fix(openai): validate embedding input before API call

Author: pandego

semantic_router/encoders/openai.py

@@ -118,6 +118,33 @@ def __init__(
         # get token encoder
         self._token_encoder = tiktoken.encoding_for_model(name)
 
+    def _validate_docs(self, docs: List[str]) -> List[str]:
+        """Validate embedding input and return normalized docs list."""
+        if not isinstance(docs, list):
+            raise ValueError(
+                "OpenAI encoder input must be a list of strings, got "
+                f"{type(docs).__name__}."
+            )
+        if len(docs) == 0:
+            raise ValueError("OpenAI encoder input cannot be empty.")
+
+        normalized_docs: List[str] = []
+        for idx, doc in enumerate(docs):
+            if not isinstance(doc, str):
+                raise ValueError(
+                    "OpenAI encoder input must contain only strings, got "
+                    f"{type(doc).__name__} at index {idx}."
+                )
+            cleaned = doc.strip()
+            if cleaned == "":
+                raise ValueError(
+                    "OpenAI encoder input contains an empty string at "
+                    f"index {idx}."
+                )
+            normalized_docs.append(cleaned)
+
+        return normalized_docs
+
     def __call__(self, docs: List[str], truncate: bool = True) -> List[List[float]]:
         """Encode a list of text documents into embeddings using OpenAI API.
 
@@ -130,6 +157,8 @@ def __call__(self, docs: List[str], truncate: bool = True) -> List[List[float]]:
             raise ValueError("OpenAI client is not initialized.")
         embeds = None
 
+        docs = self._validate_docs(docs)
+
         if truncate:
             # check if any document exceeds token limit and truncate if so
             docs = [self._truncate(doc) for doc in docs]
@@ -202,6 +231,8 @@ async def acall(self, docs: List[str], truncate: bool = True) -> List[List[float
             raise ValueError("OpenAI async client is not initialized.")
         embeds = None
 
+        docs = self._validate_docs(docs)
+
         if truncate:
             # check if any document exceeds token limit and truncate if so
             docs = [self._truncate(doc) for doc in docs]

tests/unit/encoders/test_openai.py

@@ -95,6 +95,24 @@ def test_openai_encoder_call_failure_non_openai_error(self, openai_encoder, mock
 
         assert "OpenAI API call failed. Error: Non-OpenAIError" in str(e.value)
 
+    def test_openai_encoder_call_rejects_empty_input(self, openai_encoder):
+        with pytest.raises(ValueError) as e:
+            openai_encoder([])
+
+        assert "OpenAI encoder input cannot be empty." in str(e.value)
+
+    def test_openai_encoder_call_rejects_blank_string_input(self, openai_encoder):
+        with pytest.raises(ValueError) as e:
+            openai_encoder(["   "])
+
+        assert "contains an empty string" in str(e.value)
+
+    def test_openai_encoder_call_rejects_non_string_input(self, openai_encoder):
+        with pytest.raises(ValueError) as e:
+            openai_encoder(["ok", 123])  # type: ignore[list-item]
+
+        assert "must contain only strings" in str(e.value)
+
     def test_openai_encoder_call_successful_retry(self, openai_encoder, mocker):
         mock_embeddings = mocker.Mock()
         mock_embeddings.data = [