Summarize progress counts reasoning tokens: thinking models stream most of the run in reasoning_content, which the ticker ignored — UI sat at 'waiting for model' then jumped to done with no percentage

backend/shonar/services/ai/ollama.py

@@ -117,9 +117,13 @@ async def _collect_reply(resp: httpx.Response, ticker) -> str:
         except ValueError:
             continue
         piece = (obj.get("message") or {}).get("content") or ""
-        if piece:
+        # Like the openai_compat adapter: a thinking channel is real
+        # work and must move the progress ticker even when the server
+        # ignored think:false.
+        thinking = (obj.get("message") or {}).get("thinking") or ""
+        if piece or thinking:
             parts.append(piece)
-            total += len(piece)
+            total += len(piece) + len(thinking)
             ticker(total)
         if obj.get("done"):
             break

backend/shonar/services/ai/openai_compat.py

@@ -48,11 +48,18 @@ async def _collect_reply(resp: httpx.Response, ticker) -> str:
             chunk = _json.loads(data)
             delta = chunk["choices"][0].get("delta") or {}
             piece = delta.get("content") or ""
+            # Thinking models (qwen3 on llama.cpp) stream a long
+            # reasoning_content channel BEFORE any content: counting only
+            # content left the progress bar frozen at "waiting for model"
+            # for the entire run, then jumping straight to done. Reasoning
+            # is real work — count it toward progress (never into the
+            # reply itself).
+            thinking = delta.get("reasoning_content") or ""
         except (ValueError, KeyError, IndexError, TypeError):
             continue  # keep-alives / usage chunks / odd frames: not content
-        if piece:
+        if piece or thinking:
             parts.append(piece)
-            total += len(piece)
+            total += len(piece) + len(thinking)
             ticker(total)
     content = "".join(parts)
     if not content: