fix: default SDK retrieve_model to gpt-5.4 instead of following model

The local SDK path builds IndexConfig directly (not from config.yaml), so
retrieve_model defaulting to None made agentic QA silently follow the cheaper
indexing model (gpt-4o) instead of the stronger gpt-5.4 the packaged config
and CLI use. Set the IndexConfig field default to gpt-5.4; retrieve_model=None
still follows model as an explicit escape hatch. Cloud path and the
index-only CLI are unaffected.

pageindex/client.py

@@ -35,7 +35,7 @@ class PageIndexClient:
         api_key: PageIndex cloud API key. When provided, cloud mode is used
             and local-only params (model, storage_path, index_config, …) are ignored.
         model: LLM model for indexing (local mode only, default: gpt-4o-2024-11-20).
-        retrieve_model: LLM model for agent QA (local mode only, default: same as model).
+        retrieve_model: LLM model for agent QA (local mode only, default: gpt-5.4).
         storage_path: Directory for SQLite DB and files (local mode only, default: ./.pageindex).
         storage: Custom StorageEngine instance (local mode only).
         index_config: Advanced indexing parameters (local mode only, optional).
@@ -312,7 +312,7 @@ class LocalClient(PageIndexClient):
 
     Args:
         model: LLM model for indexing (default: gpt-4o-2024-11-20)
-        retrieve_model: LLM model for agent QA (default: same as model)
+        retrieve_model: LLM model for agent QA (default: gpt-5.4)
         storage_path: Directory for SQLite DB and files (default: ./.pageindex)
         storage: Custom StorageEngine instance (default: SQLiteStorage)
         index_config: Advanced indexing parameters. Pass an IndexConfig instance

pageindex/config.py

@@ -18,7 +18,7 @@ class IndexConfig(BaseModel):
     model_config = {"extra": "forbid"}
 
     model: str = "gpt-4o-2024-11-20"
-    retrieve_model: str | None = None
+    retrieve_model: str | None = "gpt-5.4"  # None = follow `model`
     toc_check_page_num: int = 20
     max_page_num_each_node: int = 10
     max_token_num_each_node: int = 20000

pageindex/config.yaml

@@ -1,6 +1,6 @@
 model: "gpt-4o-2024-11-20"
 # model: "anthropic/claude-sonnet-4-6"
-retrieve_model: "gpt-5.4"  # defaults to `model` if not set
+retrieve_model: "gpt-5.4"  # set to null to follow `model`
 toc_check_page_num: 20
 max_page_num_each_node: 10
 max_token_num_each_node: 20000

tests/test_client.py

@@ -69,6 +69,25 @@ def test_delete_collection(tmp_path):
     assert "papers" not in client.list_collections()
 
 
+def test_retrieve_model_defaults_to_strong_reasoner(tmp_path):
+    """Retrieval must not silently follow the (cheaper) indexing model."""
+    client = LocalClient(model="gpt-4o", storage_path=str(tmp_path / "pi"))
+    assert client._backend.get_retrieve_model() == "gpt-5.4"
+
+
+def test_explicit_retrieve_model_wins(tmp_path):
+    client = LocalClient(model="gpt-4o", retrieve_model="ollama/llama3",
+                         storage_path=str(tmp_path / "pi"))
+    assert client._backend.get_retrieve_model() == "litellm/ollama/llama3"
+
+
+def test_retrieve_model_none_follows_model(tmp_path):
+    from pageindex.config import IndexConfig
+    client = LocalClient(model="gpt-4o", storage_path=str(tmp_path / "pi"),
+                         index_config=IndexConfig(retrieve_model=None))
+    assert client._backend.get_retrieve_model() == "gpt-4o"
+
+
 def test_register_parser(tmp_path):
     client = LocalClient(model="gpt-4o", storage_path=str(tmp_path / "pi"))
     class FakeParser:

tests/test_config.py

@@ -6,7 +6,7 @@ from pageindex.config import IndexConfig
 def test_defaults():
     config = IndexConfig()
     assert config.model == "gpt-4o-2024-11-20"
-    assert config.retrieve_model is None
+    assert config.retrieve_model == "gpt-5.4"
     assert config.toc_check_page_num == 20