fix: Handle encrypted images + link text to recent images

- Add RoomEncryptedImage callback with decrypt_attachment for E2E rooms
- Cache recent images per room (60s TTL) so follow-up text messages
  like "was ist das" get the image context instead of hallucinating
- Treat filenames (containing dots) as no-caption, default to
  "What's in this image?"

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

bot.py

@@ -19,6 +19,7 @@ from nio import (
     LoginResponse,
     InviteMemberEvent,
     MegolmEvent,
+    RoomEncryptedImage,
     RoomMessageFile,
     RoomMessageImage,
     RoomMessageText,
@@ -31,6 +32,7 @@ from nio import (
     KeyVerificationMac,
     ToDeviceError,
 )
+from nio.crypto.attachments import decrypt_attachment
 from livekit import api
 
 BOT_DEVICE_ID = "AIBOT"
@@ -200,6 +202,7 @@ class Bot:
         self.user_keys: dict[str, str] = self._load_user_keys()  # matrix_user_id -> api_key
         self.room_models: dict[str, str] = {}  # room_id -> model name
         self.auto_rename_rooms: set[str] = set()  # rooms with auto-rename enabled
+        self._recent_images: dict[str, tuple[str, str, float]] = {}  # room_id -> (b64, mime, timestamp)
         self.renamed_rooms: dict[str, float] = {}  # room_id -> timestamp of last rename
         self._loaded_rooms: set[str] = set()  # rooms where we've loaded state
         self._sync_token_received = False
@@ -259,6 +262,7 @@ class Bot:
         self.client.add_event_callback(self.on_unknown, UnknownEvent)
         self.client.add_event_callback(self.on_text_message, RoomMessageText)
         self.client.add_event_callback(self.on_image_message, RoomMessageImage)
+        self.client.add_event_callback(self.on_encrypted_image_message, RoomEncryptedImage)
         self.client.add_event_callback(self.on_file_message, RoomMessageFile)
         self.client.add_event_callback(self.on_room_unknown, RoomMessageUnknown)
         self.client.add_response_callback(self.on_sync, SyncResponse)
@@ -435,9 +439,18 @@ class Bot:
             await self._send_text(room.room_id, "LLM not configured (LITELLM_BASE_URL not set).")
             return
 
+        # Check if a recent image was sent in this room (within 60s)
+        image_data = None
+        cached = self._recent_images.get(room.room_id)
+        if cached:
+            b64, mime, ts = cached
+            if time.time() - ts < 60:
+                image_data = (b64, mime)
+                del self._recent_images[room.room_id]
+
         await self.client.room_typing(room.room_id, typing_state=True)
         try:
-            await self._respond_with_ai(room, body, sender=event.sender)
+            await self._respond_with_ai(room, body, sender=event.sender, image_data=image_data)
         finally:
             await self.client.room_typing(room.room_id, typing_state=False)
 
@@ -490,7 +503,70 @@ class Bot:
         b64_data = base64.b64encode(img_bytes).decode("utf-8")
 
         caption = (event.body or "").strip()
-        text = caption if caption and caption != "image" else "What's in this image?"
+        # Treat filenames (contain dots or are very long) as no caption
+        is_filename = not caption or caption == "image" or "." in caption or len(caption) > 100
+        text = "What's in this image?" if is_filename else caption
+
+        # Cache image for follow-up text messages
+        self._recent_images[room.room_id] = (b64_data, mime_type, time.time())
+
+        await self.client.room_typing(room.room_id, typing_state=True)
+        try:
+            await self._respond_with_ai(room, text, sender=event.sender, image_data=(b64_data, mime_type))
+        finally:
+            await self.client.room_typing(room.room_id, typing_state=False)
+
+    async def on_encrypted_image_message(self, room, event: RoomEncryptedImage):
+        """Handle encrypted image messages: decrypt, encode, and send to AI."""
+        if event.sender == BOT_USER:
+            return
+        if not self._sync_token_received:
+            return
+        server_ts = event.server_timestamp / 1000
+        if time.time() - server_ts > 30:
+            return
+
+        await self._load_room_settings(room.room_id)
+
+        is_dm = room.member_count == 2
+        if not is_dm:
+            body = (event.body or "").strip()
+            bot_display = self.client.user_id.split(":")[0].lstrip("@")
+            mentioned = (
+                BOT_USER in body
+                or f"@{bot_display}" in body.lower()
+                or bot_display.lower() in body.lower()
+            )
+            if not mentioned:
+                return
+
+        if not self.llm:
+            await self._send_text(room.room_id, "LLM not configured (LITELLM_BASE_URL not set).")
+            return
+
+        mxc_url = event.url
+        if not mxc_url:
+            return
+        try:
+            resp = await self.client.download(mxc=mxc_url)
+            if not hasattr(resp, "body"):
+                logger.warning("Encrypted image download failed for %s", mxc_url)
+                return
+            # Decrypt the attachment
+            img_bytes = decrypt_attachment(resp.body, event.key["k"], event.hashes["sha256"], event.iv)
+        except Exception:
+            logger.exception("Failed to download/decrypt encrypted image %s", mxc_url)
+            return
+
+        mime_type = getattr(event, "mimetype", None) or "image/png"
+        b64_data = base64.b64encode(img_bytes).decode("utf-8")
+
+        caption = (event.body or "").strip()
+        is_filename = not caption or caption == "image" or "." in caption or len(caption) > 100
+        text = "What's in this image?" if is_filename else caption
+
+        # Cache image for follow-up text messages
+        self._recent_images[room.room_id] = (b64_data, mime_type, time.time())
 
         await self.client.room_typing(room.room_id, typing_state=True)
         try: