Update app.py

app.py

@@ -5,6 +5,7 @@ from google.genai import types
 from pymongo import UpdateOne
 from pymongo import MongoClient
 from bson import ObjectId
+import base64
 import bcrypt
 import os
 import re
@@ -106,6 +107,14 @@ Mode: English-guided learning.
   pushing.
 - Correct mistakes gently and explain the grammar point in English."""
 
+def parse_data_url(data_url):
+    try:
+        header, b64data = data_url.split(",", 1)
+        mime = header.split(":")[1].split(";")[0]
+        return mime, base64.b64decode(b64data)
+    except Exception:
+        return None, None
+
 
 def normalize_word(w):
     return re.sub(r"[^\w']", "", w, flags=re.UNICODE).strip().lower()
@@ -164,12 +173,18 @@ def build_config(mode, pronunciation_notes=""):
     )
 
 
-def call_intyoza(message, history, mode):
+def call_intyoza(message, history, mode, image_data_url=None):
     contents = []
     for turn in history:
         role = "user" if turn["role"] == "user" else "model"
         contents.append(types.Content(role=role, parts=[types.Part(text=str(turn["content"]))]))
-    contents.append(types.Content(role="user", parts=[types.Part(text=message)]))
+
+    parts = [types.Part(text=message)]
+    if image_data_url:
+        mime, raw = parse_data_url(image_data_url)
+        if raw:
+            parts.append(types.Part.from_bytes(data=raw, mime_type=mime))
+    contents.append(types.Content(role="user", parts=parts))
 
     config = build_config(mode, lookup_pronunciation_notes(message, history))
 
@@ -457,14 +472,17 @@ def send_message():
     message = (data.get("message") or "").strip()
     mode = data.get("mode", "Kinyarwanda immersion")
     chat_id = data.get("chat_id")
+    image_data_url = data.get("image")
 
-    if not message:
+    if not message and not image_data_url:
         return jsonify(ok=False, message="Empty message."), 400
+    if image_data_url and len(image_data_url) > 4_000_000:
+        return jsonify(ok=False, message="Image is too large (4MB max)."), 400
 
     username = require_login()
 
     if not (username and chat_id):
-        reply = call_intyoza(message, data.get("history", []), mode)
+        reply = call_intyoza(message or "What is in this image?", data.get("history", []), mode, image_data_url)
         return jsonify(ok=True, reply=reply)
 
     chat_doc = chats_col.find_one({"_id": ObjectId(chat_id), "username": username})
@@ -476,24 +494,24 @@ def send_message():
         for d in messages_col.find({"chat_id": chat_id}).sort("timestamp", 1)
     ]
 
-    reply = call_intyoza(message, history, mode)
+    reply = call_intyoza(message or "What is in this image?", history, mode, image_data_url)
     now = datetime.datetime.utcnow()
 
+    stored_message = message or "[Sent an image]"
     user_doc = messages_col.insert_one(
-        {"chat_id": chat_id, "role": "user", "content": message, "timestamp": now})
+        {"chat_id": chat_id, "role": "user", "content": stored_message, "timestamp": now})
     bot_doc = messages_col.insert_one(
         {"chat_id": chat_id, "role": "assistant", "content": reply,
          "timestamp": now + datetime.timedelta(milliseconds=1)})
 
     full_history = history + [
-        {"role": "user", "content": message},
+        {"role": "user", "content": stored_message},
         {"role": "assistant", "content": reply},
     ]
     exchanges = len(full_history) // 2
     stage = chat_doc.get("auto_titled_at", 0)
 
     new_title = None
-    # Title after the first exchange, then refine once the conversation has shape.
     if stage < 999 and (exchanges == 1 or (exchanges >= 4 and stage < 4)):
         candidate = generate_title(snippet_from(full_history))
         if candidate: