feat(api): replace SSE chat with sync POST /api/search

src/mediastorm/api.py

@@ -6,16 +6,13 @@ from typing import Any
 import httpx
 from fastapi import FastAPI
 from pydantic import BaseModel
-from sse_starlette.sse import EventSourceResponse
-
 from mediastorm.config import CHROMADB_PATH, BM25_INDEX_PATH, INGEST_CACHE_DIR
 from mediastorm.vectorize.store import VectorStore
 from mediastorm.vectorize.embedder import Embedder
 from mediastorm.vectorize.bm25_store import BM25Store
 from mediastorm.rag.router import QueryRouter
 from mediastorm.rag.retriever import HybridRetriever
-from mediastorm.rag.generator import generate_response_stream
-from mediastorm.rag.rewriter import rewrite_if_needed
+from mediastorm.rag.generator import generate_response
 
 store: VectorStore | None = None
 embedder: Embedder | None = None
@@ -155,10 +152,8 @@ async def lifespan(app: FastAPI):
 app = FastAPI(title="MediaStorm Archive Explorer", lifespan=lifespan)
 
 
-class ChatRequest(BaseModel):
+class SearchRequest(BaseModel):
     query: str
-    history: list[dict] = []
-    filters: dict = {}
 
 
 def _extract_story_cards(stories: list[dict]) -> list[dict]:
@@ -202,44 +197,19 @@ async def stories():
     return {"stories": enriched}
 
 
-@app.post("/api/chat")
-async def chat(req: ChatRequest):
-    async def event_generator():
-        # Convert history to Gemini format
-        chat_history = []
-        for msg in (req.history or []):
-            role = msg.get("role", "user")
-            content = msg.get("content", "")
-            if role == "user":
-                chat_history.append({"role": "user", "parts": [{"text": content}]})
-            else:
-                chat_history.append({"role": "model", "parts": [{"text": content}]})
-
-        # Rewrite query for retrieval if it's a follow-up with implicit references
-        rewrite_history = req.history or []
-        search_query = await rewrite_if_needed(req.query, rewrite_history)
-
-        ui_filters = req.filters if req.filters else None
-        result = await retriever.retrieve(search_query, ui_filters=ui_filters)
-
-        all_cards = _extract_story_cards(result.stories)
-
-        # Stream text first, accumulate full response
-        full_text = ""
-        async for text in generate_response_stream(
-            req.query, result, chat_history or None, link_lookup=_link_lookup,
-        ):
-            full_text = text
-            yield {"event": "text", "data": json.dumps({"content": text})}
-
-        # Only show cards for stories Gemini actually cited (by link URL in text)
-        cited_cards = [
-            c for c in all_cards
-            if c.get("link") and c["link"] in full_text
-        ]
-        if cited_cards:
-            yield {"event": "stories", "data": json.dumps({"stories": cited_cards})}
-
-        yield {"event": "done", "data": "{}"}
-
-    return EventSourceResponse(event_generator())
+@app.post("/api/search")
+async def search(req: SearchRequest):
+    result = await retriever.retrieve(req.query)
+    all_cards = _extract_story_cards(result.stories)
+
+    # Use Gemini as relevance filter — generate text, keep only cited stories
+    full_text = await generate_response(
+        req.query, result, link_lookup=_link_lookup,
+    )
+
+    cited_cards = [
+        c for c in all_cards
+        if c.get("link") and c["link"] in full_text
+    ]
+
+    return {"stories": cited_cards}

src/mediastorm/rag/retriever.py

@@ -21,7 +21,7 @@ class HybridRetriever:
         embedder: Embedder,
         router: QueryRouter,
         top_k_retrieval: int = 50,
-        top_k_final: int = 4,
+        top_k_final: int = 8,
     ):
         self._store = vector_store
         self._bm25 = bm25_store

tests/test_api.py

@@ -6,13 +6,27 @@ from mediastorm.api import app
 
 
 @pytest.mark.asyncio
-async def test_chat_requires_query():
+async def test_search_requires_query():
     transport = ASGITransport(app=app)
     async with AsyncClient(transport=transport, base_url="http://test") as client:
-        resp = await client.post("/api/chat", json={})
+        resp = await client.post("/api/search", json={})
     assert resp.status_code == 422
 
 
+@pytest.mark.asyncio
+async def test_search_returns_stories():
+    """Integration test — requires GEMINI_API_KEY and populated ChromaDB."""
+    if not os.environ.get("GEMINI_API_KEY"):
+        pytest.skip("GEMINI_API_KEY not set")
+    transport = ASGITransport(app=app)
+    async with AsyncClient(transport=transport, base_url="http://test") as client:
+        resp = await client.post("/api/search", json={"query": "Emmy winning documentaries"})
+    assert resp.status_code == 200
+    data = resp.json()
+    assert "stories" in data
+    assert isinstance(data["stories"], list)
+
+
 @pytest.mark.asyncio
 async def test_health():
     transport = ASGITransport(app=app)