Upload app.py

app.py

@@ -277,7 +277,13 @@ def gemini_reply(client, convo, n_followups):
     cfg = types.GenerateContentConfig(
         system_instruction=sys,
         temperature=0.3,          # lower temp -> stays closer to the sources
-        max_output_tokens=2048,
+        # gemini-2.5-flash has "thinking" on by default, and those thinking
+        # tokens are drawn from max_output_tokens. That could exhaust the budget
+        # and truncate the visible answer mid-sentence (e.g. the long treatment-
+        # options response). Disable thinking (budget=0) so the whole budget goes
+        # to the answer, and raise the ceiling for extra headroom.
+        max_output_tokens=8192,
+        thinking_config=types.ThinkingConfig(thinking_budget=0),
     )
     resp = client.models.generate_content(model=MODEL, contents=contents, config=cfg)
     return (resp.text or "").strip()