7 +31,7 @@
 ]
 
 
-def __getattr__(name: str):
+def __getattr__(name: str) -> Any:
     """Lazy import providers to avoid requiring all dependencies."""
     if name == "OCIEmbeddings":
         from locus.rag.embeddings.oci import OCIEmbeddings
diff --git a/src/locus/rag/embeddings/oci.py b/src/locus/rag/embeddings/oci.py
index 43c3bf3..b380981 100644
--- a/src/locus/rag/embeddings/oci.py
+++ b/src/locus/rag/embeddings/oci.py
@@ -182,15 +182,19 @@ async def _get_client(self) -> GenerativeAiInferenceClient:
 
             config_file = os.path.expanduser(config_file)
 
-        self._oci_config_dict = oci.config.from_file(
+        # ``oci.config.from_file`` returns the parsed config dict; the
+        # ``_oci_config_dict`` field is declared as ``... | None`` to
+        # represent the pre-init state, so bind to a local before use.
+        config_dict: dict[str, Any] = oci.config.from_file(
             config_file,
             self.oci_config.profile_name,
         )
+        self._oci_config_dict = config_dict
 
         # Determine service endpoint
         endpoint = self.oci_config.service_endpoint
         if endpoint is None:
-            region = self._oci_config_dict.get("region", "us-chicago-1")
+            region = config_dict.get("region", "us-chicago-1")
             endpoint = f"https://inference.generativeai.{region}.oci.oraclecloud.com"
 
         # Determine auth type - respect explicit setting, only auto-detect if needed
@@ -202,21 +206,21 @@ async def _get_client(self) -> GenerativeAiInferenceClient:
         # 3. Config doesn't have user field (api_key profiles have user)
         if (
             auth_type != "api_key"
-            and "security_token_file" in self._oci_config_dict
-            and "user" not in self._oci_config_dict
+            and "security_token_file" in config_dict
+            and "user" not in config_dict
         ):
             auth_type = "security_token"
 
         # Create client based on auth type
         if auth_type == "security_token":
-            token_file = self._oci_config_dict.get("security_token_file")
+            token_file = config_dict.get("security_token_file")
             if token_file:
                 import os as os_module
 
                 token_file = os_module.path.expanduser(token_file)
                 with open(token_file) as f:
                     token = f.read().strip()
-                key_file = os_module.path.expanduser(self._oci_config_dict["key_file"])
+                key_file = os_module.path.expanduser(config_dict["key_file"])
                 private_key = oci.signer.load_private_key_from_file(key_file)
                 signer = oci.auth.signers.SecurityTokenSigner(token, private_key)
                 self._client = GenerativeAiInferenceClient(
@@ -259,7 +263,8 @@ def _get_compartment_id(self) -> str:
         if self.oci_config.compartment_id:
             return self.oci_config.compartment_id
         if self._oci_config_dict:
-            return self._oci_config_dict.get("tenancy", "")
+            tenancy: str = self._oci_config_dict.get("tenancy", "")
+            return tenancy
         return ""
 
     async def embed(self, text: str) -> EmbeddingResult:
diff --git a/src/locus/rag/multimodal.py b/src/locus/rag/multimodal.py
index 370a6a1..82fb3cf 100644
--- a/src/locus/rag/multimodal.py
+++ b/src/locus/rag/multimodal.py
@@ -363,7 +363,7 @@ def __init__(
     ):
         self.use_whisper = use_whisper
         self.whisper_model = whisper_model
-        self._whisper = None
+        self._whisper: Any = None
 
     def supports(self, content_type: ContentType) -> bool:
         return content_type == ContentType.AUDIO
@@ -448,8 +448,9 @@ async def _transcribe_whisper(self, audio_bytes: bytes, audio_format: str) -> st
                 temp_path = f.name
 
             try:
-                result = self._whisper.transcribe(temp_path)
-                return result["text"].strip()
+                result: dict[str, Any] = self._whisper.transcribe(temp_path)
+                text: str = result["text"].strip()
+                return text
             finally:
                 os.unlink(temp_path)
 
@@ -476,7 +477,7 @@ def __init__(
         use_ocr: bool = True,
         use_whisper: bool = True,
     ):
-        self.processors = {
+        self.processors: dict[ContentType, ContentProcessor] = {
             ContentType.TEXT: TextProcessor(),
             ContentType.MARKDOWN: TextProcessor(),
             ContentType.HTML: TextProcessor(),
diff --git a/src/locus/rag/retriever.py b/src/locus/rag/retriever.py
index 4794361..ebd01ae 100644
--- a/src/locus/rag/retriever.py
+++ b/src/locus/rag/retriever.py
@@ -204,7 +204,8 @@ async def add_document(
             documents.append(doc)
 
         # Store all documents
-        return await self.store.add_batch(documents)
+        added: list[str] = await self.store.add_batch(documents)
+        return added
 
     async def add_documents(
         self,
@@ -307,7 +308,8 @@ async def add_file(
             )
             documents.append(doc)
 
-        return await self.store.add_batch(documents)
+        added: list[str] = await self.store.add_batch(documents)
+        return added
 
     async def add_image(
         self,
@@ -345,7 +347,8 @@ async def add_image(
             raw_content=result.raw_content,
         )
 
-        return await self.store.add(doc)
+        doc_added: str = await self.store.add(doc)
+        return doc_added
 
     async def add_pdf(
         self,
@@ -412,7 +415,8 @@ async def add_audio(
             raw_content=result.raw_content,
         )
 
-        return await self.store.add(doc)
+        doc_added: str = await self.store.add(doc)
+        return doc_added
 
     async def retrieve(
         self,
@@ -495,21 +499,24 @@ async def retrieve_text(
 
     async def delete_document(self, doc_id: str) -> bool:
         """Delete a document by ID."""
-        return await self.store.delete(doc_id)
+        deleted: bool = await self.store.delete(doc_id)
+        return deleted
 
     async def clear(self) -> int:
         """Delete all documents."""
-        return await self.store.clear()
+        cleared: int = await self.store.clear()
+        return cleared
 
     async def count(self) -> int:
         """Count documents in store."""
-        return await self.store.count()
+        n: int = await self.store.count()
+        return n
 
     async def close(self) -> None:
         """Close resources."""
         await self.store.close()
 
-    def as_tool(self, name: str = "search_knowledge", description: str | None = None):
+    def as_tool(self, name: str = "search_knowledge", description: str | None = None) -> Any:
         """
         Create a tool function for agent use.
 
diff --git a/src/locus/rag/stores/__init__.py b/src/locus/rag/stores/__init__.py
index d82311c..3688717 100644
--- a/src/locus/rag/stores/__init__.py
+++ b/src/locus/rag/stores/__init__.py
@@ -14,6 +14,8 @@
 - InMemoryVectorStore: In-memory store (testing)
 """
 
+from typing import Any
+
 from locus.rag.stores.base import (
     BaseVectorStore,
     Document,
@@ -41,7 +43,7 @@
 ]
 
 
-def __getattr__(name: str):
+def __getattr__(name: str) -> Any:
     """Lazy import stores to avoid requiring all dependencies."""
     if name == "OracleVectorStore":
         from locus.rag.stores.oracle import OracleVectorStore
diff --git a/src/locus/rag/stores/chroma.py b/src/locus/rag/stores/chroma.py
index 73202f0..2ee0714 100644
--- a/src/locus/rag/stores/chroma.py
+++ b/src/locus/rag/stores/chroma.py
@@ -229,9 +229,11 @@ async def add(self, document: Document) -> str:
             },
         }
 
+        # chromadb's typed shapes ask for ``Sequence[Sequence[float]]``;
+        # we pass the wider ``list[list[float]]`` we have at hand.
         collection.upsert(
             ids=[doc_id],
-            embeddings=[document.embedding],
+            embeddings=[document.embedding],  # type: ignore[arg-type, unused-ignore]
             documents=[document.content],
             metadatas=[metadata],
         )
@@ -269,9 +271,9 @@ async def add_batch(self, documents: list[Document]) -> list[str]:
         if ids:
             collection.upsert(
                 ids=ids,
-                embeddings=embeddings,
+                embeddings=embeddings,  # type: ignore[arg-type, unused-ignore]
                 documents=docs,
-                metadatas=metadatas,
+                metadatas=metadatas,  # type: ignore[arg-type, unused-ignore]
             )
 
         return ids
@@ -291,7 +293,7 @@ async def get(self, doc_id: str) -> Document | None:
         if not result["ids"]:
             return None
 
-        metadata = result["metadatas"][0] if result["metadatas"] else {}
+        metadata: dict[str, Any] = dict(result["metadatas"][0]) if result["metadatas"] else {}
         created_at_str = metadata.pop("created_at", None)
         created_at = datetime.fromisoformat(created_at_str) if created_at_str else datetime.now(UTC)
 
@@ -303,7 +305,7 @@ async def get(self, doc_id: str) -> Document | None:
         return Document(
             id=result["ids"][0],
             content=result["documents"][0] if result["documents"] else "",
-            embedding=embedding,
+            embedding=embedding,  # type: ignore[arg-type, unused-ignore]
             metadata=metadata,
             created_at=created_at,
         )
@@ -334,7 +336,7 @@ async def search(
         collection = self._get_collection()
 
         # Build where filter for metadata
-        where = None
+        where: dict[str, Any] | None = None
         if metadata_filter:
             if len(metadata_filter) == 1:
                 key, value = next(iter(metadata_filter.items()))
@@ -343,7 +345,7 @@ async def search(
                 where = {"$and": [{k: {"$eq": v}} for k, v in metadata_filter.items()]}
 
         result = collection.query(
-            query_embeddings=[query_embedding],
+            query_embeddings=[query_embedding],  # type: ignore[arg-type, unused-ignore]
             n_results=limit,
             where=where,
             include=["embeddings", "documents", "metadatas", "distances"],
@@ -374,7 +376,7 @@ async def search(
             if threshold is not None and score < threshold:
                 continue
 
-            metadata = metadatas[i] if i < len(metadatas) else {}
+            metadata: dict[str, Any] = dict(metadatas[i]) if i < len(metadatas) else {}
             created_at_str = metadata.pop("created_at", None)
             created_at = (
                 datetime.fromisoformat(created_at_str) if created_at_str else datetime.now(UTC)
@@ -383,7 +385,7 @@ async def search(
             doc = Document(
                 id=doc_id,
                 content=documents[i] if i < len(documents) else "",
-                embedding=embeddings[i] if i < len(embeddings) else None,
+                embedding=embeddings[i] if i < len(embeddings) else None,  # type: ignore[arg-type, unused-ignore]
                 metadata=metadata,
                 created_at=created_at,
             )
@@ -401,12 +403,13 @@ async def search(
     async def count(self) -> int:
         """Count documents."""
         collection = self._get_collection()
-        return collection.count()
+        n: int = collection.count()
+        return n
 
     async def clear(self) -> int:
         """Delete all documents."""
         collection = self._get_collection()
-        count = collection.count()
+        count: int = collection.count()
 
         # Delete collection and recreate
         client = self._get_client()
diff --git a/src/locus/rag/stores/opensearch.py b/src/locus/rag/stores/opensearch.py
index 251d43c..0bf8279 100644
--- a/src/locus/rag/stores/opensearch.py
+++ b/src/locus/rag/stores/opensearch.py
@@ -217,7 +217,9 @@ async def add_batch(self, documents: list[Document]) -> list[str]:
         await self._ensure_index()
         client = await self._get_client()
 
-        actions = []
+        # The OpenSearch bulk API alternates control headers and source bodies;
+        # both shapes are dicts with disparate value types, so widen to ``Any``.
+        actions: list[dict[str, Any]] = []
         ids = []
 
         for doc in documents:
@@ -273,7 +275,7 @@ async def delete(self, doc_id: str) -> bool:
         client = await self._get_client()
 
         try:
-            result = await client.delete(
+            result: dict[str, Any] = await client.delete(
                 index=self.os_config.index_name,
                 id=doc_id,
                 refresh=True,
@@ -304,8 +306,9 @@ async def search(
         }
 
