Update app.py

app.py

@@ -726,98 +726,61 @@ def chat():
     # ======================
     # 🖼️ VISION MODE (AUTO DETECT - BASE64 ONLY)
     # ======================
-    image_base64 = data.get("image_base64")
-    if image_base64:
-        print("[VISION] 🖼️ Image detected → using c4ai-aya-vision-32b (Cohere)")
+    # ======================
+# 🖼️ VISION MODE (Cohere v2) — TEXT ONLY
+# ======================
+image_base64 = data.get("image_base64")
+if image_base64:
+    print("[VISION] 🖼️ Image detected → Cohere c4ai-aya-vision-32b")
 
-        # Validate base64 quickly
-        try:
-            # If user sends full data URI like "data:image/png;base64,....", strip prefix for decoding test
-            test_b64 = image_base64
-            if test_b64.startswith("data:"):
-                test_b64 = test_b64.split(",", 1)[1]
-            base64.b64decode(test_b64, validate=True)
-        except Exception as e:
-            print(f"[VISION] ❌ Invalid base64: {e}")
-            return jsonify({"error": "Invalid base64 image"}), 400
-
-        # Build payload for Cohere v2 chat (vision)
-        cohere_url = "https://api.cohere.ai/v2/chat"
-        payload = {
-            "model": "c4ai-aya-vision-32b",
-            "messages": [
-                {
-                    "role": "user",
-                    "content": [
-                        {
-                            "type": "text",
-                            "text": prompt if prompt else "Describe this image."
-                        },
-                        {
-                            # Cohere accepts data URI as image url - keep the data URI if present
-                            "type": "image_url",
-                            "image_url": {
-                                "url": image_base64,
-                                "detail": "auto"
-                            }
+    # validate base64
+    try:
+        test_b64 = image_base64.split(",", 1)[1] if image_base64.startswith("data:") else image_base64
+        base64.b64decode(test_b64, validate=True)
+    except Exception:
+        return Response("Invalid base64 image", mimetype="text/plain", status=400)
+
+    cohere_url = "https://api.cohere.ai/v2/chat"
+    payload = {
+        "model": "c4ai-aya-vision-32b",
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": prompt or "Describe this image."},
+                    {
+                        "type": "image_url",
+                        "image_url": {
+                            "url": image_base64,
+                            "detail": "auto"
                         }
-                    ]
-                }
-            ],
-            
-        }
-
-        headers = {
-            "Authorization": f"Bearer {COHERE_API_KEY}",
-            "Content-Type": "application/json"
-        }
-
-        try:
-            res = requests.post(cohere_url, json=payload, headers=headers, timeout=60)
-            try:
-                res_json = res.json()
-            except Exception:
-                print("[VISION] ❌ Cohere returned non-json response")
-                return jsonify({"error": "Cohere returned non-json response", "status_code": res.status_code, "text": res.text}), 500
+                    }
+                ]
+            }
+        ]
+    }
 
-            # Robust extraction of text reply (support several possible shapes)
-            ai_reply = None
-            # Try common v2 shape: res_json["messages"][0]["content"][0]["text"]
-            try:
-                ai_reply = res_json.get("messages", [])[0].get("content", [])[0].get("text")
-            except Exception:
-                ai_reply = None
-
-            # Fallback: sometimes Cohere returns 'output' or 'text' fields
-            if not ai_reply:
-                # Try to find any 'text' in nested dicts
-                def find_text(obj):
-                    if isinstance(obj, dict):
-                        for k, v in obj.items():
-                            if k == "text" and isinstance(v, str):
-                                return v
-                            else:
-                                found = find_text(v)
-                                if found:
-                                    return found
-                    elif isinstance(obj, list):
-                        for item in obj:
-                            found = find_text(item)
-                            if found:
-                                return found
-                    return None
-                ai_reply = find_text(res_json) or ""
-
-            return jsonify({
-                "mode": "vision",
-                "reply": ai_reply,
-                "raw": res_json
-            })
+    headers = {
+        "Authorization": f"Bearer {COHERE_API_KEY}",
+        "Content-Type": "application/json"
+    }
 
-        except requests.exceptions.RequestException as e:
-            print(f"[VISION] ❌ RequestException: {e}")
-            return jsonify({"error": "Failed to call Cohere", "details": str(e)}), 500
+    res = requests.post(cohere_url, json=payload, headers=headers, timeout=60)
 
+    try:
+        res_json = res.json()
+    except Exception:
+        return Response("Cohere non-json response", mimetype="text/plain", status=500)
+
+    # 🔥 AMBIL TEKS SAJA
+    ai_text = (
+        res_json
+        .get("message", {})
+        .get("content", [{}])[0]
+        .get("text", "")
+    )
+
+    return Response(ai_text, mimetype="text/plain")
     # =====================================================
     # 🧩 🎚️ (VISION DONE) — LANJUTKAN MODE TEXT SEPERTI BIASA
     # =====================================================