feat(frontend): optimize photo requesting for slower internet

Makefile

@@ -1,4 +1,4 @@
-.PHONY: recreate down restart frontend deploy rebuild migrate convex-key script
+.PHONY: recreate down reset hard-reset restart frontend deploy rebuild migrate convex-key script
 
 recreate:
 	docker compose --profile services up -d
@@ -6,11 +6,20 @@ recreate:
 down:
 	docker compose --profile services down
 
+reset:
+	$(MAKE) down
+	$(MAKE) recreate
+
+hard-reset:
+	docker compose down
+	docker compose up -d
+
 restart:
 	docker compose --profile services restart
 
 frontend:
 	docker compose build frontend
+	$(MAKE) migrate
 	docker compose up -d frontend
 
 deploy:
@@ -22,10 +31,16 @@ rebuild:
 	docker compose --profile services up -d
 
 migrate:
-	docker compose --profile migrate run --rm migrator $(filter-out $@,$(MAKECMDGOALS))
+	docker compose run --rm migrate
 
 convex-key:
-	docker compose exec convex ./generate_admin_key.sh
+	@output=$$(docker compose exec convex ./generate_admin_key.sh 2>&1); \
+	echo "$$output"; \
+	if echo "$$output" | grep -q "Admin key:"; then \
+		key=$$(echo "$$output" | tail -1); \
+		sed -i '' 's#^CONVEX_SELF_HOSTED_ADMIN_KEY=.*#CONVEX_SELF_HOSTED_ADMIN_KEY='"$$key"'#' frontend/.env; \
+		echo "Updated frontend/.env with new admin key"; \
+	fi
 
 script:
 	@cd backend && docker compose --profile scripts run --rm script-runner scripts/$(subst .,/,$(word 2,$(MAKECMDGOALS))).py $(wordlist 3,$(words $(MAKECMDGOALS)),$(MAKECMDGOALS))

backend/src/bot/handlers/message/handler.py

@@ -1,11 +1,21 @@
 import asyncio
+import base64
 import contextlib
 import io
 import time
+from collections.abc import Awaitable, Callable
+from typing import Any
 
-from aiogram import Bot, F, Router, html, types
+from aiogram import BaseMiddleware, Bot, F, Router, html, types
 from aiogram.enums import ChatAction
-from aiogram.types import KeyboardButton, ReplyKeyboardMarkup, ReplyKeyboardRemove
+from aiogram.types import (
+    BufferedInputFile,
+    InputMediaPhoto,
+    KeyboardButton,
+    ReplyKeyboardMarkup,
+    ReplyKeyboardRemove,
+    TelegramObject,
+)
 from convex import ConvexInt64
 
 from bot.modules.ai import (
@@ -22,10 +32,57 @@ from utils.convex import ConvexClient
 router = Router()
 convex = ConvexClient(env.convex_url)
 
+ALBUM_COLLECT_DELAY = 0.5
+
+
+class AlbumMiddleware(BaseMiddleware):
+    def __init__(self) -> None:
+        self.albums: dict[str, list[types.Message]] = {}
+        self.scheduled: set[str] = set()
+
+    async def __call__(
+        self,
+        handler: Callable[[TelegramObject, dict[str, Any]], Awaitable[Any]],
+        event: TelegramObject,
+        data: dict[str, Any],
+    ) -> Any:  # noqa: ANN401
+        if not isinstance(event, types.Message) or not event.media_group_id:
+            return await handler(event, data)
+
+        album_id = event.media_group_id
+        if album_id not in self.albums:
+            self.albums[album_id] = []
+        self.albums[album_id].append(event)
+
+        if album_id in self.scheduled:
+            return None
+
+        self.scheduled.add(album_id)
+        await asyncio.sleep(ALBUM_COLLECT_DELAY)
+
+        messages = self.albums.pop(album_id, [])
+        self.scheduled.discard(album_id)
+
+        if messages:
+            data["album"] = messages
+            return await handler(messages[0], data)
+        return None
+
+
+router.message.middleware(AlbumMiddleware())
+
 EDIT_THROTTLE_SECONDS = 1.0
 TELEGRAM_MAX_LENGTH = 4096
 
 
+async def fetch_chat_images(chat_id: str) -> list[ImageData]:
+    chat_images = await convex.query("messages:getChatImages", {"chatId": chat_id})
+    return [
+        ImageData(data=base64.b64decode(img["base64"]), media_type=img["mediaType"])
+        for img in (chat_images or [])
+    ]
+
+
 def make_follow_up_keyboard(options: list[str]) -> ReplyKeyboardMarkup:
     buttons = [[KeyboardButton(text=opt)] for opt in options]
     return ReplyKeyboardMarkup(
@@ -114,8 +171,13 @@ async def send_long_message(
         )
 
 
-async def process_message_from_web(  # noqa: C901, PLR0915
+async def process_message_from_web(  # noqa: C901, PLR0912, PLR0913, PLR0915
-    convex_user_id: str, text: str, bot: Bot, convex_chat_id: str
+    convex_user_id: str,
+    text: str,
+    bot: Bot,
+    convex_chat_id: str,
+    images_base64: list[str] | None = None,
+    images_media_types: list[str] | None = None,
 ) -> None:
     user = await convex.query("users:getById", {"userId": convex_user_id})
 
@@ -126,9 +188,32 @@ async def process_message_from_web(  # noqa: C901, PLR0915
     is_summarize = text == "/summarize"
 
     if tg_chat_id and not is_summarize:
-        await bot.send_message(
+        if images_base64 and images_media_types:
-            tg_chat_id, f"📱 {html.quote(text)}", reply_markup=ReplyKeyboardRemove()
+            if len(images_base64) == 1:
-        )
+                photo_bytes = base64.b64decode(images_base64[0])
+                await bot.send_photo(
+                    tg_chat_id,
+                    BufferedInputFile(photo_bytes, "photo.jpg"),
+                    caption=f"📱 {text}" if text else "📱",
+                    reply_markup=ReplyKeyboardRemove(),
+                )
+            else:
+                media = []
+                img_pairs = zip(images_base64, images_media_types, strict=True)
+                for i, (img_b64, _) in enumerate(img_pairs):
+                    photo_bytes = base64.b64decode(img_b64)
+                    caption = f"📱 {text}" if i == 0 and text else None
+                    media.append(
+                        InputMediaPhoto(
+                            media=BufferedInputFile(photo_bytes, f"photo_{i}.jpg"),
+                            caption=caption,
+                        )
+                    )
+                await bot.send_media_group(tg_chat_id, media)
+        else:
+            await bot.send_message(
+                tg_chat_id, f"📱 {html.quote(text)}", reply_markup=ReplyKeyboardRemove()
+            )
 
     api_key = user["geminiApiKey"]
     model_name = user.get("model", "gemini-3-pro-preview")
@@ -178,7 +263,11 @@ async def process_message_from_web(  # noqa: C901, PLR0915
             prompt_text = text
             hist = history[:-1]
 
-        final_answer = await stream_response(text_agent, prompt_text, hist, on_chunk)
+        chat_images = await fetch_chat_images(convex_chat_id)
+
+        final_answer = await stream_response(
+            text_agent, prompt_text, hist, on_chunk, images=chat_images
+        )
 
         if state:
             await state.flush()
@@ -189,7 +278,7 @@ async def process_message_from_web(  # noqa: C901, PLR0915
         follow_up_agent = create_follow_up_agent(
             api_key=api_key, model_name=follow_up_model, system_prompt=follow_up_prompt
         )
-        follow_ups = await get_follow_ups(follow_up_agent, full_history)
+        follow_ups = await get_follow_ups(follow_up_agent, full_history, chat_images)
 
         if state:
             await state.stop_typing()
@@ -204,6 +293,21 @@ async def process_message_from_web(  # noqa: C901, PLR0915
             },
         )
 
+        if is_summarize:
+            await convex.mutation(
+                "chats:clear", {"chatId": convex_chat_id, "preserveImages": True}
+            )
+            await convex.mutation(
+                "messages:create",
+                {
+                    "chatId": convex_chat_id,
+                    "role": "assistant",
+                    "content": final_answer,
+                    "source": "web",
+                    "followUpOptions": follow_ups,
+                },
+            )
+
         if tg_chat_id and processing_msg:
             with contextlib.suppress(Exception):
                 await processing_msg.delete()
@@ -229,7 +333,7 @@ async def process_message_from_web(  # noqa: C901, PLR0915
 
 
 async def process_message(
-    user_id: int, text: str, bot: Bot, chat_id: int, image: ImageData | None = None
+    user_id: int, text: str, bot: Bot, chat_id: int, *, skip_user_message: bool = False
 ) -> None:
     user = await convex.query(
         "users:getByTelegramId", {"telegramId": ConvexInt64(user_id)}
@@ -251,15 +355,16 @@ async def process_message(
     api_key = user["geminiApiKey"]
     model_name = user.get("model", "gemini-3-pro-preview")
 
-    await convex.mutation(
+    if not skip_user_message:
-        "messages:create",
+        await convex.mutation(
-        {
+            "messages:create",
-            "chatId": active_chat_id,
+            {
-            "role": "user",
+                "chatId": active_chat_id,
-            "content": text,
+                "role": "user",
-            "source": "telegram",
+                "content": text,
-        },
+                "source": "telegram",
-    )
+            },
+        )
 
     assistant_message_id = await convex.mutation(
         "messages:create",
@@ -293,8 +398,10 @@ async def process_message(
                 {"messageId": assistant_message_id, "content": content},
             )
 
+        chat_images = await fetch_chat_images(active_chat_id)
+
         final_answer = await stream_response(
-            text_agent, text, history[:-2], on_chunk, image=image
+            text_agent, text, history[:-2], on_chunk, images=chat_images
         )
 
         await state.flush()
@@ -305,7 +412,7 @@ async def process_message(
         follow_up_agent = create_follow_up_agent(
             api_key=api_key, model_name=follow_up_model, system_prompt=follow_up_prompt
         )
-        follow_ups = await get_follow_ups(follow_up_agent, full_history, image=image)
+        follow_ups = await get_follow_ups(follow_up_agent, full_history, chat_images)
 
         await state.stop_typing()
 
@@ -367,6 +474,74 @@ async def on_text_message(message: types.Message, bot: Bot) -> None:
     await process_message(message.from_user.id, message.text, bot, message.chat.id)
 
 
+@router.message(F.media_group_id, F.photo)
+async def on_album_message(
+    message: types.Message, bot: Bot, album: list[types.Message]
+) -> None:
+    if not message.from_user:
+        return
+
+    await convex.mutation(
+        "users:getOrCreate",
+        {
+            "telegramId": ConvexInt64(message.from_user.id),
+            "telegramChatId": ConvexInt64(message.chat.id),
+        },
+    )
+
+    user = await convex.query(
+        "users:getByTelegramId", {"telegramId": ConvexInt64(message.from_user.id)}
+    )
+
+    if not user or not user.get("activeChatId"):
+        await message.answer("Use /new first to create a chat.")
+        return
+
+    caption = message.caption or "Process the images according to your task"
+
+    images_base64: list[str] = []
+    images_media_types: list[str] = []
+
+    for msg in album:
+        if not msg.photo:
+            continue
+
+        photo = msg.photo[-1]
+        file = await bot.get_file(photo.file_id)
+        if not file.file_path:
+            continue
+
+        buffer = io.BytesIO()
+        await bot.download_file(file.file_path, buffer)
+        image_bytes = buffer.getvalue()
+        images_base64.append(base64.b64encode(image_bytes).decode())
+
+        ext = file.file_path.rsplit(".", 1)[-1].lower()
+        media_type = f"image/{ext}" if ext in ("png", "gif", "webp") else "image/jpeg"
+        images_media_types.append(media_type)
+
+    if not images_base64:
+        await message.answer("Failed to get photos.")
+        return
+
+    active_chat_id = user["activeChatId"]
+    await convex.mutation(
+        "messages:create",
+        {
+            "chatId": active_chat_id,
+            "role": "user",
+            "content": caption,
+            "source": "telegram",
+            "imagesBase64": images_base64,
+            "imagesMediaTypes": images_media_types,
+        },
+    )
+
+    await process_message(
+        message.from_user.id, caption, bot, message.chat.id, skip_user_message=True
+    )
+
+
 @router.message(F.photo)
 async def on_photo_message(message: types.Message, bot: Bot) -> None:
     if not message.from_user or not message.photo:
@@ -380,6 +555,14 @@ async def on_photo_message(message: types.Message, bot: Bot) -> None:
         },
     )
 
+    user = await convex.query(
+        "users:getByTelegramId", {"telegramId": ConvexInt64(message.from_user.id)}
+    )
+
+    if not user or not user.get("activeChatId"):
+        await message.answer("Use /new first to create a chat.")
+        return
+
     caption = message.caption or "Process the image according to your task"
     photo = message.photo[-1]
 
@@ -391,11 +574,24 @@ async def on_photo_message(message: types.Message, bot: Bot) -> None:
     buffer = io.BytesIO()
     await bot.download_file(file.file_path, buffer)
     image_bytes = buffer.getvalue()
+    image_base64 = base64.b64encode(image_bytes).decode()
 
     ext = file.file_path.rsplit(".", 1)[-1].lower()
     media_type = f"image/{ext}" if ext in ("png", "gif", "webp") else "image/jpeg"
-    image = ImageData(data=image_bytes, media_type=media_type)
+
+    active_chat_id = user["activeChatId"]
+    await convex.mutation(
+        "messages:create",
+        {
+            "chatId": active_chat_id,
+            "role": "user",
+            "content": caption,
+            "source": "telegram",
+            "imageBase64": image_base64,
+            "imageMediaType": media_type,
+        },
+    )
 
     await process_message(
-        message.from_user.id, caption, bot, message.chat.id, image=image
+        message.from_user.id, caption, bot, message.chat.id, skip_user_message=True
     )

backend/src/bot/modules/ai/agent.py

@@ -41,7 +41,7 @@ def create_text_agent(
     model = GoogleModel(model_name, provider=provider)
     base_prompt = system_prompt or DEFAULT_SYSTEM_PROMPT
     full_prompt = f"{base_prompt} {LATEX_INSTRUCTION}"
-    return Agent(model, system_prompt=full_prompt)
+    return Agent(model, instructions=full_prompt)
 
 
 def create_follow_up_agent(
@@ -52,7 +52,7 @@ def create_follow_up_agent(
     provider = GoogleProvider(api_key=api_key)
     model = GoogleModel(model_name, provider=provider)
     prompt = system_prompt or DEFAULT_FOLLOW_UP
-    return Agent(model, output_type=FollowUpOptions, system_prompt=prompt)
+    return Agent(model, output_type=FollowUpOptions, instructions=prompt)
 
 
 def build_message_history(history: list[dict[str, str]]) -> list[ModelMessage]:
@@ -99,17 +99,17 @@ async def stream_response(  # noqa: PLR0913
 async def get_follow_ups(
     follow_up_agent: Agent[None, FollowUpOptions],
     history: list[dict[str, str]],
-    image: ImageData | None = None,
+    images: list[ImageData] | None = None,
 ) -> list[str]:
     message_history = build_message_history(history) if history else None
 
-    if image:
+    if images:
-        prompt: list[str | BinaryContent] = [
+        prompt: list[str | BinaryContent] = ["Process this:"]
-            "Suggest follow-up options based on this conversation and image.",
+        prompt.extend(
-            BinaryContent(data=image.data, media_type=image.media_type),
+            BinaryContent(data=img.data, media_type=img.media_type) for img in images
-        ]
+        )
     else:
-        prompt = "Suggest follow-up questions based on this conversation."  # type: ignore[assignment]
+        prompt = "Process this conversation."  # type: ignore[assignment]
 
     result = await follow_up_agent.run(prompt, message_history=message_history)
     return result.output["options"]

backend/src/bot/modules/ai/prompts.py

@@ -1,6 +1,6 @@
 EXAM_SYSTEM = """You help solve problem sets and exams.
 
