Games / Too Many Requests
Too Many Requests
Everybody wants an AI now, and every question they ask is a request. You run the datacenter behind the company that answers them. You start in a garage and end on a huge datacenter site, with an IPO and a crash along the way.
- Status
- On pause
- Languages
- English, Russian, Ukrainian
- Developer
- Armatura

Trailer
Trailer coming soonAbout
Too Many Requests is a building game about the datacenter behind an AI company. Chat, images, code, agents and video come in as packets, and you wire up whatever answers them. Some of the chat is jailbreaks. Then the bubble pops, and you keep answering.
Everybody wants an AI now, and every question they ask is a request. You run the datacenter behind the company that answers them. You start in a garage and end on a huge datacenter site, with an IPO and a crash along the way.
Routing
Requests come in as small packets, and you can watch them move along the wires. There are five kinds, chat, image, code, agent and video, and each has its own color and mark. Every request carries the prompt a customer typed. Hover over a wire to read one.
You build what answers them, one piece and one wire at a time. Routers sort requests by type, queues hold extra traffic, caches answer repeated questions, and each GPU pool runs one AI model. When something jams, you can see where.
Products
Most people don't talk to a model. They use an app. Put an Assistant App, an Image Studio, a Cloud IDE or an Agent Platform between your pools and the Customers, and their answers pay more. Each one has a catch. Assistant App users ask follow-ups, and the Cloud IDE's tests come back as agent requests.
The hype moves
At every stage, people want something new. First everyone wants pictures, then a coding helper, then an agent to do their job, then video. Each new craze jams what you built for the last one.
Some of the chat is jailbreaks. They look like normal chat and pay nothing. If one gets answered, it's a scandal, and you earn half for a minute. A Safety Filter in front of your GPUs stops them.
Research and training
Your own GPUs do the research. A Research Lab turns data into tests, and a Journal writes a Paper for every right answer. Papers unlock a tree of 95 nodes in six lanes, from Batching up to Custom Silicon and Recursive Self-Improvement.
Scrape the web, scan books or make your own data, and a Training Run turns it into a new model. Scraped data gets worse as the web fills up with your own AI's answers, and too much made-up data breaks the model.
The crash
There is one crash, and you choose when it happens. Buy the IPO when you're ready. The bell rings, and then the bubble pops. Demand drops by half, and your only job is to keep answering. If you fail, you keep what you built and can try again. If you make it through, you can build AGI.
After AGI
AGI ends the campaign with a short epilogue written from your own run. Then the sandbox opens with everything unlocked. Three challenges, Lunch Rush, Chat Flood and Red Team, give you set traffic and a set budget, and score your build on cost, GPUs and speed.
Credits
I bought the music and the sound effects. Thank you to the people who made them.