1. Choosing a Deployment Platform#

The original architecture uses Python, Pillow, and Telegram Long Polling, so the priority is to evaluate the compatibility and migration cost of the existing code.

Recommendation: Continue using Render for existing projects, prioritizing improvements in data center location, image processing, and API latency.

If you plan to rewrite with TypeScript Webhook, then consider Cloudflare Workers.

2. Telegram Permissions and Security Settings#

To allow the bot to receive images within a group, you need to adjust Telegram’s privacy mode.

1. Open @BotFather and execute /setprivacy.
2. Select your Bot and set it to Disable.
3. If necessary, remove the Bot from the group and re-add it.
4. Use chat_id and user_id whitelists to restrict access.

1
# Only allow specified groups to use the bot
2
if update.effective_chat.id not in ALLOWED_CHAT_IDS:
3
    return

Additionally, non-image messages should be filtered first to avoid unnecessary Gemini API calls.

3. Render Deployment and UptimeRobot Anti-Sleep#

Render’s free Web Service may go to sleep after a period of inactivity, causing longer waits for the next Telegram operation.

You can add an HTTP health check endpoint to your Python Bot to allow external monitoring tools to access it periodically.

1
import os
2
import threading
3
from http.server import HTTPServer, BaseHTTPRequestHandler
4

5
class HealthHandler(BaseHTTPRequestHandler):
6
    def do_GET(self):
7
        self.send_response(200)
8
        self.end_headers()
9
        self.wfile.write(b"OK")
10

11
# Provide health checks in the background, not blocking the Telegram Bot
12
threading.Thread(
13
    target=lambda: HTTPServer(
14
        ("0.0.0.0", int(os.getenv("PORT", 8080))),
15
        HealthHandler
16
    ).serve_forever(),
17
    daemon=True
18
).start()

Next, create an HTTP(s) Monitor in UptimeRobot:

1. Log in to the UptimeRobot Dashboard.
2. Click Add New Monitor.
3. Select HTTP(s) for Monitor Type.
4. Enter your Render service URL, e.g., https://your-bot.onrender.com.
5. Set Monitoring Interval to every 5 minutes (depending on your plan).
6. Save settings and confirm that the health check returns HTTP 200. Effect:

UptimeRobot periodically sends HTTP GET requests, reducing the chance of Render going to sleep due to inactivity.

However, this method does not guarantee that free services will remain permanently active, nor does it prevent platform maintenance, deployment restarts, or resource limitations.

4. Gemini Recognition and Performance Optimization#

Telegram image accounting’s main time consumption usually focuses on:

• Image download.
• Pillow image processing.
• Gemini model inference.
• Google Sheets / Tricount accounting API write operations.

It is recommended to adopt the following optimization methods:

Image processing can be changed to:

1
import asyncio
2

3
# Move synchronous image processing to a thread
4
image_bytes = await asyncio.to_thread(
5
    optimize_receipt_image,
6
    raw_bytes
7
)

After Telegram receives the image, first display the processing status:

1
# First notify the user that recognition is in progress
2
status = await update.message.reply_text(
3
    "Receipt recognition in progress..."
4
)
5

6
# Execute Gemini recognition and accounting write operations
7
# result_text = ...
8

9
# Update the original message after completion
10
await status.edit_text(result_text)

This method does not directly shorten Gemini inference time but prevents users from mistakenly thinking the Bot is unresponsive.

1. Render Data Center Location#

If you were originally using the US East data center, consider recreating the service in Singapore.

• Add a new Render Web Service.
• Select Singapore for Region.
• Configure the original GitHub Repository and environment variables.
• After successful deployment, stop the old service. Note: Do not run two Long Polling instances simultaneously to avoid Telegram update reception conflicts.

2. Environment Variables and Restart#

Common Render environment variables:

• TELEGRAM_BOT_TOKEN
• GEMINI_API_KEY
• ALLOWED_CHAT_IDS

After modifying environment variables, confirm whether Render has completed deployment or restarted.

If a manual restart is needed, you can select Restart Service in the Render Dashboard.

3. Gemini Free Tier and Model Fallback#

Gemini model names, free tier quotas, and RPM / RPD limits may change over time.

Recommendations:

• Confirm currently available models in Google AI Studio.
• When encountering HTTP 429, decide whether to retry or switch models based on the error type.
• Do not hardcode historical free tier quotas or deprecated models in your code.