-When you receive an IMAGE with problems:
+When you receive just an IMAGE to process with problems:
 - Give HINTS in Russian for each problem
 - Focus on key insights and potential difficulties,
 give all formulas that will be helpful
@@ -9,18 +9,22 @@ give all formulas that will be helpful
 
 When asked for DETAILS on a specific problem (or a problem number):
 - Provide full structured solution in English
-- Academic style, as it would be written in a notebook
+- Academic style, as it would be written in a notebook on real exam
-- Step by step, clean, no fluff"""
+- Step by step, clean, no fluff, no overcompications, reuse thoughts inside
+ one task, as you would write it on an exam, be consistent
+- This is also true if you get a summary, and then problem number is asked"""
 
-EXAM_FOLLOW_UP = """You see a problem set image. List available problem numbers.
+EXAM_FOLLOW_UP = """Look at the problem set image and list ALL problem numbers as
-Output only the numbers that exist in the image, like: 1, 2, 3, 4, 5
+ options. Split by subparts ONLY if they are totally different tasks, not the steps of
-If problems have letters (a, b, c), list them as: 1a, 1b, 2a, etc.
+ one.
-Keep it minimal - just the identifiers.
+If there are multiple problem sets/sheets, break it down logically and specify set,
-Then, if applicable, output some possible followups of conversation"""
+for example Group A: 1, Group A: 2a, Group B: 2b, etc.
+Or, Theory: 1, Theory: 2a, Practice: 1, etc.
+Only output identifiers that exist in the image."""
 
 DEFAULT_FOLLOW_UP = (
     "Based on the conversation, suggest 3 short follow-up questions "
-    "the user might want to ask. Be concise, each under 50 chars."
+    "the user might want to ask. Each option should be under 50 characters."
 )
 
 SUMMARIZE_PROMPT = """You are summarize agent. You may receive:

backend/src/bot/sync.py

@@ -51,6 +51,8 @@ async def handle_pending_generation(bot: Bot, item: dict, item_id: str) -> None:
             text=item["userMessage"],
             bot=bot,
             convex_chat_id=item["chatId"],
+            images_base64=item.get("imagesBase64"),
+            images_media_types=item.get("imagesMediaTypes"),
         )
     except Exception as e:  # noqa: BLE001
         logger.error(f"Error processing {item_id}: {e}")

caddy/Caddyfile.example

@@ -26,6 +26,9 @@
     }
 
     handle {
+        request_body {
+            max_size 50MB
+        }
         reverse_proxy stealth-ai-relay-frontend:3000
     }
 }

docker-compose.yml

@@ -69,6 +69,8 @@ services:
     build:
       context: ./frontend
       dockerfile: Dockerfile
+      args:
+        - PUBLIC_CONVEX_URL=${PUBLIC_CONVEX_URL}
     image: stealth-ai-relay/frontend
     profiles:
       - frontend
@@ -85,6 +87,8 @@ services:
     build:
       context: ./frontend
       dockerfile: Dockerfile
+      args:
+        - PUBLIC_CONVEX_URL=${PUBLIC_CONVEX_URL}
     image: stealth-ai-relay/frontend
     volumes:
       - ./frontend:/app
@@ -121,6 +125,20 @@ services:
       browserless:
     entrypoint: [ "python" ]
 
+  migrate:
+    image: stealth-ai-relay/frontend
+    volumes:
+      - ./frontend:/app
+      - /app/node_modules
+    env_file:
+      - "frontend/.env"
+      - ".env"
+    profiles:
+      - migrate
+    networks:
+      database:
+    command: x convex deploy
+
   convex-dashboard:
     image: ghcr.io/get-convex/convex-dashboard:latest
     stop_grace_period: 10s

frontend/.gitignore

@@ -21,6 +21,3 @@ Thumbs.db
 # Vite
 vite.config.js.timestamp-*
 vite.config.ts.timestamp-*
-
-# Convex
-src/lib/convex/_generated

frontend/.prettierignore

@@ -5,5 +5,8 @@ yarn.lock
 bun.lock
 bun.lockb
 
+# Convex generated files
+src/lib/convex/_generated/
+
 # Miscellaneous
 /static/

frontend/Dockerfile

@@ -3,6 +3,9 @@ FROM oven/bun:alpine
 ENV TERM=xterm-256color
 ENV COLORTERM=truecolor
 
+ARG PUBLIC_CONVEX_URL
+ENV PUBLIC_CONVEX_URL=$PUBLIC_CONVEX_URL
+
 WORKDIR /app
 
 COPY package.json bun.lock* ./

frontend/eslint.config.js

@@ -22,9 +22,11 @@ export default defineConfig(
 		languageOptions: { globals: { ...globals.browser, ...globals.node } },
 
 		rules: {
-			// typescript-eslint strongly recommend that you do not use the no-undef lint rule on TypeScript projects.
+			'no-undef': 'off',
-			// see: https://typescript-eslint.io/troubleshooting/faqs/eslint/#i-get-errors-from-the-no-undef-rule-about-global-variables-not-being-defined-even-though-there-are-no-typescript-errors
+			'@typescript-eslint/no-unused-vars': [
-			'no-undef': 'off'
+				'error',
+				{ argsIgnorePattern: '^_', varsIgnorePattern: '^_' }
+			]
 		}
 	},
 	{

frontend/src/app.d.ts

@@ -1,12 +1,10 @@
-// See https://svelte.dev/docs/kit/types#app.d.ts
+import type { ConvexHttpClient } from 'convex/browser';
-// for information about these interfaces
+
 declare global {
 	namespace App {
-		// interface Error {}
+		interface Locals {
-		// interface Locals {}
+			convex: ConvexHttpClient;
-		// interface PageData {}
+		}
-		// interface PageState {}
-		// interface Platform {}
 	}
 }
 

frontend/src/convex/_generated/api.d.ts

@@ -1,35 +0,0 @@
-/* eslint-disable */
-/**
- * Generated `api` utility.
- *
- * THIS CODE IS AUTOMATICALLY GENERATED.
- *
- * To regenerate, run `npx convex dev`.
- * @module
- */
-
-import type { ApiFromModules, FilterApi, FunctionReference } from 'convex/server';
-
-declare const fullApi: ApiFromModules<{}>;
-
-/**
- * A utility for referencing Convex functions in your app's public API.
- *
- * Usage:
- * ```js
- * const myFunctionReference = api.myModule.myFunction;
- * ```
- */
-export declare const api: FilterApi<typeof fullApi, FunctionReference<any, 'public'>>;
-
-/**
- * A utility for referencing Convex functions in your app's internal API.
- *
- * Usage:
- * ```js
- * const myFunctionReference = internal.myModule.myFunction;
- * ```
- */
-export declare const internal: FilterApi<typeof fullApi, FunctionReference<any, 'internal'>>;
-
-export declare const components: {};

frontend/src/hooks.server.ts

@@ -0,0 +1,8 @@
+import { ConvexHttpClient } from 'convex/browser';
+import { PUBLIC_CONVEX_URL } from '$env/static/public';
+import type { Handle } from '@sveltejs/kit';
+
+export const handle: Handle = async ({ event, resolve }) => {
+	event.locals.convex = new ConvexHttpClient(PUBLIC_CONVEX_URL);
+	return resolve(event);
+};

frontend/src/lib/components/CameraCapture.svelte

@@ -0,0 +1,184 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+
+	interface Props {
+		showPreview?: boolean;
+		oncapture: (base64: string, mediaType: string) => void;
+		onclose: () => void;
+	}
+
+	let { showPreview = true, oncapture, onclose }: Props = $props();
+
+	let videoElement: HTMLVideoElement | null = $state(null);
+	let stream: MediaStream | null = $state(null);
+	let capturedImage: { base64: string; mediaType: string } | null = $state(null);
+	let error: string | null = $state(null);
+	let closed = false;
+
+	async function findUltraWideCamera(): Promise<string | null> {
+		try {
+			const devices = await navigator.mediaDevices.enumerateDevices();
+			const videoDevices = devices.filter((d) => d.kind === 'videoinput');
+			const ultraWide = videoDevices.find(
+				(d) => d.label.toLowerCase().includes('ultra') && d.label.toLowerCase().includes('back')
+			);
+			return ultraWide?.deviceId ?? null;
+		} catch {
+			return null;
+		}
+	}
+
+	async function startCamera() {
+		if (closed) return;
+
+		if (!navigator.mediaDevices?.getUserMedia) {
+			error = 'Camera not supported (requires HTTPS)';
+			return;
+		}
+
+		try {
+			stream = await navigator.mediaDevices.getUserMedia({
+				video: { facingMode: { ideal: 'environment' } },
+				audio: false
+			});
+
+			const ultraWideId = await findUltraWideCamera();
+
+			if (ultraWideId) {
+				stream.getTracks().forEach((t) => t.stop());
+				stream = await navigator.mediaDevices.getUserMedia({
+					video: {
+						deviceId: { exact: ultraWideId },
+						width: { ideal: 4032 },
+						height: { ideal: 3024 }
+					},
+					audio: false
+				});
+			} else {
+				stream.getTracks().forEach((t) => t.stop());
+				stream = await navigator.mediaDevices.getUserMedia({
+					video: {
+						facingMode: { ideal: 'environment' },
+						width: { ideal: 4032 },
+						height: { ideal: 3024 }
+					},
+					audio: false
+				});
+			}
+
+			if (videoElement && !closed) {
+				videoElement.srcObject = stream;
+			}
+		} catch (e) {
+			error = e instanceof Error ? e.message : 'Camera access denied';
+		}
+	}
+
+	function stopCamera() {
+		if (stream) {
+			stream.getTracks().forEach((track) => track.stop());
+			stream = null;
+		}
+	}
+
+	function capture() {
+		if (!videoElement) return;
+
+		const canvas = document.createElement('canvas');
+		canvas.width = videoElement.videoWidth;
+		canvas.height = videoElement.videoHeight;
+
+		const ctx = canvas.getContext('2d');
+		if (!ctx) return;
+
+		ctx.drawImage(videoElement, 0, 0);
+
+		const maxSize = 1920;
+		const scale = Math.min(maxSize / canvas.width, maxSize / canvas.height, 1);
+		const outCanvas = document.createElement('canvas');
+		outCanvas.width = Math.round(canvas.width * scale);
+		outCanvas.height = Math.round(canvas.height * scale);
+		const outCtx = outCanvas.getContext('2d');
+		if (!outCtx) return;
+		outCtx.drawImage(canvas, 0, 0, outCanvas.width, outCanvas.height);
+
+		const base64 = outCanvas.toDataURL('image/jpeg', 0.65).split(',')[1];
+		const mediaType = 'image/jpeg';
+
+		stopCamera();
+
+		if (showPreview) {
+			capturedImage = { base64, mediaType };
+		} else {
+			oncapture(base64, mediaType);
+		}
+	}
+
+	function acceptCapture() {
+		if (capturedImage) {
+			oncapture(capturedImage.base64, capturedImage.mediaType);
+		}
+	}
+
+	function retake() {
+		capturedImage = null;
+		startCamera();
+	}
+
+	function close() {
+		closed = true;
+		stopCamera();
+		onclose();
+	}
+
+	onMount(() => {
+		startCamera();
+		return () => {
+			closed = true;
+			stopCamera();
+		};
+	});
+</script>
+
+<div class="fixed inset-0 z-50 flex flex-col bg-black" data-camera-ui>
+	{#if error}
+		<div class="flex flex-1 flex-col items-center justify-center p-4">
+			<p class="mb-4 text-center text-sm text-red-400">{error}</p>
+			<button onclick={close} class="rounded bg-neutral-700 px-4 py-2 text-sm text-white">
+				Close
+			</button>
+		</div>
+	{:else if capturedImage}
+		<div class="relative min-h-0 flex-1">
+			<button class="absolute inset-0" onclick={acceptCapture}>
+				<img
+					src="data:{capturedImage.mediaType};base64,{capturedImage.base64}"
+					alt="Captured"
+					class="h-full w-full object-contain"
+				/>
+			</button>
+		</div>
+		<div class="flex gap-4 p-4">
+			<button onclick={retake} class="flex-1 rounded bg-neutral-700 py-3 text-sm text-white">
+				Retake
+			</button>
+			<button onclick={acceptCapture} class="flex-1 rounded bg-blue-600 py-3 text-sm text-white">
+				Use
+			</button>
+		</div>
+	{:else}
+		<video bind:this={videoElement} autoplay playsinline muted class="h-full w-full object-cover"
+		></video>
+		<button
+			onclick={close}
+			class="absolute top-4 right-4 flex h-8 w-8 items-center justify-center rounded-full bg-black/50 text-white"
+		>
+			×
+		</button>
+		<button
+			onclick={capture}
+			aria-label="Capture photo"
+			class="absolute bottom-8 left-1/2 h-16 w-16 -translate-x-1/2 rounded-full border-4 border-white bg-white/20"
+		></button>
+	{/if}
+</div>

frontend/src/lib/components/CaptureCountdown.svelte

@@ -0,0 +1,144 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+
+	interface Props {
+		oncomplete: (base64: string, mediaType: string, thumbnailBase64: string) => void;
+		oncancel: () => void;
+	}
+
+	let { oncomplete, oncancel }: Props = $props();
+
+	let count = $state(3);
+	let videoElement: HTMLVideoElement | null = $state(null);
+	let stream: MediaStream | null = $state(null);
+	let error: string | null = $state(null);
+	let cancelled = false;
+	let countdownInterval: ReturnType<typeof setInterval> | null = null;
+
+	async function startCamera() {
+		if (cancelled) return;
+
+		if (!navigator.mediaDevices?.getUserMedia) {
+			error = 'Camera not supported (requires HTTPS)';
+			return;
+		}
+
+		try {
+			const constraints: MediaStreamConstraints = {
+				video: {
+					facingMode: { ideal: 'environment' },
+					width: { ideal: 1920 },
+					height: { ideal: 1080 }
+				},
+				audio: false
+			};
+
+			stream = await navigator.mediaDevices.getUserMedia(constraints);
+
+			if (videoElement && !cancelled) {
+				videoElement.srcObject = stream;
+				startCountdown();
+			}
+		} catch (e) {
+			error = e instanceof Error ? e.message : 'Camera access denied';
+		}
+	}
+
+	function stopCamera() {
+		if (countdownInterval) {
+			clearInterval(countdownInterval);
+			countdownInterval = null;
+		}
+		if (stream) {
+			stream.getTracks().forEach((track) => track.stop());
+			stream = null;
+		}
+	}
+
+	function startCountdown() {
+		countdownInterval = setInterval(() => {
+			if (cancelled) {
+				stopCamera();
+				return;
+			}
+			count--;
+			if (count === 0) {
+				if (countdownInterval) clearInterval(countdownInterval);
+				capture();
+			}
+		}, 1000);
+	}
+
+	function capture() {
+		if (cancelled || !videoElement) {
+			stopCamera();
+			return;
+		}
+
+		const canvas = document.createElement('canvas');
+		canvas.width = videoElement.videoWidth;
+		canvas.height = videoElement.videoHeight;
+
+		const ctx = canvas.getContext('2d');
+		if (!ctx) {
+			stopCamera();
+			oncancel();
+			return;
+		}
+
+		ctx.drawImage(videoElement, 0, 0);
+
+		const base64 = canvas.toDataURL('image/jpeg', 0.85).split(',')[1];
+		const mediaType = 'image/jpeg';
+
+		const thumbMaxSize = 800;
+		const scale = Math.min(thumbMaxSize / canvas.width, thumbMaxSize / canvas.height, 1);
+		const thumbCanvas = document.createElement('canvas');
+		thumbCanvas.width = Math.round(canvas.width * scale);
+		thumbCanvas.height = Math.round(canvas.height * scale);
+		const thumbCtx = thumbCanvas.getContext('2d');
+		if (thumbCtx) {
+			thumbCtx.drawImage(canvas, 0, 0, thumbCanvas.width, thumbCanvas.height);
+		}
+		const thumbnailBase64 = thumbCanvas.toDataURL('image/jpeg', 0.7).split(',')[1];
+
+		stopCamera();
+		oncomplete(base64, mediaType, thumbnailBase64);
+	}
+
+	function handleCancel() {
+		cancelled = true;
+		stopCamera();
+		oncancel();
+	}
+
+	onMount(() => {
+		startCamera();
+		return () => {
+			cancelled = true;
+			stopCamera();
+		};
+	});
+</script>
+
+<div class="fixed inset-0 z-50 flex flex-col bg-black" data-camera-ui>
+	{#if error}
+		<div class="flex flex-1 flex-col items-center justify-center p-4">
+			<p class="mb-4 text-center text-sm text-red-400">{error}</p>
+			<button onclick={handleCancel} class="rounded bg-neutral-700 px-4 py-2 text-sm text-white">
+				Close
+			</button>
+		</div>
+	{:else}
+		<div class="relative flex-1">
+			<video bind:this={videoElement} autoplay playsinline muted class="h-full w-full object-cover"
+			></video>
+			<div class="absolute inset-0 flex items-center justify-center">
+				<span class="text-8xl font-bold text-white drop-shadow-lg">{count}</span>
+			</div>
+		</div>
+		<div class="p-4 text-center">
+			<button onclick={handleCancel} class="text-sm text-neutral-400">Cancel</button>
+		</div>
+	{/if}
+</div>

frontend/src/lib/components/ChatInput.svelte

@@ -2,34 +2,35 @@
 	interface Props {
 		onsubmit: (message: string) => void;
 		disabled?: boolean;
+		allowEmpty?: boolean;
 	}
 
-	let { onsubmit, disabled = false }: Props = $props();
+	let { onsubmit, disabled = false, allowEmpty = false }: Props = $props();
 	let value = $state('');
 
 	function handleSubmit(e: Event) {
 		e.preventDefault();
 		const trimmed = value.trim();
-		if (trimmed && !disabled) {
+		if ((trimmed || allowEmpty) && !disabled) {
 			onsubmit(trimmed);
 			value = '';
 		}
 	}
 </script>
 
-<form onsubmit={handleSubmit} class="flex gap-2">
+<form onsubmit={handleSubmit} class="flex gap-1">
 	<input
 		type="text"
 		bind:value
 		{disabled}
-		placeholder="Message..."
+		placeholder="..."
-		class="flex-1 rounded-lg bg-neutral-800 px-3 py-2 text-[11px] text-white placeholder-neutral-500 outline-none focus:ring-1 focus:ring-neutral-600"
+		class="min-w-0 flex-1 rounded bg-neutral-800 px-2 py-1 text-[10px] text-white placeholder-neutral-500 outline-none"
 	/>
 	<button
 		type="submit"
 		{disabled}
-		class="rounded-lg bg-blue-600 px-3 py-2 text-[11px] text-white transition-colors hover:bg-blue-500 disabled:opacity-50"
+		class="shrink-0 rounded bg-blue-600 px-2 py-1 text-[10px] text-white disabled:opacity-50"
 	>
-		Send
+		&gt;
 	</button>
 </form>

frontend/src/lib/components/DraftBadge.svelte

@@ -0,0 +1,29 @@
+<script lang="ts">
+	import type { Id } from '$lib/convex/_generated/dataModel';
+
+	interface Photo {
+		_id: Id<'photoDrafts'>;
+		mediaType: string;
+	}
+
+	interface Props {
+		photos: Photo[];
+		onremove: (index: number) => void;
+	}
+
+	let { photos, onremove }: Props = $props();
+</script>
+
+{#if photos.length > 0}
+	<div class="flex flex-wrap gap-1">
+		{#each photos as _photo, i (i)}
+			<button
+				onclick={() => onremove(i)}
+				class="flex items-center gap-1 rounded bg-blue-600/30 px-1.5 py-0.5 text-[8px] text-blue-300"
+			>
+				<span>photo {i + 1}</span>
+				<span class="text-blue-400">&times;</span>
+			</button>
+		{/each}
+	</div>
+{/if}

frontend/src/lib/components/PhotoButton.svelte

@@ -0,0 +1,65 @@
+<script lang="ts">
+	interface Props {
+		hasCamera: boolean;
+		hasOnlineDevices: boolean;
+		ontakephoto: () => void;
+		onrequestphoto: () => void;
+	}
+
+	let { hasCamera, hasOnlineDevices, ontakephoto, onrequestphoto }: Props = $props();
+	let menuOpen = $state(false);
+
+	function handleClick() {
+		if (hasCamera && hasOnlineDevices) {
+			menuOpen = !menuOpen;
+		} else if (hasOnlineDevices) {
+			onrequestphoto();
+		} else {
+			ontakephoto();
+		}
+	}
+
+	function handleTakePhoto() {
+		menuOpen = false;
+		ontakephoto();
+	}
+
+	function handleRequestPhoto() {
+		menuOpen = false;
+		onrequestphoto();
+	}
+
+	function handleBackdropClick() {
+		menuOpen = false;
+	}
+</script>
+
+<div class="relative">
+	<button
+		onclick={handleClick}
+		class="shrink-0 rounded bg-neutral-800 px-1.5 py-0.5 text-[8px] text-neutral-400"
+	>
+		+
+	</button>
+
+	{#if menuOpen}
+		<button class="fixed inset-0 z-40" onclick={handleBackdropClick} aria-label="Close menu"
+		></button>
+		<div
+			class="absolute bottom-full left-0 z-50 mb-1 overflow-hidden rounded bg-neutral-800 shadow-lg"
+		>
+			<button
+				onclick={handleTakePhoto}
+				class="block w-full px-3 py-2 text-left text-[10px] whitespace-nowrap text-white hover:bg-neutral-700"
+			>
+				Take photo
+			</button>
+			<button
+				onclick={handleRequestPhoto}
+				class="block w-full px-3 py-2 text-left text-[10px] whitespace-nowrap text-white hover:bg-neutral-700"
+			>
+				Request photo
+			</button>
+		</div>
+	{/if}
+</div>

frontend/src/lib/components/PhotoPreview.svelte

@@ -0,0 +1,27 @@
+<script lang="ts">
+	interface Props {
+		base64: string;
+		mediaType: string;
+		onaccept: () => void;
+		onreject: () => void;
+	}
+
+	let { base64, mediaType, onaccept, onreject }: Props = $props();
+</script>
+
+<div class="fixed inset-0 z-50 overflow-auto bg-black" data-camera-ui>
+	<button class="block min-h-full min-w-full" onclick={onaccept}>
+		<img
+			src="data:{mediaType};base64,{base64}"
+			alt="Preview"
+			class="min-h-dvh min-w-full object-cover"
+		/>
+	</button>
+</div>
+<button
+	onclick={onreject}
+	class="fixed top-4 right-4 z-[9999] flex h-10 w-10 items-center justify-center rounded-full bg-red-600 text-xl text-white shadow-lg"
+	data-camera-ui
+>
+	×
+</button>

frontend/src/lib/components/PhotoRequestPopup.svelte

@@ -0,0 +1,22 @@
+<script lang="ts">
+	interface Props {
+		onaccept: () => void;
+		ondecline: () => void;
+	}
+
+	let { onaccept, ondecline }: Props = $props();
+</script>
+
+<div class="fixed inset-0 z-50 flex items-center justify-center bg-black/80 p-4" data-camera-ui>
+	<div class="w-full max-w-xs rounded-lg bg-neutral-900 p-4">
+		<p class="mb-4 text-center text-sm text-white">Photo requested</p>
+		<div class="flex gap-3">
+			<button onclick={ondecline} class="flex-1 rounded bg-neutral-700 py-2 text-sm text-white">
+				Decline
+			</button>
+			<button onclick={onaccept} class="flex-1 rounded bg-blue-600 py-2 text-sm text-white">
+				Capture
+			</button>
+		</div>
+	</div>
+</div>

frontend/src/lib/components/SilentCapture.svelte

@@ -0,0 +1,135 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+
+	interface Props {
+		oncapture: (base64: string, mediaType: string, thumbnailBase64: string) => void;
+		onunpair?: () => void;
+	}
+
+	let { oncapture, onunpair }: Props = $props();
+
+	let videoElement: HTMLVideoElement | null = $state(null);
+	let stream: MediaStream | null = $state(null);
+	let ready = $state(false);
+
+	async function findUltraWideCamera(): Promise<string | null> {
+		try {
+			const devices = await navigator.mediaDevices.enumerateDevices();
+			const videoDevices = devices.filter((d) => d.kind === 'videoinput');
+			const ultraWide = videoDevices.find(
+				(d) => d.label.toLowerCase().includes('ultra') && d.label.toLowerCase().includes('back')
+			);
+			return ultraWide?.deviceId ?? null;
+		} catch {
+			return null;
+		}
+	}
+
+	async function startCamera() {
+		if (!navigator.mediaDevices?.getUserMedia) return;
+
+		try {
+			stream = await navigator.mediaDevices.getUserMedia({
+				video: { facingMode: { ideal: 'environment' } },
+				audio: false
+			});
+
+			const ultraWideId = await findUltraWideCamera();
+
+			if (ultraWideId) {
+				stream.getTracks().forEach((t) => t.stop());
+				stream = await navigator.mediaDevices.getUserMedia({
+					video: {
+						deviceId: { exact: ultraWideId },
+						width: { ideal: 4032 },
+						height: { ideal: 3024 }
+					},
+					audio: false
+				});
+			} else {
+				stream.getTracks().forEach((t) => t.stop());
+				stream = await navigator.mediaDevices.getUserMedia({
+					video: {
+						facingMode: { ideal: 'environment' },
+						width: { ideal: 4032 },
+						height: { ideal: 3024 }
+					},
+					audio: false
+				});
+			}
+
+			if (videoElement) {
+				videoElement.srcObject = stream;
+				await new Promise<void>((resolve) => {
+					if (videoElement) {
+						videoElement.onloadedmetadata = () => resolve();
+					}
+				});
+				ready = true;
+			}
+		} catch {
+			ready = false;
+		}
+	}
+
+	export function capture() {
+		if (!ready || !videoElement) return false;
+
+		const canvas = document.createElement('canvas');
+		canvas.width = videoElement.videoWidth;
+		canvas.height = videoElement.videoHeight;
+
+		const ctx = canvas.getContext('2d');
+		if (!ctx) return false;
+
+		ctx.drawImage(videoElement, 0, 0);
+
+		const maxSize = 1920;
+		const scale = Math.min(maxSize / canvas.width, maxSize / canvas.height, 1);
+		const outCanvas = document.createElement('canvas');
+		outCanvas.width = Math.round(canvas.width * scale);
+		outCanvas.height = Math.round(canvas.height * scale);
+		const outCtx = outCanvas.getContext('2d');
+		if (!outCtx) return false;
+		outCtx.drawImage(canvas, 0, 0, outCanvas.width, outCanvas.height);
+
+		const base64 = outCanvas.toDataURL('image/jpeg', 0.65).split(',')[1];
+		const mediaType = 'image/jpeg';
+
+		const thumbMaxSize = 800;
+		const thumbScale = Math.min(thumbMaxSize / outCanvas.width, thumbMaxSize / outCanvas.height, 1);
+		const thumbCanvas = document.createElement('canvas');
+		thumbCanvas.width = Math.round(outCanvas.width * thumbScale);
+		thumbCanvas.height = Math.round(outCanvas.height * thumbScale);
+		const thumbCtx = thumbCanvas.getContext('2d');
+		if (thumbCtx) {
+			thumbCtx.drawImage(outCanvas, 0, 0, thumbCanvas.width, thumbCanvas.height);
+		}
+		const thumbnailBase64 = thumbCanvas.toDataURL('image/jpeg', 0.6).split(',')[1];
+
+		oncapture(base64, mediaType, thumbnailBase64);
+		return true;
+	}
+
+	onMount(() => {
+		startCamera();
+		return () => {
+			if (stream) {
+				stream.getTracks().forEach((track) => track.stop());
+			}
+		};
+	});
+</script>
+
+<div class="fixed inset-0 z-40 bg-black">
+	<video bind:this={videoElement} autoplay playsinline muted class="h-full w-full object-cover"
+	></video>
+	{#if onunpair}
+		<button
+			onclick={onunpair}
+			class="absolute top-4 left-4 z-10 rounded-full bg-red-600/80 px-3 py-1.5 text-xs text-white"
+		>
+			unpair
+		</button>
+	{/if}
+</div>

