fix: skip index build when a source produces no text chunks

fenn/agents/rag/retriever.py

@@ -262,6 +262,12 @@ def index(self, docs):
             for chunk in chunk_text(doc, mode=self.chunk_mode):
                 self.chunks.append(chunk)
 
+        if not self.chunks:
+            logger.warning(
+                "[cofone] nothing to index: the source(s) produced no text chunks."
+            )
+            return
+
         if self.use_faiss:
             self._build_faiss()
             if self.persist_path:
@@ -344,6 +350,8 @@ def query(self, text, top_k=5):
         return self._query_bm25(text, top_k)
 
     def _query_faiss(self, text, top_k):
+        if self._faiss_index is None or not self.chunks:
+            return []
         faiss_mod = self._get_faiss()
         vec = self._embed([text], input_type="search_query")
         faiss_mod.normalize_L2(vec)

tests/unit/agents/test_retrievers.py

@@ -215,6 +215,31 @@ def test_build_faiss_raises_without_faiss(self):
                 r._build_faiss()
 
 
+class TestIndexWithoutChunks:
+    def test_faiss_index_with_no_docs_does_not_build(self):
+        r = Retriever(use_faiss=True)
+        with patch.object(r, "_build_faiss") as build:
+            r.index([])
+        build.assert_not_called()
+        assert r._faiss_index is None
+
+    def test_faiss_index_with_only_blank_docs_does_not_build(self):
+        r = Retriever(use_faiss=True)
+        with patch.object(r, "_build_faiss") as build:
+            r.index(["   \n\n  "])
+        build.assert_not_called()
+        assert r.chunks == []
+
+    def test_bm25_index_with_no_docs_does_not_raise(self):
+        r = Retriever(use_faiss=False)
+        r.index([])
+        assert r.query("anything") == []
+
+    def test_query_faiss_returns_empty_when_index_never_built(self):
+        r = Retriever(use_faiss=True)
+        assert r.query("anything", top_k=5) == []
+
+
 # ── Embedding method ImportError paths ────────────────────────────────────────