         # Add metadata filter if provided
+        query: dict[str, Any]
         if metadata_filter:
-            must_clauses = [knn_query]
+            must_clauses: list[dict[str, Any]] = [knn_query]
             for key, value in metadata_filter.items():
                 must_clauses.append({"term": {f"metadata.{key}": value}})
             query = {"bool": {"must": must_clauses}}
@@ -364,7 +367,8 @@ async def count(self) -> int:
         client = await self._get_client()
 
         result = await client.count(index=self.os_config.index_name)
-        return result["count"]
+        n: int = result["count"]
+        return n
 
     async def clear(self) -> int:
         """Delete all documents."""
diff --git a/src/locus/rag/stores/oracle.py b/src/locus/rag/stores/oracle.py
index c20161f..2220265 100644
--- a/src/locus/rag/stores/oracle.py
+++ b/src/locus/rag/stores/oracle.py
@@ -376,7 +376,7 @@ async def delete(self, doc_id: str) -> bool:
                 f"DELETE FROM {self._full_table_name} WHERE id = :id",
                 {"id": doc_id},
             )
-            deleted = cursor.rowcount > 0
+            deleted: bool = cursor.rowcount > 0
             await conn.commit()
 
         return deleted
diff --git a/src/locus/rag/stores/pgvector.py b/src/locus/rag/stores/pgvector.py
index 76bb4eb..b7059e3 100644
--- a/src/locus/rag/stores/pgvector.py
+++ b/src/locus/rag/stores/pgvector.py
@@ -370,7 +370,7 @@ async def delete(self, doc_id: str) -> bool:
         pool = await self._get_pool()
 
         async with pool.acquire() as conn:
-            result = await conn.execute(
+            result: str = await conn.execute(
                 f"""
                 DELETE FROM {self._full_table_name}
                 WHERE id = $1
@@ -590,12 +590,13 @@ async def has_index(self) -> bool:
         table_name = self.pgvector_config.table_name
 
         async with pool.acquire() as conn:
-            return await conn.fetchval(f"""
+            exists: bool = await conn.fetchval(f"""
                 SELECT EXISTS (
                     SELECT 1 FROM pg_indexes
                     WHERE indexname = 'idx_{table_name}_embedding'
                 )
             """)
+            return exists
 
     async def close(self) -> None:
         """Close connection pool."""
diff --git a/src/locus/rag/stores/pinecone.py b/src/locus/rag/stores/pinecone.py
index 3610e02..e041f1a 100644
--- a/src/locus/rag/stores/pinecone.py
+++ b/src/locus/rag/stores/pinecone.py
@@ -380,9 +380,11 @@ async def count(self) -> int:
 
         if self.pinecone_config.namespace:
             ns_stats = stats.namespaces.get(self.pinecone_config.namespace, {})
-            return ns_stats.get("vector_count", 0)
+            ns_count: int = ns_stats.get("vector_count", 0)
+            return ns_count
 
-        return stats.total_vector_count
+        total: int = stats.total_vector_count
+        return total
 
     async def clear(self) -> int:
         """Delete all documents."""
diff --git a/src/locus/rag/stores/qdrant.py b/src/locus/rag/stores/qdrant.py
index 83f57c7..42e16f2 100644
--- a/src/locus/rag/stores/qdrant.py
+++ b/src/locus/rag/stores/qdrant.py
@@ -308,7 +308,7 @@ async def get(self, doc_id: str) -> Document | None:
         return Document(
             id=payload.get("doc_id", str(point.id)),
             content=payload.get("content", ""),
-            embedding=list(point.vector) if point.vector else None,
+            embedding=list(point.vector) if point.vector else None,  # type: ignore[arg-type, unused-ignore]
             metadata=payload.get("metadata", {}),
             created_at=datetime.fromisoformat(payload["created_at"])
             if payload.get("created_at")
@@ -358,15 +358,19 @@ async def search(
             except ImportError:
                 pass
             else:
-                conditions = []
-                for key, value in metadata_filter.items():
-                    conditions.append(
-                        FieldCondition(
-                            key=f"metadata.{key}",
-                            match=MatchValue(value=value),
-                        )
+                # ``Filter.must`` is typed as the broad union of all possible
+                # condition kinds; ``list[FieldCondition]`` is invariant and
+                # would be rejected. Sequence is covariant and accepted.
+                from collections.abc import Sequence
+
+                conditions: Sequence[FieldCondition] = [
+                    FieldCondition(
+                        key=f"metadata.{key}",
+                        match=MatchValue(value=value),
                     )
-                query_filter = Filter(must=conditions)
+                    for key, value in metadata_filter.items()
+                ]
+                query_filter = Filter(must=list(conditions))
 
         # Search using query_points (newer API)
 
@@ -378,8 +382,18 @@ async def search(
             with_vectors=True,
         )
 
-        # Get points from result
-        points = search_result.points if hasattr(search_result, "points") else search_result
+        # ``query_points`` returns a ``QueryResponse`` whose ``.points`` is
+        # ``list[ScoredPoint]``; older builds fall through to the bare
+        # response object. Either way, the loop body below uses
+        # ScoredPoint-shape attributes.
+        from typing import cast as _cast
+
+        from qdrant_client.models import ScoredPoint as _ScoredPoint
+
+        points: list[_ScoredPoint] = _cast(
+            "list[_ScoredPoint]",
+            search_result.points if hasattr(search_result, "points") else search_result,
+        )
 
         results = []
         for point in points:
@@ -394,7 +408,7 @@ async def search(
             doc = Document(
                 id=payload.get("doc_id", str(point.id)),
                 content=payload.get("content", ""),
-                embedding=list(point.vector) if point.vector else None,
+                embedding=list(point.vector) if point.vector else None,  # type: ignore[arg-type, unused-ignore]
                 metadata=payload.get("metadata", {}),
                 created_at=datetime.fromisoformat(payload["created_at"])
                 if payload.get("created_at")
diff --git a/src/locus/rag/tools.py b/src/locus/rag/tools.py
index 28b9a24..19552c6 100644
--- a/src/locus/rag/tools.py
+++ b/src/locus/rag/tools.py
@@ -39,7 +39,7 @@ def create_rag_tool(
     description: str | None = None,
     limit: int = 5,
     threshold: float | None = 0.5,
-):
+) -> Any:
     """
     Create a RAG search tool for agent use.
 
@@ -125,7 +125,7 @@ def create_rag_context_tool(
     name: str = "get_context",
     description: str | None = None,
     limit: int = 3,
-):
+) -> Any:
     """
     Create a RAG tool that returns context as formatted text.
 
@@ -207,7 +207,7 @@ def __init__(
         self.retriever = retriever
         self.prefix = prefix
 
-    def get_tools(self) -> list:
+    def get_tools(self) -> list[Any]:
         """Get all RAG tools."""
         return [
             self.search_tool(),
@@ -215,7 +215,7 @@ def get_tools(self) -> list:
             self.lookup_tool(),
         ]
 
-    def search_tool(self):
+    def search_tool(self) -> Any:
         """Get the search tool."""
         return create_rag_tool(
             self.retriever,
@@ -223,7 +223,7 @@ def search_tool(self):
             description="Search the knowledge base for relevant documents.",
         )
 
-    def context_tool(self):
+    def context_tool(self) -> Any:
         """Get the context tool."""
         return create_rag_context_tool(
             self.retriever,
@@ -231,7 +231,7 @@ def context_tool(self):
             description="Get formatted context from the knowledge base.",
         )
 
-    def lookup_tool(self):
+    def lookup_tool(self) -> Any:
         """Get the lookup tool."""
         from locus.tools import tool as tool_decorator