frontend/src/lib/components/StealthOverlay.svelte

@@ -0,0 +1,64 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+
+	let stealthMode = $state(false);
+	let lastTap = $state({ time: 0, x: 0, y: 0 });
+
+	onMount(() => {
+		document.body.style.touchAction = 'manipulation';
+		return () => {
+			document.body.style.touchAction = '';
+		};
+	});
+
+	function isInCenterZone(x: number, y: number): boolean {
+		const w = window.innerWidth;
+		const h = window.innerHeight;
+		return x > w * 0.3 && x < w * 0.7 && y > h * 0.3 && y < h * 0.7;
+	}
+
+	function handleTouchEnd(e: TouchEvent) {
+		if (e.touches.length > 0) return;
+
+		const target = e.target as HTMLElement;
+		if (target?.closest('[data-camera-ui]')) return;
+
+		const touch = e.changedTouches[0];
+		const now = Date.now();
+		const x = touch.clientX;
+		const y = touch.clientY;
+
+		if (!isInCenterZone(x, y)) {
+			lastTap = { time: 0, x: 0, y: 0 };
+			return;
+		}
+
+		const timeDiff = now - lastTap.time;
+		const distX = Math.abs(x - lastTap.x);
+		const distY = Math.abs(y - lastTap.y);
+
+		if (timeDiff < 500 && distX < 50 && distY < 50) {
+			stealthMode = !stealthMode;
+			lastTap = { time: 0, x: 0, y: 0 };
+			e.preventDefault();
+		} else {
+			lastTap = { time: now, x, y };
+		}
+	}
+</script>
+
+<svelte:document ontouchend={handleTouchEnd} />
+
+{#if stealthMode}
+	<div class="stealth-overlay" ontouchend={handleTouchEnd}></div>
+{/if}
+
+<style>
+	.stealth-overlay {
+		position: fixed;
+		inset: 0;
+		z-index: 9999;
+		background: #000;
+		touch-action: manipulation;
+	}
+</style>

frontend/src/lib/components/WatchCountdown.svelte

@@ -0,0 +1,29 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+
+	interface Props {
+		oncomplete: () => void;
+		oncancel: () => void;
+	}
+
+	let { oncomplete, oncancel }: Props = $props();
+
+	let count = $state(3);
+
+	onMount(() => {
+		const interval = setInterval(() => {
+			count--;
+			if (count === 0) {
+				clearInterval(interval);
+				oncomplete();
+			}
+		}, 1000);
+
+		return () => clearInterval(interval);
+	});
+</script>
+
+<div class="fixed inset-0 z-50 flex flex-col items-center justify-center bg-black" data-camera-ui>
+	<span class="text-8xl font-bold text-white">{count}</span>
+	<button onclick={oncancel} class="mt-8 text-sm text-neutral-400">Cancel</button>
+</div>

frontend/src/lib/convex-polling.svelte.ts

@@ -0,0 +1,113 @@
+import { ConvexHttpClient } from 'convex/browser';
+import { getContext, setContext } from 'svelte';
+import type { FunctionReference, FunctionArgs, FunctionReturnType } from 'convex/server';
+
+const POLLING_CONTEXT_KEY = 'convex-polling';
+const POLL_INTERVAL = 1000;
+
+type PollingContext = {
+	client: ConvexHttpClient;
+};
+
+export function hasWebSocketSupport(): boolean {
+	if (typeof window === 'undefined') return true;
+	try {
+		return 'WebSocket' in window && typeof WebSocket !== 'undefined';
+	} catch {
+		return false;
+	}
+}
+
+export function setupPollingConvex(url: string): void {
+	const client = new ConvexHttpClient(url);
+	setContext<PollingContext>(POLLING_CONTEXT_KEY, { client });
+}
+
+export function usePollingClient(): ConvexHttpClient {
+	const ctx = getContext<PollingContext>(POLLING_CONTEXT_KEY);
+	if (!ctx) {
+		throw new Error('Convex polling client not set up. Call setupPollingConvex first.');
+	}
+	return ctx.client;
+}
+
+type QueryState<T> = {
+	data: T | undefined;
+	error: Error | null;
+	isLoading: boolean;
+};
+
+export function usePollingMutation<Mutation extends FunctionReference<'mutation'>>(
+	mutation: Mutation
+): (args: FunctionArgs<Mutation>) => Promise<FunctionReturnType<Mutation>> {
+	const client = usePollingClient();
+	return (args: FunctionArgs<Mutation>) => client.mutation(mutation, args);
+}
+
+export function usePollingQuery<Query extends FunctionReference<'query'>>(
+	query: Query,
+	argsGetter: () => FunctionArgs<Query> | 'skip'
+): {
+	data: FunctionReturnType<Query> | undefined;
+	error: Error | null;
+	isLoading: boolean;
+} {
+	const client = usePollingClient();
+
+	// eslint-disable-next-line prefer-const
+	let state = $state<QueryState<FunctionReturnType<Query>>>({
+		data: undefined,
+		error: null,
+		isLoading: true
+	});
+
+	let intervalId: ReturnType<typeof setInterval> | null = null;
+	let lastArgsJson = '';
+
+	async function poll() {
+		const args = argsGetter();
+		if (args === 'skip') {
+			state.isLoading = false;
+			return;
+		}
+
+		const argsJson = JSON.stringify(args);
+		if (argsJson !== lastArgsJson) {
+			state.isLoading = true;
+			lastArgsJson = argsJson;
+		}
+
+		try {
+			const result = await client.query(query, args);
+			state.data = result;
+			state.error = null;
+			state.isLoading = false;
+		} catch (err) {
+			state.error = err instanceof Error ? err : new Error(String(err));
+			state.isLoading = false;
+		}
+	}
+
+	$effect(() => {
+		poll();
+		intervalId = setInterval(poll, POLL_INTERVAL);
+
+		return () => {
+			if (intervalId) {
+				clearInterval(intervalId);
+			}
+		};
+	});
+
+	return {
+		get data() {
+			return state.data;
+		},
+		get error() {
+			return state.error;
+		},
+		get isLoading() {
+			return state.isLoading;
+		}
+	};
+}

frontend/src/lib/convex/_generated/api.d.ts

@@ -0,0 +1,63 @@
+/* eslint-disable */
+/**
+ * Generated `api` utility.
+ *
+ * THIS CODE IS AUTOMATICALLY GENERATED.
+ *
+ * To regenerate, run `npx convex dev`.
+ * @module
+ */
+
+import type * as chats from "../chats.js";
+import type * as devicePairings from "../devicePairings.js";
+import type * as messages from "../messages.js";
+import type * as pairingRequests from "../pairingRequests.js";
+import type * as pendingGenerations from "../pendingGenerations.js";
+import type * as photoDrafts from "../photoDrafts.js";
+import type * as photoRequests from "../photoRequests.js";
+import type * as users from "../users.js";
+
+import type {
+  ApiFromModules,
+  FilterApi,
+  FunctionReference,
+} from "convex/server";
+
+declare const fullApi: ApiFromModules<{
+  chats: typeof chats;
+  devicePairings: typeof devicePairings;
+  messages: typeof messages;
+  pairingRequests: typeof pairingRequests;
+  pendingGenerations: typeof pendingGenerations;
+  photoDrafts: typeof photoDrafts;
+  photoRequests: typeof photoRequests;
+  users: typeof users;
+}>;
+
+/**
+ * A utility for referencing Convex functions in your app's public API.
+ *
+ * Usage:
+ * ```js
+ * const myFunctionReference = api.myModule.myFunction;
+ * ```
+ */
+export declare const api: FilterApi<
+  typeof fullApi,
+  FunctionReference<any, "public">
+>;
+
+/**
+ * A utility for referencing Convex functions in your app's internal API.
+ *
+ * Usage:
+ * ```js
+ * const myFunctionReference = internal.myModule.myFunction;
+ * ```
+ */
+export declare const internal: FilterApi<
+  typeof fullApi,
+  FunctionReference<any, "internal">
+>;
+
+export declare const components: {};

frontend/src/lib/convex/_generated/api.js

@@ -8,7 +8,7 @@
  * @module
  */
 
-import { anyApi, componentsGeneric } from 'convex/server';
+import { anyApi, componentsGeneric } from "convex/server";
 
 /**
  * A utility for referencing Convex functions in your app's API.

frontend/src/lib/convex/_generated/dataModel.d.ts

@@ -8,29 +8,29 @@
  * @module
  */
 
-import { AnyDataModel } from 'convex/server';
+import type {
-import type { GenericId } from 'convex/values';
+  DataModelFromSchemaDefinition,
-
+  DocumentByName,
-/**
+  TableNamesInDataModel,
- * No `schema.ts` file found!
+  SystemTableNames,
- *
+} from "convex/server";
- * This generated code has permissive types like `Doc = any` because
+import type { GenericId } from "convex/values";
- * Convex doesn't know your schema. If you'd like more type safety, see
+import schema from "../schema.js";
- * https://docs.convex.dev/using/schemas for instructions on how to add a
- * schema file.
- *
- * After you change a schema, rerun codegen with `npx convex dev`.
- */
 
 /**
  * The names of all of your Convex tables.
  */
-export type TableNames = string;
+export type TableNames = TableNamesInDataModel<DataModel>;
 
 /**
  * The type of a document stored in Convex.
+ *
+ * @typeParam TableName - A string literal type of the table name (like "users").
  */
-export type Doc = any;
+export type Doc<TableName extends TableNames> = DocumentByName<
+  DataModel,
+  TableName
+>;
 
 /**
  * An identifier for a document in Convex.
@@ -42,8 +42,11 @@ export type Doc = any;
  *
  * IDs are just strings at runtime, but this type can be used to distinguish them from other
  * strings when type checking.
+ *
+ * @typeParam TableName - A string literal type of the table name (like "users").
  */
-export type Id<TableName extends TableNames = TableNames> = GenericId<TableName>;
+export type Id<TableName extends TableNames | SystemTableNames> =
+  GenericId<TableName>;
 
 /**
  * A type describing your Convex data model.
@@ -54,4 +57,4 @@ export type Id<TableName extends TableNames = TableNames> = GenericId<TableName>
  * This type is used to parameterize methods like `queryGeneric` and
  * `mutationGeneric` to make them type-safe.
  */
-export type DataModel = AnyDataModel;
+export type DataModel = DataModelFromSchemaDefinition<typeof schema>;

frontend/src/lib/convex/_generated/server.d.ts

@@ -9,17 +9,17 @@
  */
 
 import {
-	ActionBuilder,
+  ActionBuilder,
-	HttpActionBuilder,
+  HttpActionBuilder,
-	MutationBuilder,
+  MutationBuilder,
-	QueryBuilder,
+  QueryBuilder,
-	GenericActionCtx,
+  GenericActionCtx,
-	GenericMutationCtx,
+  GenericMutationCtx,
-	GenericQueryCtx,
+  GenericQueryCtx,
-	GenericDatabaseReader,
+  GenericDatabaseReader,
-	GenericDatabaseWriter
+  GenericDatabaseWriter,
-} from 'convex/server';
+} from "convex/server";
-import type { DataModel } from './dataModel.js';
+import type { DataModel } from "./dataModel.js";
 
 /**
  * Define a query in this Convex app's public API.
@@ -29,7 +29,7 @@ import type { DataModel } from './dataModel.js';
  * @param func - The query function. It receives a {@link QueryCtx} as its first argument.
  * @returns The wrapped query. Include this as an `export` to name it and make it accessible.
  */
-export declare const query: QueryBuilder<DataModel, 'public'>;
+export declare const query: QueryBuilder<DataModel, "public">;
 
 /**
  * Define a query that is only accessible from other Convex functions (but not from the client).
@@ -39,7 +39,7 @@ export declare const query: QueryBuilder<DataModel, 'public'>;
  * @param func - The query function. It receives a {@link QueryCtx} as its first argument.
  * @returns The wrapped query. Include this as an `export` to name it and make it accessible.
  */
-export declare const internalQuery: QueryBuilder<DataModel, 'internal'>;
+export declare const internalQuery: QueryBuilder<DataModel, "internal">;
 
 /**
  * Define a mutation in this Convex app's public API.
@@ -49,7 +49,7 @@ export declare const internalQuery: QueryBuilder<DataModel, 'internal'>;
  * @param func - The mutation function. It receives a {@link MutationCtx} as its first argument.
  * @returns The wrapped mutation. Include this as an `export` to name it and make it accessible.
  */
-export declare const mutation: MutationBuilder<DataModel, 'public'>;
+export declare const mutation: MutationBuilder<DataModel, "public">;
 
 /**
  * Define a mutation that is only accessible from other Convex functions (but not from the client).
@@ -59,7 +59,7 @@ export declare const mutation: MutationBuilder<DataModel, 'public'>;
  * @param func - The mutation function. It receives a {@link MutationCtx} as its first argument.
  * @returns The wrapped mutation. Include this as an `export` to name it and make it accessible.
  */
-export declare const internalMutation: MutationBuilder<DataModel, 'internal'>;
+export declare const internalMutation: MutationBuilder<DataModel, "internal">;
 
 /**
  * Define an action in this Convex app's public API.
@@ -72,7 +72,7 @@ export declare const internalMutation: MutationBuilder<DataModel, 'internal'>;
  * @param func - The action. It receives an {@link ActionCtx} as its first argument.
  * @returns The wrapped action. Include this as an `export` to name it and make it accessible.
  */
-export declare const action: ActionBuilder<DataModel, 'public'>;
+export declare const action: ActionBuilder<DataModel, "public">;
 
 /**
  * Define an action that is only accessible from other Convex functions (but not from the client).
@@ -80,7 +80,7 @@ export declare const action: ActionBuilder<DataModel, 'public'>;
  * @param func - The function. It receives an {@link ActionCtx} as its first argument.
  * @returns The wrapped function. Include this as an `export` to name it and make it accessible.
  */
-export declare const internalAction: ActionBuilder<DataModel, 'internal'>;
+export declare const internalAction: ActionBuilder<DataModel, "internal">;
 
 /**
  * Define an HTTP action.

frontend/src/lib/convex/_generated/server.js

@@ -9,14 +9,14 @@
  */
 
 import {
-	actionGeneric,
+  actionGeneric,
-	httpActionGeneric,
+  httpActionGeneric,
-	queryGeneric,
+  queryGeneric,
-	mutationGeneric,
+  mutationGeneric,
-	internalActionGeneric,
+  internalActionGeneric,
-	internalMutationGeneric,
+  internalMutationGeneric,
-	internalQueryGeneric
+  internalQueryGeneric,
-} from 'convex/server';
+} from "convex/server";
 
 /**
  * Define a query in this Convex app's public API.

frontend/src/lib/convex/chats.ts

@@ -43,8 +43,15 @@ export const clear = mutation({
 			.collect();
 
 		for (const message of messages) {
-			if (args.preserveImages && message.imageStorageId) {
+			if (args.preserveImages) {
-				continue;
+				const hasLegacyImage = message.imageBase64 || message.imagesBase64?.length;
+				const messageImages = await ctx.db
+					.query('messageImages')
+					.withIndex('by_message_id', (q) => q.eq('messageId', message._id))
+					.first();
+				if (hasLegacyImage || messageImages) {
+					continue;
+				}
 			}
 			await ctx.db.delete(message._id);
 		}

frontend/src/lib/convex/devicePairings.ts

@@ -0,0 +1,109 @@
+import { v } from 'convex/values';
+import { mutation, query } from './_generated/server';
+
+export const register = mutation({
+	args: {
+		chatId: v.id('chats'),
+		deviceId: v.string(),
+		hasCamera: v.boolean()
+	},
+	returns: v.id('devicePairings'),
+	handler: async (ctx, args) => {
+		const existing = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		const device = existing.find((d) => d.deviceId === args.deviceId);
+
+		if (device) {
+			await ctx.db.patch(device._id, {
+				hasCamera: args.hasCamera,
+				lastSeen: Date.now()
+			});
+			return device._id;
+		}
+
+		return await ctx.db.insert('devicePairings', {
+			chatId: args.chatId,
+			deviceId: args.deviceId,
+			hasCamera: args.hasCamera,
+			lastSeen: Date.now()
+		});
+	}
+});
+
+export const heartbeat = mutation({
+	args: {
+		chatId: v.id('chats'),
+		deviceId: v.string()
+	},
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const devices = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		const device = devices.find((d) => d.deviceId === args.deviceId);
+		if (device) {
+			await ctx.db.patch(device._id, { lastSeen: Date.now() });
+		}
+		return null;
+	}
+});
+
+export const getMyDevice = query({
+	args: { chatId: v.id('chats'), deviceId: v.string() },
+	returns: v.union(
+		v.object({
+			_id: v.id('devicePairings'),
+			_creationTime: v.number(),
+			chatId: v.id('chats'),
+			deviceId: v.string(),
+			hasCamera: v.boolean(),
+			pairedWithDeviceId: v.optional(v.string()),
+			lastSeen: v.number()
+		}),
+		v.null()
+	),
+	handler: async (ctx, args) => {
+		const devices = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		return devices.find((d) => d.deviceId === args.deviceId) ?? null;
+	}
+});
+
+export const getPairedDevice = query({
+	args: { chatId: v.id('chats'), deviceId: v.string() },
+	returns: v.union(
+		v.object({
+			_id: v.id('devicePairings'),
+			_creationTime: v.number(),
+			chatId: v.id('chats'),
+			deviceId: v.string(),
+			hasCamera: v.boolean(),
+			pairedWithDeviceId: v.optional(v.string()),
+			lastSeen: v.number()
+		}),
+		v.null()
+	),
+	handler: async (ctx, args) => {
+		const devices = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		const myDevice = devices.find((d) => d.deviceId === args.deviceId);
+		if (!myDevice?.pairedWithDeviceId) return null;
+
+		const thirtySecondsAgo = Date.now() - 30000;
+		const paired = devices.find(
+			(d) => d.deviceId === myDevice.pairedWithDeviceId && d.lastSeen > thirtySecondsAgo
+		);
+		return paired ?? null;
+	}
+});

frontend/src/lib/convex/messages.ts

@@ -1,5 +1,6 @@
 import { v } from 'convex/values';
-import { internalMutation, mutation, query } from './_generated/server';
+import { mutation, query } from './_generated/server';
+import type { Id } from './_generated/dataModel';
 
 export const listByChat = query({
 	args: { chatId: v.id('chats') },
@@ -10,8 +11,6 @@ export const listByChat = query({
 			chatId: v.id('chats'),
 			role: v.union(v.literal('user'), v.literal('assistant')),
 			content: v.string(),
-			imageStorageId: v.optional(v.id('_storage')),
-			imageMediaType: v.optional(v.string()),
 			followUpOptions: v.optional(v.array(v.string())),
 			source: v.union(v.literal('telegram'), v.literal('web')),
 			createdAt: v.number(),
@@ -19,11 +18,23 @@ export const listByChat = query({
 		})
 	),
 	handler: async (ctx, args) => {
-		return await ctx.db
+		const messages = await ctx.db
 			.query('messages')
 			.withIndex('by_chat_id_and_created_at', (q) => q.eq('chatId', args.chatId))
 			.order('asc')
 			.collect();
+
+		return messages.map((m) => ({
+			_id: m._id,
+			_creationTime: m._creationTime,
+			chatId: m.chatId,
+			role: m.role,
+			content: m.content,
+			followUpOptions: m.followUpOptions,
+			source: m.source,
+			createdAt: m.createdAt,
+			isStreaming: m.isStreaming
+		}));
 	}
 });
 
@@ -33,8 +44,11 @@ export const create = mutation({
 		role: v.union(v.literal('user'), v.literal('assistant')),
 		content: v.string(),
 		source: v.union(v.literal('telegram'), v.literal('web')),
-		imageStorageId: v.optional(v.id('_storage')),
+		imageBase64: v.optional(v.string()),
 		imageMediaType: v.optional(v.string()),
+		imagesBase64: v.optional(v.array(v.string())),
+		imagesMediaTypes: v.optional(v.array(v.string())),
+		photoDraftIds: v.optional(v.array(v.id('photoDrafts'))),
 		followUpOptions: v.optional(v.array(v.string())),
 		isStreaming: v.optional(v.boolean())
 	},
@@ -45,25 +59,59 @@ export const create = mutation({
 			role: args.role,
 			content: args.content,
 			source: args.source,
-			imageStorageId: args.imageStorageId,
+			imageBase64: args.imageBase64,
 			imageMediaType: args.imageMediaType,
+			imagesBase64: args.imagesBase64,
+			imagesMediaTypes: args.imagesMediaTypes,
 			followUpOptions: args.followUpOptions,
 			createdAt: Date.now(),
 			isStreaming: args.isStreaming
 		});
 
+		const drafts: Array<{ base64: string; mediaType: string; id: Id<'photoDrafts'> }> = [];
+		if (args.photoDraftIds && args.photoDraftIds.length > 0) {
+			for (const draftId of args.photoDraftIds) {
+				const draft = await ctx.db.get(draftId);
+				if (draft) {
+					drafts.push({ base64: draft.base64, mediaType: draft.mediaType, id: draft._id });
+				}
+			}
+		}
+
+		for (let i = 0; i < drafts.length; i++) {
+			await ctx.db.insert('messageImages', {
+				messageId,
+				base64: drafts[i].base64,
+				mediaType: drafts[i].mediaType,
+				order: i
+			});
+		}
+
 		if (args.source === 'web' && args.role === 'user') {
 			const chat = await ctx.db.get(args.chatId);
 			if (chat) {
-				await ctx.db.insert('pendingGenerations', {
+				const pendingGenId = await ctx.db.insert('pendingGenerations', {
 					userId: chat.userId,
 					chatId: args.chatId,
 					userMessage: args.content,
 					createdAt: Date.now()
 				});
+
+				for (let i = 0; i < drafts.length; i++) {
+					await ctx.db.insert('pendingGenerationImages', {
+						pendingGenerationId: pendingGenId,
+						base64: drafts[i].base64,
+						mediaType: drafts[i].mediaType,
+						order: i
+					});
+				}
 			}
 		}
 
+		for (const draft of drafts) {
+			await ctx.db.delete(draft.id);
+		}
+
 		return messageId;
 	}
 });
@@ -132,8 +180,10 @@ export const getLastAssistantMessage = query({
 			chatId: v.id('chats'),
 			role: v.union(v.literal('user'), v.literal('assistant')),
 			content: v.string(),
-			imageStorageId: v.optional(v.id('_storage')),
+			imageBase64: v.optional(v.string()),
 			imageMediaType: v.optional(v.string()),
+			imagesBase64: v.optional(v.array(v.string())),
+			imagesMediaTypes: v.optional(v.array(v.string())),
 			followUpOptions: v.optional(v.array(v.string())),
 			source: v.union(v.literal('telegram'), v.literal('web')),
 			createdAt: v.number(),
@@ -152,21 +202,12 @@ export const getLastAssistantMessage = query({
 	}
 });
 
-export const generateUploadUrl = mutation({
+export const getChatImages = query({
-	args: {},
-	returns: v.string(),
-	handler: async (ctx) => {
-		return await ctx.storage.generateUploadUrl();
-	}
-});
-
-export const getImageUrls = query({
 	args: { chatId: v.id('chats') },
 	returns: v.array(
 		v.object({
-			storageId: v.id('_storage'),
+			base64: v.string(),
-			mediaType: v.string(),
+			mediaType: v.string()
-			url: v.union(v.string(), v.null())
 		})
 	),
 	handler: async (ctx, args) => {
@@ -175,41 +216,33 @@ export const getImageUrls = query({
 			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
 			.collect();
 
-		const imageMessages = messages.filter((m) => m.imageStorageId && m.imageMediaType);
+		const images: Array<{ base64: string; mediaType: string }> = [];
-		const results = [];
 
-		for (const msg of imageMessages) {
+		for (const m of messages) {
-			if (msg.imageStorageId && msg.imageMediaType) {
+			const msgImages = await ctx.db
-				const url = await ctx.storage.getUrl(msg.imageStorageId);
+				.query('messageImages')
-				results.push({
+				.withIndex('by_message_id', (q) => q.eq('messageId', m._id))
-					storageId: msg.imageStorageId,
+				.collect();
-					mediaType: msg.imageMediaType,
+
-					url
+			for (const img of msgImages.sort((a, b) => a.order - b.order)) {
+				images.push({ base64: img.base64, mediaType: img.mediaType });
+			}
+
+			if (m.imagesBase64 && m.imagesMediaTypes) {
+				for (let i = 0; i < m.imagesBase64.length; i++) {
+					images.push({
+						base64: m.imagesBase64[i],
+						mediaType: m.imagesMediaTypes[i]
+					});
+				}
+			} else if (m.imageBase64 && m.imageMediaType) {
+				images.push({
+					base64: m.imageBase64,
+					mediaType: m.imageMediaType
 				});
 			}
 		}
 
-		return results;
+		return images;
-	}
-});
-
-export const createWithImage = internalMutation({
-	args: {
-		chatId: v.id('chats'),
-		content: v.string(),
-		imageStorageId: v.id('_storage'),
-		imageMediaType: v.string()
-	},
-	returns: v.id('messages'),
-	handler: async (ctx, args) => {
-		return await ctx.db.insert('messages', {
-			chatId: args.chatId,
-			role: 'user' as const,
-			content: args.content,
-			source: 'telegram' as const,
-			imageStorageId: args.imageStorageId,
-			imageMediaType: args.imageMediaType,
-			createdAt: Date.now()
-		});
 	}
 });

frontend/src/lib/convex/pairingRequests.ts

@@ -0,0 +1,122 @@
+import { v } from 'convex/values';
+import { mutation, query } from './_generated/server';
+
+export const create = mutation({
+	args: {
+		chatId: v.id('chats'),
+		fromDeviceId: v.string()
+	},
+	returns: v.id('pairingRequests'),
+	handler: async (ctx, args) => {
+		const existing = await ctx.db
+			.query('pairingRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		const pending = existing.find(
+			(r) => r.fromDeviceId === args.fromDeviceId && r.status === 'pending'
+		);
+		if (pending) return pending._id;
+
+		return await ctx.db.insert('pairingRequests', {
+			chatId: args.chatId,
+			fromDeviceId: args.fromDeviceId,
+			status: 'pending',
+			createdAt: Date.now()
+		});
+	}
+});
+
+export const accept = mutation({
+	args: {
+		requestId: v.id('pairingRequests'),
+		acceptingDeviceId: v.string()
+	},
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const request = await ctx.db.get(args.requestId);
+		if (!request || request.status !== 'pending') return null;
+
+		await ctx.db.patch(args.requestId, { status: 'accepted' });
+
+		const devices = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', request.chatId))
+			.collect();
+
+		const fromDevice = devices.find((d) => d.deviceId === request.fromDeviceId);
+		const acceptingDevice = devices.find((d) => d.deviceId === args.acceptingDeviceId);
+
+		if (fromDevice) {
+			await ctx.db.patch(fromDevice._id, { pairedWithDeviceId: args.acceptingDeviceId });
+		}
+		if (acceptingDevice) {
+			await ctx.db.patch(acceptingDevice._id, { pairedWithDeviceId: request.fromDeviceId });
+		}
+
+		return null;
+	}
+});
+
+export const reject = mutation({
+	args: { requestId: v.id('pairingRequests') },
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const request = await ctx.db.get(args.requestId);
+		if (!request || request.status !== 'pending') return null;
+		await ctx.db.patch(args.requestId, { status: 'rejected' });
+		return null;
+	}
+});
+
+export const getPending = query({
+	args: { chatId: v.id('chats'), excludeDeviceId: v.string() },
+	returns: v.union(
+		v.object({
+			_id: v.id('pairingRequests'),
+			_creationTime: v.number(),
+			chatId: v.id('chats'),
+			fromDeviceId: v.string(),
+			status: v.union(v.literal('pending'), v.literal('accepted'), v.literal('rejected')),
+			createdAt: v.number()
+		}),
+		v.null()
+	),
+	handler: async (ctx, args) => {
+		const requests = await ctx.db
+			.query('pairingRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		return (
+			requests.find((r) => r.status === 'pending' && r.fromDeviceId !== args.excludeDeviceId) ??
+			null
+		);
+	}
+});
+
+export const unpair = mutation({
+	args: {
+		chatId: v.id('chats'),
+		deviceId: v.string()
+	},
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const devices = await ctx.db
+			.query('devicePairings')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.collect();
+
+		const myDevice = devices.find((d) => d.deviceId === args.deviceId);
+		if (!myDevice?.pairedWithDeviceId) return null;
+
+		const pairedDevice = devices.find((d) => d.deviceId === myDevice.pairedWithDeviceId);
+
+		await ctx.db.patch(myDevice._id, { pairedWithDeviceId: undefined });
+		if (pairedDevice) {
+			await ctx.db.patch(pairedDevice._id, { pairedWithDeviceId: undefined });
+		}
+
+		return null;
+	}
+});

frontend/src/lib/convex/pendingGenerations.ts

@@ -10,11 +10,33 @@ export const list = query({
 			userId: v.id('users'),
 			chatId: v.id('chats'),
 			userMessage: v.string(),
+			imagesBase64: v.optional(v.array(v.string())),
+			imagesMediaTypes: v.optional(v.array(v.string())),
 			createdAt: v.number()
 		})
 	),
 	handler: async (ctx) => {
-		return await ctx.db.query('pendingGenerations').collect();
+		const pending = await ctx.db.query('pendingGenerations').collect();
+
+		const result = [];
+		for (const p of pending) {
+			const images = await ctx.db
+				.query('pendingGenerationImages')
+				.withIndex('by_pending_generation_id', (q) => q.eq('pendingGenerationId', p._id))
+				.collect();
+
+			const sortedImages = images.sort((a, b) => a.order - b.order);
+
+			result.push({
+				...p,
+				imagesBase64:
+					sortedImages.length > 0 ? sortedImages.map((img) => img.base64) : p.imagesBase64,
+				imagesMediaTypes:
+					sortedImages.length > 0 ? sortedImages.map((img) => img.mediaType) : p.imagesMediaTypes
+			});
+		}
+
+		return result;
 	}
 });
 
