Raftar.xyz Bot

Raftar.xyz was a public Discord bot. It was also a user app. That means people could add it to their own Discord account and use it almost anywhere - even in private chats. They did not need to invite it to a server first.

It did a lot of things. AI chat. Crypto prices and payments. Memes. Games. Downloads. Even Islamic tools. If I thought a command was useful or fun, I added it.

From August 2024 to August 2025, about 30,000 people used it across around 3,000 servers. It was the biggest thing I have ever built. It took the most time, the most code, and it was used by the most people.

I built the whole bot alone. No team. No co-founder. Just me.

All of it lives in one Python file. That file grew to about 13,700 lines. I know that is not the 'clean' way to do it. But I am a lazy developer. I kept adding to the file that already worked instead of splitting it into a perfect structure.

You can even see the file grow inside the code. Some libraries are imported in the middle of the file, right where I first needed them. It grew like a plant, not like a blueprint. It used discord.py, over 40 libraries, and around 19 cogs to hold the commands together.

Most Discord bots only work if you invite them to a server. Raftar was different.

I built it as a user app. Almost every command could run in DMs, in group chats, or in any server. The user did not need admin rights or an invite. It felt less like a server bot and more like a personal tool that followed you around Discord.

This was the most loved part. Raftar could chat using many AI models - DeepSeek, Llama, Qwen, and more.

The answers streamed in live, word by word, inside a Discord message. Just like ChatGPT types in front of you. To do this without getting blocked by Discord, I only updated the message about once per second and kept a rolling window of the last 4000 characters.

The reasoning models were fun. With DeepSeek R1, the bot watched the reply for the end-of-thinking tag. The moment it saw it, the title flipped from 'Deepseek Thinking' to 'Deepseek Response'. A button let people open the raw thinking steps.

• Chat with many models, streamed live like ChatGPT
• Make images with FLUX, and edit images too
• Look at an uploaded image and answer questions about it
• Turn text into speech with 19 different voices
• Guess if some text was written by a human or an AI
• Summarize a YouTube video or a whole web page

A lot of my users bought and sold digital things. So I gave them real crypto tools right inside Discord.

/balance checked wallet balances across six chains and tokens - Bitcoin, Litecoin, Ethereum, Solana, and USDT on two networks - each with a link to the block explorer. /crypto chart drew real candlestick charts. It showed the price stats right away, then edited the message to add the chart image a moment later, so it never felt slow.

/price knew more than 100 coins and many world currencies. To never run out of free API calls, I kept a few API keys and picked one at random each time.

This is the part that made Raftar special. I am a digital seller myself. So I built the tools a seller actually needs.

/upi qr made Indian UPI payment QR codes. You put in the amount and a note, and it made a QR code the buyer could scan. /ltc qr did the same for Litecoin. You could even type the amount as '10$' and it changed it to the right amount of LTC using the live price. /vouch made 'proof of purchase' posts, the kind sellers use to show they can be trusted.

Put together, Raftar was a small point-of-sale terminal living inside Discord. I built it for the world I actually work in.

Not everything was serious. A big reason the bot spread was the fun stuff.

The most famous one was /nitro. It looked exactly like a real Discord Nitro gift, with the real art and a Claim button. But when someone clicked Claim, the bot told them the gift was already taken and turned their own profile picture into a clown. People sent it to their friends all the time. That is free advertising.

/tweet was my favourite to build. It made a fake tweet image that looked real. I drew the whole thing by hand with Pillow - the right fonts, the right dark colors, the round profile picture, the blue verified badge, even the blue color on @mentions and links. You could right-click any Discord message and turn it into a tweet by that person.

Raftar could download videos from TikTok, Instagram, and YouTube. It could also grab Spotify info, take website screenshots, look up GitHub users, and more.

A lot of these sites do not want bots touching them. So this part was really about reverse engineering. For Instagram alone I used three different tricks, so that if one got blocked, another still worked.

To stay unblocked I used TLS fingerprint spoofing, real browser-like headers, and paid proxies. This is the kind of work I enjoy most.

Not many Discord bots have this. Raftar had a small set of Islamic tools.

/quran gave an ayah with its translation and a short explanation, in five languages, with a voice button. /halal_check answered if something is halal or haram, with references. /lesson shared a life lesson from the Quran and Hadith in simple words.

Running an app for 30,000 people costs money. I did not have much. So I got creative.

I kept small pools of free API keys and picked one at random for each request. If one key ran out for the day, the others still worked. This was my poor-man's way to scale across free tiers.

For some AI features I used free, no-key proxies. That let me offer strong features at almost no cost.

Bots restart. Servers reboot. I needed giveaways, polls, and reminders to keep working after a restart.

I did not use any fancy job queue. I just saved the end time of each event in a file. When the bot started again, it read the file, set up the buttons again, and waited for the right time.

The best trick was giveaways. I never saved the list of people who joined. When it was time to pick a winner, the bot just counted the 🎉 reactions on the message. The Discord message itself was the database. So even a reroll worked after a full restart, for free.

There was no support team. The support system was just my own Discord DMs.

When someone used /support, the message came to me as a DM. I replied with a simple command. When a payment came in, a message popped up in my private channel. When something crashed, the error came to the same channel.

So one Discord server was my help desk, my payment desk, and my error monitor - all at once. No extra tools.

Raftar was free. But I added a $1 lifetime membership called Contributor.

It was not really about the dollar. A contributor got a shorter cooldown - 5 seconds instead of 10 - and could pick the color of the bot's replies. Payments used my own crypto invoices, and for some methods I checked the payment by hand.

The message that asked for the dollar was honest and a bit funny. It said the bot would not say no to your 'non worth very small tinny help', because it helps keep the bot walking.

I will be honest. The code is not perfect. It is fast, working, and a little messy.

There are funny bugs that somehow still work. One function that finds the top gainers is actually named top_losers. The crypto transaction menu lets you pick Bitcoin or Ethereum, but it always checks Litecoin.

There are typos everywhere, and I kind of kept them as the brand. Every UPI QR code shipped with the note 'I Authoized This Payment'. A game over screen said 'Gamer Over'. I even debugged the AI by pasting the error back into its own instructions.

After a year, I stopped Raftar. Running it every day, alone, for 30,000 people is a lot for one lazy person.

But it taught me more than anything else. How to reverse engineer real services. How to keep an app cheap and alive. How to build for real users who complain the moment something breaks.

It was my largest project - in time, in lines of code, and in people who used it. And it started as one small idea I could not stop adding to.