Update routers/editor.py

routers/editor.py

@@ -122,6 +122,28 @@ EASING_TYPES = [
 # editorAgent.ts's stream-reading helper on the frontend, which expects the
 # exact same shape.
 
+# Headers asking any reverse proxy in front of this app (Hugging Face
+# Spaces' Docker gateway, in particular) to NOT buffer the response until
+# it's complete. `StreamingResponse` already streams correctly out of
+# uvicorn on its own (confirmed: no compression/buffering middleware in
+# main.py) — but a hop between here and the browser can still coalesce
+# everything into one delivery despite that, which reads exactly like "no
+# streaming" client-side even though the server sent it live.
+# `X-Accel-Buffering: no` is the standard nginx-family opt-out (harmless to
+# send even if the fronting proxy isn't nginx-based); Cache-Control/
+# Connection reinforce the same intent for any other caching layer. If a
+# proxy hop still buffers despite these, that's a platform-level behavior no
+# response header can override — `curl -N <url>` straight against the
+# deployed endpoint (bypassing the browser entirely) is the way to tell
+# "backend isn't streaming" apart from "something downstream is buffering
+# it anyway".
+_NO_BUFFER_HEADERS = {
+    "X-Accel-Buffering": "no",
+    "Cache-Control": "no-cache, no-transform",
+    "Connection": "keep-alive",
+}
+
+
 def _cleanup_temp_files(*paths: Optional[str]) -> None:
     """Shared cleanup for every temp file this router creates — dedups via
     `set()` since a couple of these paths can legitimately be the same file
@@ -396,7 +418,7 @@ async def editor_chat(request: EditorChatRequest):
                 save_session(request.chat_id, chat.metadata)
             _cleanup_temp_files(context_path)
 
-    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8")
+    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8", headers=_NO_BUFFER_HEADERS)
 
 
 # ── Smart video cut (Fase 2) ────────────────────────────────────────────────
@@ -539,7 +561,7 @@ async def editor_smart_cut(request: SmartCutRequest):
             async def empty_body():
                 yield '{"segments": []}\n' + "I couldn't find any speech in this clip to analyze."
 
-            return StreamingResponse(empty_body(), media_type="text/plain; charset=utf-8")
+            return StreamingResponse(empty_body(), media_type="text/plain; charset=utf-8", headers=_NO_BUFFER_HEADERS)
 
         print(f"🕒 [SMART-CUT] Transcrição em tempo absoluto ({len(transcript_blocks)} blocos):")
         for b in transcript_blocks:
@@ -589,7 +611,7 @@ async def editor_smart_cut(request: SmartCutRequest):
         finally:
             _cleanup_temp_files(original_media_path, context_path)
 
-    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8")
+    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8", headers=_NO_BUFFER_HEADERS)
 
 
 # ── Media analysis (analyze_media) ─────────────────────────────────────────
@@ -740,4 +762,4 @@ async def editor_analyze_media(
         finally:
             _cleanup_temp_files(upload_path, gemini_filepath, context_path)
 
-    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8")
+    return StreamingResponse(body_stream(), media_type="text/plain; charset=utf-8", headers=_NO_BUFFER_HEADERS)