@@ -39,7 +61,35 @@ export const remove = mutation({
 	args: { id: v.id('pendingGenerations') },
 	returns: v.null(),
 	handler: async (ctx, args) => {
+		const images = await ctx.db
+			.query('pendingGenerationImages')
+			.withIndex('by_pending_generation_id', (q) => q.eq('pendingGenerationId', args.id))
+			.collect();
+		for (const img of images) {
+			await ctx.db.delete(img._id);
+		}
 		await ctx.db.delete(args.id);
 		return null;
 	}
 });
+
+export const getImages = query({
+	args: { pendingGenerationId: v.id('pendingGenerations') },
+	returns: v.array(
+		v.object({
+			base64: v.string(),
+			mediaType: v.string()
+		})
+	),
+	handler: async (ctx, args) => {
+		const images = await ctx.db
+			.query('pendingGenerationImages')
+			.withIndex('by_pending_generation_id', (q) =>
+				q.eq('pendingGenerationId', args.pendingGenerationId)
+			)
+			.collect();
+		return images
+			.sort((a, b) => a.order - b.order)
+			.map((img) => ({ base64: img.base64, mediaType: img.mediaType }));
+	}
+});

frontend/src/lib/convex/photoDrafts.ts

@@ -0,0 +1,93 @@
+import { v } from 'convex/values';
+import { mutation, query } from './_generated/server';
+
+const photoValidator = v.object({
+	base64: v.string(),
+	mediaType: v.string()
+});
+
+export const get = query({
+	args: { chatId: v.id('chats'), deviceId: v.string() },
+	returns: v.object({
+		photos: v.array(
+			v.object({
+				_id: v.id('photoDrafts'),
+				mediaType: v.string()
+			})
+		)
+	}),
+	handler: async (ctx, args) => {
+		const drafts = await ctx.db
+			.query('photoDrafts')
+			.withIndex('by_chat_id_and_device_id', (q) =>
+				q.eq('chatId', args.chatId).eq('deviceId', args.deviceId)
+			)
+			.collect();
+
+		return {
+			photos: drafts.map((d) => ({
+				_id: d._id,
+				mediaType: d.mediaType
+			}))
+		};
+	}
+});
+
+export const addPhoto = mutation({
+	args: {
+		chatId: v.id('chats'),
+		deviceId: v.string(),
+		photo: photoValidator
+	},
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		await ctx.db.insert('photoDrafts', {
+			chatId: args.chatId,
+			deviceId: args.deviceId,
+			base64: args.photo.base64,
+			mediaType: args.photo.mediaType,
+			createdAt: Date.now()
+		});
+		return null;
+	}
+});
+
+export const removePhoto = mutation({
+	args: {
+		chatId: v.id('chats'),
+		deviceId: v.string(),
+		index: v.number()
+	},
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const drafts = await ctx.db
+			.query('photoDrafts')
+			.withIndex('by_chat_id_and_device_id', (q) =>
+				q.eq('chatId', args.chatId).eq('deviceId', args.deviceId)
+			)
+			.collect();
+
+		if (drafts[args.index]) {
+			await ctx.db.delete(drafts[args.index]._id);
+		}
+		return null;
+	}
+});
+
+export const clear = mutation({
+	args: { chatId: v.id('chats'), deviceId: v.string() },
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		const drafts = await ctx.db
+			.query('photoDrafts')
+			.withIndex('by_chat_id_and_device_id', (q) =>
+				q.eq('chatId', args.chatId).eq('deviceId', args.deviceId)
+			)
+			.collect();
+
+		for (const draft of drafts) {
+			await ctx.db.delete(draft._id);
+		}
+		return null;
+	}
+});

frontend/src/lib/convex/photoRequests.ts

@@ -0,0 +1,293 @@
+import { v } from 'convex/values';
+import { mutation, query } from './_generated/server';
+
+const photoRequestValidator = v.object({
+	_id: v.id('photoRequests'),
+	_creationTime: v.number(),
+	chatId: v.id('chats'),
+	requesterId: v.string(),
+	captureDeviceId: v.optional(v.string()),
+	status: v.union(
+		v.literal('pending'),
+		v.literal('countdown'),
+		v.literal('capture_now'),
+		v.literal('captured'),
+		v.literal('accepted'),
+		v.literal('rejected')
+	),
+	photoBase64: v.optional(v.string()),
+	photoMediaType: v.optional(v.string()),
+	thumbnailBase64: v.optional(v.string()),
+	createdAt: v.number()
+});
+
+const photoRequestLightValidator = v.object({
+	_id: v.id('photoRequests'),
+	status: v.union(
+		v.literal('pending'),
+		v.literal('countdown'),
+		v.literal('capture_now'),
+		v.literal('captured'),
+		v.literal('accepted'),
+		v.literal('rejected')
+	),
+	photoMediaType: v.optional(v.string()),
+	thumbnailBase64: v.optional(v.string())
+});
+
+export const create = mutation({
+	args: {
+		chatId: v.id('chats'),
+		requesterId: v.string(),
+		captureDeviceId: v.string()
+	},
+	returns: v.id('photoRequests'),
+	handler: async (ctx, args) => {
+		const oldRequests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.take(20);
+
+		for (const req of oldRequests) {
+			if (req.status === 'pending' || req.status === 'countdown' || req.status === 'capture_now') {
+				await ctx.db.patch(req._id, { status: 'rejected' });
+			}
+		}
+
+		return await ctx.db.insert('photoRequests', {
+			chatId: args.chatId,
+			requesterId: args.requesterId,
+			captureDeviceId: args.captureDeviceId,
+			status: 'countdown',
+			createdAt: Date.now()
+		});
+	}
+});
+
+export const markCaptureNow = mutation({
+	args: { requestId: v.id('photoRequests') },
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		await ctx.db.patch(args.requestId, { status: 'capture_now' });
+		return null;
+	}
+});
+
+export const submitPhoto = mutation({
+	args: {
+		requestId: v.id('photoRequests'),
+		photoBase64: v.string(),
+		photoMediaType: v.string(),
+		thumbnailBase64: v.optional(v.string())
+	},
+	returns: v.boolean(),
+	handler: async (ctx, args) => {
+		const req = await ctx.db.get(args.requestId);
+		if (!req || req.status !== 'capture_now') {
+			return false;
+		}
+
+		await ctx.db.patch(args.requestId, {
+			status: 'captured',
+			photoBase64: args.photoBase64,
+			photoMediaType: args.photoMediaType,
+			thumbnailBase64: args.thumbnailBase64
+		});
+		return true;
+	}
+});
+
+export const markAccepted = mutation({
+	args: { requestId: v.id('photoRequests') },
+	returns: v.null(),
+	handler: async (ctx, args) => {
+		await ctx.db.patch(args.requestId, { status: 'accepted' });
+		return null;
+	}
+});
+
+export const markRejected = mutation({
+	args: { requestId: v.id('photoRequests') },
+	returns: v.boolean(),
+	handler: async (ctx, args) => {
+		const req = await ctx.db.get(args.requestId);
+		if (!req || req.status === 'accepted' || req.status === 'rejected') {
+			return false;
+		}
+		await ctx.db.patch(req._id, { status: 'rejected' });
+		return true;
+	}
+});
+
+const captureNowLightValidator = v.object({
+	_id: v.id('photoRequests'),
+	status: v.literal('capture_now')
+});
+
+export const getCaptureNowRequest = query({
+	args: { chatId: v.id('chats'), deviceId: v.optional(v.string()) },
+	returns: v.union(captureNowLightValidator, v.null()),
+	handler: async (ctx, args) => {
+		const now = Date.now();
+		const maxAge = 60 * 1000;
+
+		const requests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.order('desc')
+			.take(50);
+
+		const found = requests.find((r) => r.status === 'capture_now' && now - r.createdAt < maxAge);
+		if (!found) return null;
+
+		return { _id: found._id, status: 'capture_now' as const };
+	}
+});
+
+export const getActiveForCapture = query({
+	args: { chatId: v.id('chats'), deviceId: v.optional(v.string()) },
+	returns: v.union(photoRequestValidator, v.null()),
+	handler: async (ctx, args) => {
+		const requests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.order('desc')
+			.take(50);
+
+		return requests.find((r) => r.status === 'countdown' || r.status === 'capture_now') ?? null;
+	}
+});
+
+export const getMyActiveRequest = query({
+	args: { chatId: v.id('chats'), deviceId: v.optional(v.string()) },
+	returns: v.union(photoRequestLightValidator, v.null()),
+	handler: async (ctx, args) => {
+		const requests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.order('desc')
+			.take(100);
+
+		if (!args.deviceId) return null;
+
+		const found = requests.find(
+			(r) =>
+				r.requesterId === args.deviceId &&
+				(r.status === 'countdown' || r.status === 'capture_now' || r.status === 'captured')
+		);
+
+		if (!found) return null;
+
+		return {
+			_id: found._id,
+			status: found.status,
+			photoMediaType: found.photoMediaType,
+			thumbnailBase64: found.thumbnailBase64
+		};
+	}
+});
+
+export const getPhotoData = query({
+	args: { requestId: v.id('photoRequests') },
+	returns: v.union(
+		v.object({
+			photoBase64: v.string(),
+			photoMediaType: v.string(),
+			thumbnailBase64: v.optional(v.string())
+		}),
+		v.null()
+	),
+	handler: async (ctx, args) => {
+		const req = await ctx.db.get(args.requestId);
+		if (!req || !req.photoBase64 || !req.photoMediaType) return null;
+
+		return {
+			photoBase64: req.photoBase64,
+			photoMediaType: req.photoMediaType,
+			thumbnailBase64: req.thumbnailBase64
+		};
+	}
+});
+
+export const getPhotoPreview = query({
+	args: { requestId: v.id('photoRequests') },
+	returns: v.union(
+		v.object({
+			thumbnailBase64: v.string(),
+			photoMediaType: v.string()
+		}),
+		v.null()
+	),
+	handler: async (ctx, args) => {
+		const req = await ctx.db.get(args.requestId);
+		if (!req || !req.photoMediaType) return null;
+
+		return {
+			thumbnailBase64: req.thumbnailBase64 || req.photoBase64 || '',
+			photoMediaType: req.photoMediaType
+		};
+	}
+});
+
+export const acceptPhotoToDraft = mutation({
+	args: {
+		requestId: v.id('photoRequests'),
+		chatId: v.id('chats'),
+		deviceId: v.string()
+	},
+	returns: v.id('photoDrafts'),
+	handler: async (ctx, args) => {
+		const req = await ctx.db.get(args.requestId);
+		if (!req || !req.photoBase64 || !req.photoMediaType) {
+			throw new Error('Photo request not found or has no photo');
+		}
+
+		const draftId = await ctx.db.insert('photoDrafts', {
+			chatId: args.chatId,
+			deviceId: args.deviceId,
+			base64: req.photoBase64,
+			mediaType: req.photoMediaType,
+			createdAt: Date.now()
+		});
+
+		await ctx.db.patch(args.requestId, { status: 'accepted' });
+
+		return draftId;
+	}
+});
+
+export const cleanup = mutation({
+	args: { chatId: v.id('chats') },
+	returns: v.number(),
+	handler: async (ctx, args) => {
+		const requests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.take(20);
+
+		let deleted = 0;
+		for (const req of requests) {
+			await ctx.db.delete(req._id);
+			deleted++;
+		}
+		return deleted;
+	}
+});
+
+export const getCapturedForPhone = query({
+	args: { chatId: v.id('chats'), deviceId: v.optional(v.string()) },
+	returns: v.union(photoRequestValidator, v.null()),
+	handler: async (ctx, args) => {
+		const requests = await ctx.db
+			.query('photoRequests')
+			.withIndex('by_chat_id', (q) => q.eq('chatId', args.chatId))
+			.order('desc')
+			.take(50);
+
+		if (!args.deviceId) return null;
+
+		return (
+			requests.find((r) => r.captureDeviceId === args.deviceId && r.status === 'captured') ?? null
+		);
+	}
+});

frontend/src/lib/convex/schema.ts

@@ -23,8 +23,11 @@ export default defineSchema({
 		chatId: v.id('chats'),
 		role: v.union(v.literal('user'), v.literal('assistant')),
 		content: v.string(),
-		imageStorageId: v.optional(v.id('_storage')),
+		imageBase64: v.optional(v.string()),
 		imageMediaType: v.optional(v.string()),
+		imageStorageId: v.optional(v.id('_storage')),
+		imagesBase64: v.optional(v.array(v.string())),
+		imagesMediaTypes: v.optional(v.array(v.string())),
 		followUpOptions: v.optional(v.array(v.string())),
 		source: v.union(v.literal('telegram'), v.literal('web')),
 		createdAt: v.number(),
@@ -37,6 +40,63 @@ export default defineSchema({
 		userId: v.id('users'),
 		chatId: v.id('chats'),
 		userMessage: v.string(),
+		imagesBase64: v.optional(v.array(v.string())),
+		imagesMediaTypes: v.optional(v.array(v.string())),
 		createdAt: v.number()
-	})
+	}),
+
+	pendingGenerationImages: defineTable({
+		pendingGenerationId: v.id('pendingGenerations'),
+		base64: v.string(),
+		mediaType: v.string(),
+		order: v.number()
+	}).index('by_pending_generation_id', ['pendingGenerationId']),
+
+	messageImages: defineTable({
+		messageId: v.id('messages'),
+		base64: v.string(),
+		mediaType: v.string(),
+		order: v.number()
+	}).index('by_message_id', ['messageId']),
+
+	devicePairings: defineTable({
+		chatId: v.id('chats'),
+		deviceId: v.string(),
+		hasCamera: v.boolean(),
+		pairedWithDeviceId: v.optional(v.string()),
+		lastSeen: v.number()
+	}).index('by_chat_id', ['chatId']),
+
+	pairingRequests: defineTable({
+		chatId: v.id('chats'),
+		fromDeviceId: v.string(),
+		status: v.union(v.literal('pending'), v.literal('accepted'), v.literal('rejected')),
+		createdAt: v.number()
+	}).index('by_chat_id', ['chatId']),
+
+	photoRequests: defineTable({
+		chatId: v.id('chats'),
+		requesterId: v.string(),
+		captureDeviceId: v.optional(v.string()),
+		status: v.union(
+			v.literal('pending'),
+			v.literal('countdown'),
+			v.literal('capture_now'),
+			v.literal('captured'),
+			v.literal('accepted'),
+			v.literal('rejected')
+		),
+		photoBase64: v.optional(v.string()),
+		photoMediaType: v.optional(v.string()),
+		thumbnailBase64: v.optional(v.string()),
+		createdAt: v.number()
+	}).index('by_chat_id', ['chatId']),
+
+	photoDrafts: defineTable({
+		chatId: v.id('chats'),
+		deviceId: v.string(),
+		base64: v.string(),
+		mediaType: v.string(),
+		createdAt: v.number()
+	}).index('by_chat_id_and_device_id', ['chatId', 'deviceId'])
 });

frontend/src/routes/+layout.svelte

@@ -3,10 +3,19 @@
 	import favicon from '$lib/assets/favicon.svg';
 	import { PUBLIC_CONVEX_URL } from '$env/static/public';
 	import { setupConvex } from 'convex-svelte';
+	import { hasWebSocketSupport, setupPollingConvex } from '$lib/convex-polling.svelte';
+	import { setContext } from 'svelte';
 
 	let { children } = $props();
 
