fix issue with responses in JSON format

app.py

@@ -16,6 +16,27 @@ if not api_key:
 client = ClientV2(api_key=api_key, client_name="hf-tiny-aya-global")
 
 
+def _extract_text(content: object) -> str:
+    """Extract plain text from any Cohere content shape.
+
+    Handles plain strings, objects with a `.text` attribute,
+    and lists of content blocks (e.g. [{'text': '...', 'type': 'text'}]).
+    """
+    if content is None:
+        return ""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = [_extract_text(block) for block in content]
+        return "".join(parts)
+    text = getattr(content, "text", None)
+    if text is not None:
+        return str(text)
+    if isinstance(content, dict):
+        return str(content.get("text", ""))
+    return ""
+
+
 def generate(
     message: str,
     history: list[dict],
@@ -29,16 +50,12 @@ def generate(
     if system_prompt:
         messages.append({"role": "system", "content": system_prompt})
 
-    # Add conversation history
     for item in history:
         role = item.get("role")
-        content = item.get("content") or ""
-        if not isinstance(content, str):
-            content = str(content)
+        content = _extract_text(item.get("content"))
         if role in ("assistant", "user") and content:
             messages.append({"role": role, "content": content})
 
-    # Add current user message
     current_text = str(message or "").strip()
     if not current_text:
         yield ""
@@ -46,7 +63,6 @@ def generate(
     messages.append({"role": "user", "content": current_text})
 
     try:
-        # Call Cohere API using the correct event type and delta access
         response = client.chat_stream(
             model=model_id,
             messages=messages,
@@ -56,11 +72,17 @@ def generate(
 
         output = ""
         for event in response:
-            if getattr(event, "type", None) == "content-delta":
-                # event.delta.message.content.text is the streamed text
-                text = getattr(event.delta.message.content, "text", "")
-                output += text
-                yield output
+            if getattr(event, "type", None) in ("content-delta", "content-start"):
+                delta = getattr(event, "delta", None)
+                if delta is None:
+                    continue
+                msg = getattr(delta, "message", None)
+                if msg is None:
+                    continue
+                text = _extract_text(getattr(msg, "content", None))
+                if text:
+                    output += text
+                    yield output
 
     except Exception:
         logger.exception("Cohere API error")