-	setupConvex(PUBLIC_CONVEX_URL);
+	const usePolling = !hasWebSocketSupport();
+	setContext('convex-use-polling', usePolling);
+
+	if (usePolling) {
+		setupPollingConvex(PUBLIC_CONVEX_URL);
+	} else {
+		setupConvex(PUBLIC_CONVEX_URL);
+	}
 </script>
 
 <svelte:head><link rel="icon" href={favicon} /></svelte:head>

frontend/src/routes/[mnemonic]/+page.svelte

@@ -1,18 +1,120 @@
 <script lang="ts">
 	import { page } from '$app/state';
+	import { browser } from '$app/environment';
+	import { getContext, onMount } from 'svelte';
+	import { SvelteSet } from 'svelte/reactivity';
 	import { useQuery, useConvexClient } from 'convex-svelte';
+	import {
+		usePollingQuery,
+		usePollingMutation,
+		usePollingClient
+	} from '$lib/convex-polling.svelte';
 	import { api } from '$lib/convex/_generated/api';
+	import type { Id } from '$lib/convex/_generated/dataModel';
 	import ChatMessage from '$lib/components/ChatMessage.svelte';
 	import ChatInput from '$lib/components/ChatInput.svelte';
 	import FollowUpButtons from '$lib/components/FollowUpButtons.svelte';
+	import StealthOverlay from '$lib/components/StealthOverlay.svelte';
+	import CameraCapture from '$lib/components/CameraCapture.svelte';
+	import WatchCountdown from '$lib/components/WatchCountdown.svelte';
+	import PhotoPreview from '$lib/components/PhotoPreview.svelte';
+	import DraftBadge from '$lib/components/DraftBadge.svelte';
+	import SilentCapture from '$lib/components/SilentCapture.svelte';
 
+	const usePolling = getContext<boolean>('convex-use-polling') ?? false;
 	let mnemonic = $derived(page.params.mnemonic);
-	const client = useConvexClient();
 
-	const chatData = useQuery(api.chats.getWithUser, () => (mnemonic ? { mnemonic } : 'skip'));
+	let lastMessageElement: HTMLDivElement | null = $state(null);
-	const messagesQuery = useQuery(api.messages.listByChat, () =>
+	let showScrollButton = $state(false);
-		chatData.data?.chat?._id ? { chatId: chatData.data.chat._id } : 'skip'
+
-	);
+	let deviceId = $state('');
+	let hasCamera = $state(false);
+	let showCamera = $state(false);
+	let showWatchCountdown = $state(false);
+	let activeRequestId: Id<'photoRequests'> | null = $state(null);
+	let previewPhoto: {
+		thumbnail: string;
+		mediaType: string;
+		requestId: Id<'photoRequests'>;
+	} | null = $state(null);
+	let shownPreviewIds = new SvelteSet<string>();
+	let silentCaptureRef: SilentCapture | null = $state(null);
+	let processedCaptureNowIds = new SvelteSet<string>();
+
+	function generateId(): string {
+		if (typeof crypto !== 'undefined' && crypto.randomUUID) {
+			return crypto.randomUUID();
+		}
+		return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, (c) => {
+			const r = (Math.random() * 16) | 0;
+			const v = c === 'x' ? r : (r & 0x3) | 0x8;
+			return v.toString(16);
+		});
+	}
+
+	function getOrCreateDeviceId(): string {
+		if (!browser) return '';
+		let id = localStorage.getItem('stealth-device-id');
+		if (!id) {
+			id = generateId();
+			localStorage.setItem('stealth-device-id', id);
+		}
+		return id;
+	}
+
+	async function checkCamera(): Promise<boolean> {
+		if (!browser) return false;
+		if (!navigator.mediaDevices?.enumerateDevices) return false;
+		try {
+			const devices = await navigator.mediaDevices.enumerateDevices();
+			return devices.some((d) => d.kind === 'videoinput');
+		} catch {
+			return false;
+		}
+	}
+
+	onMount(() => {
+		deviceId = getOrCreateDeviceId();
+		checkCamera().then((has) => {
+			hasCamera = has;
+		});
+	});
+
+	$effect(() => {
+		if (!lastMessageElement) return;
+
+		const observer = new IntersectionObserver(
+			([entry]) => {
+				showScrollButton = !entry.isIntersecting;
+			},
+			{ threshold: 0, rootMargin: '0px 0px -90% 0px' }
+		);
+
+		observer.observe(lastMessageElement);
+		return () => observer.disconnect();
+	});
+
+	function scrollToLastMessage() {
+		lastMessageElement?.scrollIntoView({ behavior: 'smooth', block: 'start' });
+	}
+
+	const chatDataWs = usePolling
+		? null
+		: useQuery(api.chats.getWithUser, () => (mnemonic ? { mnemonic } : 'skip'));
+	const chatDataPoll = usePolling
+		? usePollingQuery(api.chats.getWithUser, () => (mnemonic ? { mnemonic } : 'skip'))
+		: null;
+	const chatData = $derived(usePolling ? chatDataPoll! : chatDataWs!);
+
+	const chatId = $derived(chatData.data?.chat?._id);
+
+	const messagesQueryWs = usePolling
+		? null
+		: useQuery(api.messages.listByChat, () => (chatId ? { chatId } : 'skip'));
+	const messagesQueryPoll = usePolling
+		? usePollingQuery(api.messages.listByChat, () => (chatId ? { chatId } : 'skip'))
+		: null;
+	const messagesQuery = $derived(usePolling ? messagesQueryPoll! : messagesQueryWs!);
 
 	let messages = $derived(messagesQuery.data ?? []);
 	let lastMessage = $derived(messages[messages.length - 1]);
@@ -22,79 +124,555 @@
 			: []
 	);
 
+	const myDeviceWs = usePolling
+		? null
+		: useQuery(api.devicePairings.getMyDevice, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			);
+	const myDevicePoll = usePolling
+		? usePollingQuery(api.devicePairings.getMyDevice, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			)
+		: null;
+	const myDevice = $derived(usePolling ? myDevicePoll! : myDeviceWs!);
+
+	const pairedDeviceWs = usePolling
+		? null
+		: useQuery(api.devicePairings.getPairedDevice, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			);
+	const pairedDevicePoll = usePolling
+		? usePollingQuery(api.devicePairings.getPairedDevice, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			)
+		: null;
+	const pairedDevice = $derived(usePolling ? pairedDevicePoll! : pairedDeviceWs!);
+
+	const isPaired = $derived(!!myDevice.data?.pairedWithDeviceId && !!pairedDevice.data);
+
+	const pendingPairingWs = usePolling
+		? null
+		: useQuery(api.pairingRequests.getPending, () =>
+				chatId && deviceId ? { chatId, excludeDeviceId: deviceId } : 'skip'
+			);
+	const pendingPairingPoll = usePolling
+		? usePollingQuery(api.pairingRequests.getPending, () =>
+				chatId && deviceId ? { chatId, excludeDeviceId: deviceId } : 'skip'
+			)
+		: null;
+	const pendingPairing = $derived(usePolling ? pendingPairingPoll! : pendingPairingWs!);
+
+	const captureNowRequestWs = usePolling
+		? null
+		: useQuery(api.photoRequests.getCaptureNowRequest, () => (chatId ? { chatId } : 'skip'));
+	const captureNowRequestPoll = usePolling
+		? usePollingQuery(api.photoRequests.getCaptureNowRequest, () => (chatId ? { chatId } : 'skip'))
+		: null;
+	const captureNowRequest = $derived(usePolling ? captureNowRequestPoll! : captureNowRequestWs!);
+
+	const myActiveRequestWs = usePolling
+		? null
+		: useQuery(api.photoRequests.getMyActiveRequest, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			);
+	const myActiveRequestPoll = usePolling
+		? usePollingQuery(api.photoRequests.getMyActiveRequest, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			)
+		: null;
+	const myActiveRequest = $derived(usePolling ? myActiveRequestPoll! : myActiveRequestWs!);
+
+	const photoDraftWs = usePolling
+		? null
+		: useQuery(api.photoDrafts.get, () => (chatId && deviceId ? { chatId, deviceId } : 'skip'));
+	const photoDraftPoll = usePolling
+		? usePollingQuery(api.photoDrafts.get, () =>
+				chatId && deviceId ? { chatId, deviceId } : 'skip'
+			)
+		: null;
+	const photoDraft = $derived(usePolling ? photoDraftPoll! : photoDraftWs!);
+	const draftPhotos = $derived(photoDraft.data?.photos ?? []);
+
 	$effect(() => {
-		if (messages.length) {
+		const req = captureNowRequest.data;
+		if (req && hasCamera && !processedCaptureNowIds.has(req._id)) {
+			processedCaptureNowIds.add(req._id);
+			const tryCapture = () => {
+				const success = silentCaptureRef?.capture();
+				if (!success) {
+					setTimeout(tryCapture, 100);
+				}
+			};
+			tryCapture();
+		}
+	});
+
+	$effect(() => {
+		const req = myActiveRequest.data;
+		if (req?.status === 'captured' && req.photoMediaType) {
+			if (shownPreviewIds.has(req._id)) return;
+			shownPreviewIds.add(req._id);
+
+			const client = pollingClient ?? clientWs;
+			if (client) {
+				client.query(api.photoRequests.getPhotoPreview, { requestId: req._id }).then((data) => {
+					if (data) {
+						previewPhoto = {
+							thumbnail: data.thumbnailBase64,
+							mediaType: data.photoMediaType,
+							requestId: req._id
+						};
+					}
+				});
+			}
+		}
+	});
+
+	let prevMessageCount = 0;
+	let prevLastMessageId: string | undefined;
+
+	$effect(() => {
+		const count = messages.length;
+		const lastId = lastMessage?._id;
+		if (count > prevMessageCount || (lastId && lastId !== prevLastMessageId)) {
+			prevMessageCount = count;
+			prevLastMessageId = lastId;
 			window.scrollTo(0, document.body.scrollHeight);
 		}
 	});
 
-	async function sendMessage(content: string) {
+	const clientWs = usePolling ? null : useConvexClient();
+	const pollingClient = usePolling ? usePollingClient() : null;
+	const createMessagePoll = usePolling ? usePollingMutation(api.messages.create) : null;
+	const registerDevicePoll = usePolling ? usePollingMutation(api.devicePairings.register) : null;
+	const heartbeatPoll = usePolling ? usePollingMutation(api.devicePairings.heartbeat) : null;
+	const addPhotoPoll = usePolling ? usePollingMutation(api.photoDrafts.addPhoto) : null;
+	const removePhotoPoll = usePolling ? usePollingMutation(api.photoDrafts.removePhoto) : null;
+	const createPairingPoll = usePolling ? usePollingMutation(api.pairingRequests.create) : null;
+	const acceptPairingPoll = usePolling ? usePollingMutation(api.pairingRequests.accept) : null;
+	const rejectPairingPoll = usePolling ? usePollingMutation(api.pairingRequests.reject) : null;
+	const unpairPoll = usePolling ? usePollingMutation(api.pairingRequests.unpair) : null;
+	const createRequestPoll = usePolling ? usePollingMutation(api.photoRequests.create) : null;
+	const markCaptureNowPoll = usePolling
+		? usePollingMutation(api.photoRequests.markCaptureNow)
+		: null;
+	const submitPhotoPoll = usePolling ? usePollingMutation(api.photoRequests.submitPhoto) : null;
+	const markRejectedPoll = usePolling ? usePollingMutation(api.photoRequests.markRejected) : null;
+	const acceptPhotoToDraftPoll = usePolling
+		? usePollingMutation(api.photoRequests.acceptPhotoToDraft)
+		: null;
+
+	$effect(() => {
+		if (!chatId || !deviceId) return;
+
+		if (usePolling && registerDevicePoll) {
+			registerDevicePoll({ chatId, deviceId, hasCamera });
+		} else if (clientWs) {
+			clientWs.mutation(api.devicePairings.register, {
+				chatId,
+				deviceId,
+				hasCamera
+			});
+		}
+
+		const interval = setInterval(() => {
+			if (usePolling && heartbeatPoll) {
+				heartbeatPoll({ chatId, deviceId });
+			} else if (clientWs) {
+				clientWs.mutation(api.devicePairings.heartbeat, { chatId, deviceId });
+			}
+		}, 10000);
+
+		return () => clearInterval(interval);
+	});
+
+	function sendMessage(content: string) {
 		const chat = chatData.data?.chat;
 		if (!chat) return;
 
-		await client.mutation(api.messages.create, {
+		const photos = draftPhotos;
-			chatId: chat._id,
+		const photoDraftIds = photos.length > 0 ? photos.map((p) => p._id) : undefined;
-			role: 'user',
+
-			content,
+		const messageContent =
-			source: 'web'
+			content || (photos.length > 0 ? 'Process images according to your task' : '');
-		});
+		if (!messageContent) return;
+
+		if (usePolling && createMessagePoll) {
+			createMessagePoll({
+				chatId: chat._id,
+				role: 'user',
+				content: messageContent,
+				source: 'web',
+				photoDraftIds
+			});
+		} else if (clientWs) {
+			clientWs.mutation(api.messages.create, {
+				chatId: chat._id,
+				role: 'user',
+				content: messageContent,
+				source: 'web',
+				photoDraftIds
+			});
+		}
 	}
 
-	async function summarize() {
+	function summarize() {
 		const chat = chatData.data?.chat;
 		if (!chat) return;
 
-		await client.mutation(api.messages.create, {
+		if (usePolling && createMessagePoll) {
-			chatId: chat._id,
+			createMessagePoll({
-			role: 'user',
+				chatId: chat._id,
-			content: '/summarize',
+				role: 'user',
-			source: 'web'
+				content: '/summarize',
-		});
+				source: 'web'
+			});
+		} else if (clientWs) {
+			clientWs.mutation(api.messages.create, {
+				chatId: chat._id,
+				role: 'user',
+				content: '/summarize',
+				source: 'web'
+			});
+		}
+	}
+
+	function handleTakePhoto() {
+		showCamera = true;
+	}
+
+	function handleCameraCapture(base64: string, mediaType: string) {
+		showCamera = false;
+		if (!chatId) return;
+		if (usePolling && addPhotoPoll) {
+			addPhotoPoll({ chatId, deviceId, photo: { base64, mediaType } });
+		} else if (clientWs) {
+			clientWs.mutation(api.photoDrafts.addPhoto, {
+				chatId,
+				deviceId,
+				photo: { base64, mediaType }
+			});
+		}
+	}
+
+	function handleCameraClose() {
+		showCamera = false;
+	}
+
+	function handlePair() {
+		if (!chatId) return;
+		if (usePolling && createPairingPoll) {
+			createPairingPoll({ chatId, fromDeviceId: deviceId });
+		} else if (clientWs) {
+			clientWs.mutation(api.pairingRequests.create, {
+				chatId,
+				fromDeviceId: deviceId
+			});
+		}
+	}
+
+	function handleAcceptPairing() {
+		const req = pendingPairing.data;
+		if (!req) return;
+		if (usePolling && acceptPairingPoll) {
+			acceptPairingPoll({ requestId: req._id, acceptingDeviceId: deviceId });
+		} else if (clientWs) {
+			clientWs.mutation(api.pairingRequests.accept, {
+				requestId: req._id,
+				acceptingDeviceId: deviceId
+			});
+		}
+	}
+
+	function handleRejectPairing() {
+		const req = pendingPairing.data;
+		if (!req) return;
+		if (usePolling && rejectPairingPoll) {
+			rejectPairingPoll({ requestId: req._id });
+		} else if (clientWs) {
+			clientWs.mutation(api.pairingRequests.reject, { requestId: req._id });
+		}
+	}
+
+	function handleUnpair() {
+		if (!chatId) return;
+		if (usePolling && unpairPoll) {
+			unpairPoll({ chatId, deviceId });
+		} else if (clientWs) {
+			clientWs.mutation(api.pairingRequests.unpair, {
+				chatId,
+				deviceId
+			});
+		}
+	}
+
+	function handleRequestPhoto() {
+		if (!chatId || !pairedDevice.data) return;
+		const captureDeviceId = pairedDevice.data.deviceId;
+
+		if (usePolling && createRequestPoll) {
+			createRequestPoll({ chatId, requesterId: deviceId, captureDeviceId }).then((id) => {
+				if (id) {
+					activeRequestId = id as Id<'photoRequests'>;
+					showWatchCountdown = true;
+				}
+			});
+		} else if (clientWs) {
+			clientWs
+				.mutation(api.photoRequests.create, {
+					chatId,
+					requesterId: deviceId,
+					captureDeviceId
+				})
+				.then((id) => {
+					activeRequestId = id;
+					showWatchCountdown = true;
+				});
+		}
+	}
+
+	function handleWatchCountdownComplete() {
+		showWatchCountdown = false;
+		if (!activeRequestId) return;
+		const reqId = activeRequestId;
+		activeRequestId = null;
+
+		if (usePolling && markCaptureNowPoll) {
+			markCaptureNowPoll({ requestId: reqId });
+		} else if (clientWs) {
+			clientWs.mutation(api.photoRequests.markCaptureNow, { requestId: reqId });
+		}
+	}
+
+	function handleWatchCountdownCancel() {
+		showWatchCountdown = false;
+		if (activeRequestId && markRejectedPoll) {
+			if (usePolling) {
+				markRejectedPoll({ requestId: activeRequestId });
+			} else if (clientWs) {
+				clientWs.mutation(api.photoRequests.markRejected, { requestId: activeRequestId });
+			}
+		}
+		activeRequestId = null;
+	}
+
+	function handleSilentCapture(base64: string, mediaType: string, thumbnailBase64: string) {
+		const req = captureNowRequest.data;
+		if (!req) return;
+
+		if (usePolling && submitPhotoPoll) {
+			submitPhotoPoll({
+				requestId: req._id,
+				photoBase64: base64,
+				photoMediaType: mediaType,
+				thumbnailBase64
+			});
+		} else if (clientWs) {
+			clientWs.mutation(api.photoRequests.submitPhoto, {
+				requestId: req._id,
+				photoBase64: base64,
+				photoMediaType: mediaType,
+				thumbnailBase64
+			});
+		}
+	}
+
+	function handlePreviewAccept() {
+		if (!previewPhoto || !chatId) return;
+
+		const reqId = previewPhoto.requestId;
+		previewPhoto = null;
+
+		if (usePolling && acceptPhotoToDraftPoll) {
+			acceptPhotoToDraftPoll({ requestId: reqId, chatId, deviceId });
+		} else if (clientWs) {
+			clientWs.mutation(api.photoRequests.acceptPhotoToDraft, {
+				requestId: reqId,
+				chatId,
+				deviceId
+			});
+		}
+	}
+
+	function handlePreviewReject() {
+		if (!previewPhoto) return;
+
+		const reqId = previewPhoto.requestId;
+		previewPhoto = null;
+
+		if (usePolling && markRejectedPoll) {
+			markRejectedPoll({ requestId: reqId });
+		} else if (clientWs) {
+			clientWs.mutation(api.photoRequests.markRejected, {
+				requestId: reqId
+			});
+		}
+	}
+
+	function handleRemoveDraftPhoto(index: number) {
+		if (!chatId) return;
+		if (usePolling && removePhotoPoll) {
+			removePhotoPoll({ chatId, deviceId, index });
+		} else if (clientWs) {
+			clientWs.mutation(api.photoDrafts.removePhoto, {
+				chatId,
+				deviceId,
+				index
+			});
+		}
 	}
 </script>
 
 <svelte:head>
 	<title>Chat</title>
-	<meta name="viewport" content="width=device-width, initial-scale=1" />
+	<meta
+		name="viewport"
+		content="width=device-width, initial-scale=1, maximum-scale=1, user-scalable=no, viewport-fit=cover"
+	/>
 </svelte:head>
 
-<div class="min-h-dvh bg-black text-white">
+<div class="min-h-dvh bg-black p-1.5 text-white">
 	{#if chatData.isLoading}
-		<div class="flex min-h-dvh items-center justify-center text-neutral-500">Loading...</div>
+		<div class="py-4 text-center text-xs text-neutral-500">Loading...</div>
 	{:else if chatData.error}
-		<div class="flex min-h-dvh items-center justify-center text-red-500">
+		<div class="py-4 text-center text-red-500">
-			Error: {chatData.error.toString()}
+			<div class="text-xs">Error</div>
+			<div class="text-[8px] break-all">{chatData.error.message}</div>
 		</div>
 	{:else if !chatData.data}
-		<div class="flex min-h-dvh items-center justify-center text-neutral-500">Chat not found</div>
+		<div class="py-4 text-center text-xs text-neutral-500">Not found</div>
 	{:else}
-		<div class="space-y-1.5 p-2">
+		<div class="space-y-1">
-			{#each messages as message (message._id)}
+			{#each messages as message, i (message._id)}
-				<ChatMessage
+				{#if i === messages.length - 1}
-					role={message.role}
+					<div bind:this={lastMessageElement}>
-					content={message.content}
+						<ChatMessage
-					isStreaming={message.isStreaming}
+							role={message.role}
-				/>
+							content={message.content}
+							isStreaming={message.isStreaming}
+						/>
+					</div>
+				{:else}
+					<ChatMessage
+						role={message.role}
+						content={message.content}
+						isStreaming={message.isStreaming}
+					/>
+				{/if}
 			{/each}
 		</div>
 
 		{#if followUpOptions.length > 0}
-			<div class="border-t border-neutral-800 px-2 py-1.5">
+			<div class="mt-2">
 				<FollowUpButtons options={followUpOptions} onselect={sendMessage} />
 			</div>
 		{/if}
 
-		<div class="border-t border-neutral-800 px-2 pt-1.5">
+		<div class="mt-3 space-y-2">
-			<button
+			<div class="flex gap-2">
-				onclick={summarize}
+				{#if hasCamera}
-				class="rounded bg-neutral-800 px-2 py-1 text-[10px] text-neutral-400"
+					<button
-			>
+						onclick={handleTakePhoto}
-				/summarize
+						class="flex-1 rounded bg-neutral-800 py-2 text-xs text-neutral-300"
-			</button>
+					>
-		</div>
+						+ photo
-
+					</button>
-		<div class="p-2 pt-1">
+				{/if}
-			<ChatInput onsubmit={sendMessage} />
+				{#if isPaired && pairedDevice.data?.hasCamera}
+					<button
+						onclick={handleRequestPhoto}
+						class="flex-1 rounded bg-neutral-800 py-2 text-xs text-neutral-300"
+					>
+						request
+					</button>
+				{/if}
+				{#if isPaired}
+					<button
+						onclick={handleUnpair}
+						class="flex-1 rounded bg-red-900/50 py-2 text-xs text-red-300"
+					>
+						unpair
+					</button>
+				{:else}
+					<button
+						onclick={handlePair}
+						class="flex-1 rounded bg-neutral-800 py-2 text-xs text-neutral-300"
+					>
+						pair
+					</button>
+				{/if}
+				<button
+					onclick={summarize}
+					class="flex-1 rounded bg-neutral-800 py-2 text-xs text-neutral-300"
+				>
+					/sum
+				</button>
+			</div>
+			{#if draftPhotos.length > 0}
+				<DraftBadge photos={draftPhotos} onremove={handleRemoveDraftPhoto} />
+			{/if}
+			<ChatInput onsubmit={sendMessage} allowEmpty={draftPhotos.length > 0} />
 		</div>
 	{/if}
+
+	{#if showScrollButton}
+		<button
+			onclick={scrollToLastMessage}
+			class="fixed right-3 bottom-12 z-50 flex h-8 w-8 animate-pulse items-center justify-center rounded-full bg-blue-600 text-white shadow-lg"
+		>
+			↓
+		</button>
+	{/if}
+
+	<StealthOverlay />
+
+	{#if showCamera}
+		<CameraCapture oncapture={handleCameraCapture} onclose={handleCameraClose} />
+	{/if}
+
+	{#if pendingPairing.data && !isPaired}
+		<div class="fixed inset-0 z-50 flex items-center justify-center bg-black/90" data-camera-ui>
+			<div class="rounded-lg bg-neutral-900 p-6 text-center">
+				<p class="mb-4 text-sm text-white">Accept pairing request?</p>
+				<div class="flex gap-3">
+					<button
+						onclick={handleAcceptPairing}
+						class="flex-1 rounded bg-blue-600 py-2 text-sm text-white"
+					>
+						Accept
+					</button>
+					<button
+						onclick={handleRejectPairing}
+						class="flex-1 rounded bg-neutral-700 py-2 text-sm text-white"
+					>
+						Reject
+					</button>
+				</div>
+			</div>
+		</div>
+	{/if}
+
+	{#if showWatchCountdown}
+		<WatchCountdown
+			oncomplete={handleWatchCountdownComplete}
+			oncancel={handleWatchCountdownCancel}
+		/>
+	{/if}
+
+	{#if previewPhoto}
+		<PhotoPreview
+			base64={previewPhoto.thumbnail}
+			mediaType={previewPhoto.mediaType}
+			onaccept={handlePreviewAccept}
+			onreject={handlePreviewReject}
+		/>
+	{/if}
+
+	{#if hasCamera && isPaired}
+		<SilentCapture
+			bind:this={silentCaptureRef}
+			oncapture={handleSilentCapture}
+			onunpair={handleUnpair}
+		/>
+	{/if}
 </div>

frontend/src/routes/[mnemonic]/+server.ts

@@ -0,0 +1,91 @@
+import { api } from '$lib/convex/_generated/api';
+import { error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+
+function detectImageType(bytes: Uint8Array): string | null {
+	if (bytes[0] === 0xff && bytes[1] === 0xd8 && bytes[2] === 0xff) {
+		return 'image/jpeg';
+	}
+	if (bytes[0] === 0x89 && bytes[1] === 0x50 && bytes[2] === 0x4e && bytes[3] === 0x47) {
+		return 'image/png';
+	}
+	if (bytes[0] === 0x47 && bytes[1] === 0x49 && bytes[2] === 0x46) {
+		return 'image/gif';
+	}
+	if (bytes[0] === 0x52 && bytes[1] === 0x49 && bytes[2] === 0x46 && bytes[3] === 0x46) {
+		return 'image/webp';
+	}
+	return null;
+}
+
+export const POST: RequestHandler = async ({ params, request, locals }) => {
+	const mnemonic = params.mnemonic;
+
+	const chatData = await locals.convex.query(api.chats.getByMnemonic, { mnemonic });
+	if (!chatData) {
+		throw error(404, 'Chat not found');
+	}
+
+	const rawContentType = request.headers.get('content-type') || '';
+	const caption = request.headers.get('x-caption') || '';
+
+	console.log('[POST /{mnemonic}] headers:', Object.fromEntries(request.headers.entries()));
+	console.log('[POST /{mnemonic}] content-type:', rawContentType);
+
+	let base64: string;
+	let mediaType: string;
+
+	if (rawContentType.includes('multipart/form-data')) {
+		const formData = await request.formData();
+		const keys = [...formData.keys()];
+		console.log('[POST /{mnemonic}] formData keys:', keys);
+
+		let file: File | null = null;
+		for (const key of ['file', 'image', 'photo', 'upload', 'attachment', ...keys]) {
+			const value = formData.get(key);
+			if (value instanceof File) {
+				file = value;
+				console.log('[POST /{mnemonic}] found file in field:', key);
+				break;
+			}
+		}
+
+		if (!file) {
+			throw error(400, `No file found in form data. Keys: ${keys.join(', ')}`);
+		}
+
+		const buffer = await file.arrayBuffer();
+		const bytes = new Uint8Array(buffer);
+		base64 = Buffer.from(buffer).toString('base64');
+		mediaType = detectImageType(bytes) || file.type || 'image/jpeg';
+
+		console.log('[POST /{mnemonic}] file:', file.name, file.type, 'size:', buffer.byteLength);
+	} else if (rawContentType.includes('application/x-www-form-urlencoded')) {
+		throw error(400, 'Use Form with File field, not URL-encoded form');
+	} else {
+		const buffer = await request.arrayBuffer();
+		const bytes = new Uint8Array(buffer);
+		base64 = Buffer.from(buffer).toString('base64');
+		mediaType = detectImageType(bytes) || rawContentType || 'image/jpeg';
+
+		console.log('[POST /{mnemonic}] raw bytes size:', buffer.byteLength);
+		console.log('[POST /{mnemonic}] detected type:', mediaType);
+	}
+
+	if (!base64 || base64.length === 0) {
+		throw error(400, 'Empty image data');
+	}
+
+	await locals.convex.mutation(api.messages.create, {
+		chatId: chatData._id,
+		role: 'user',
+		content: caption,
+		source: 'web',
+		imageBase64: base64,
+		imageMediaType: mediaType
+	});
+
+	return new Response(JSON.stringify({ ok: true, mediaType, size: base64.length }), {
+		headers: { 'Content-Type': 'application/json' }
+	});
+};

frontend/src/routes/layout.css

@@ -1,5 +1,17 @@
 @import 'tailwindcss';
 
+html,
+body {
+	background: black;
+	min-height: 100%;
+	overscroll-behavior: none;
+	-webkit-overflow-scrolling: touch;
+}
+
+* {
+	-webkit-tap-highlight-color: transparent;
+}
+
 .prose-mini h1,
 .prose-mini h2,
 .prose-mini h3,

frontend/vite.config.ts

@@ -3,5 +3,6 @@ import { sveltekit } from '@sveltejs/kit/vite';
 import { defineConfig } from 'vite';
 
 export default defineConfig({
-	plugins: [tailwindcss(), sveltekit()]
+	plugins: [tailwindcss(), sveltekit()],
+	server: { allowedHosts: ['reasonable-duncan-stations-parking.trycloudflare.com'] }
 });