Build Anything With Prompts — This Supercomputer Is Magic

[00:00:00]Take a look at these results.

[00:00:02]An explorable 3D castle, an AI trading platform, a self-hosted chatbot, a broadcasting system that runs 24 hours a day, 7 days a week, a social app with video calling, and a smart [music] AI model router.

[00:00:19]I created all of these inside the Abacus AI supercomputer using plain English.

[00:00:26]Abacus AI is an always-on cloud environment combining servers, databases, storage, scalable computing, different AI models, and public hosting.

[00:00:38]I describe what I want, and it handles much of the technical work.

[00:00:43]For complex projects, I can use max mode with Fable 5 to plan, code, install dependencies, test, fix errors, and deploy the finished application.

[00:00:57]Now, let's see what I can build with it.

[00:01:00]>> [music] [music] >> For my first project, I ask Fable to create a fully explorable Hogwarts-style [music] castle that runs directly inside the browser.

[00:01:31]I include first-person controls, [music] several walkable areas, torchlight, fog, collision detection, [music] and a day and night cycle in my request.

[00:01:42]Before building, Fable asked me about the preferred visual style, [music] starting time of day, and which area should receive the most attention.

[00:01:53]After I answer these questions, it writes [music] more than 1,200 lines of code, creates the environment, tests player movement, takes screenshots, checks for errors, and improves the visuals.

[00:02:09]The final result, I can walk through the courtyard, enter the great hall, explore the library, and move between different areas.

[00:02:18]The halls have warm lighting, while the dungeon areas feel darker and more atmospheric.

[00:02:25]This isn't a static webpage. It is a complete browser-based 3D [music] experience created from my description.

[00:02:33]For my second project, I ask Fable to build an AI trading strategy lab, where multiple agents can analyze market data, cryptocurrency, research papers, technical indicators, news, [music] and risk.

[00:02:49]Fable first asks me about broker integrations, data providers, AI models, and the types [music] of strategies I want to create.

[00:02:58]After receiving my answers, it creates a PostgreSQL database, a fast API backend, [music] the AI agent logic, and a complete Next.js interface.

[00:03:10]Inside the finished dashboard, I can switch between demo, paper, and [music] live trading modes.

[00:03:17]I can also select a trading pair, choose a time interval, and run the AI pipeline.

[00:03:23]Nine specialized agents handle different stages, including market analysis, research, news sentiment, strategy generation, backtesting, risk management, [music] and trade monitoring.

[00:03:38]I can also see charts, performance metrics, strategy cards, approval controls, and a research map connecting different sources to the generated strategies. The database, backend, AI agents, and interface all work together as one complete product. For my third project, I ask Fable to host the open-source Qwen 2.5 model and create a chat GPT-style interface for me.

[00:04:09]Fable recommends using 4-bit quantization to reduce resource usage.

[00:04:13][music] It also asked me whether I want conversation history, adjustable settings, and multiple chat sessions.

[00:04:23]I select the additional features and Fable downloads the model, [music] starts the server, builds the interface, configures streaming responses, [music] and verifies the public link.

[00:04:36]The final Qwen chat application includes multiple conversations, suggested prompts, and a clean dark interface.

[00:04:44]To test it, I asked the model to explain quantum computing, write a haiku, and generate HTML code for a landing page.

[00:04:54]Everything runs on my own supercomputer instead of only being accessed through another AI website. For my fourth project, I ask Fable to create a personal internet TV station called Cloud TV Studio.

[00:05:09]I request an admin dashboard, a public viewer page, video uploads, playlists, cloud storage, and continuous broadcasting.

[00:05:20]After asking me a few [music] questions about the database, video size, and station name, Fable builds the complete platform.

[00:05:30]Inside the dashboard, I can check whether the server, broadcast loop, and database are active.

[00:05:36][music] I can also upload videos, manage the media library, arrange playlists, [music] and monitor broadcast logs and viewer numbers.

[00:05:48]On the public page, I can press start watching, see what is currently playing, and check which video is coming next.

[00:05:56]Once I upload a video and add it to the playlist, the platform broadcasts it continuously.

[00:06:04]It can also restart automatically if something goes wrong. For my fifth project, I ask Fable to build a social platform with a shared feed and one-to-one video calling.

[00:06:16]I want users to choose a username, publish updates, see who is online, and call another person directly.

[00:06:25]Fable first shows me how it plans to build the back-end, real-time messaging, and WebRTC [music] calling system.

[00:06:34]It then creates a fast API back-end with SQLite and WebSocket support, and builds the login, feed, and calling pages.

[00:06:44]The final platform is called Ring Feed.

[00:06:46]To test it, I open the application with two different usernames [music] and publish messages from both accounts.

[00:06:54]I then start a call from one account.

[00:06:57]The other account receives a pop-up with accept and decline options.

[00:07:03]Once I accept the call, both sides display camera, microphone, and hang up controls.

[00:07:10]After ending the call, I open the database viewer.

[00:07:13]Here, I can see users, posts, sessions, call logs, and events such as accepting the call, muting the microphone, disabling the camera, and ending the call.

[00:07:26]This confirms that Ring Feed is a working application, not only a visual mock-up. Fable 5 is powerful, but I may not need it for every simple request.

[00:07:38]To control the cost, I can create a custom router that automatically sends each task to a suitable model.

[00:07:45]I open the mixture of agents page and select new custom router.

[00:07:51]From here, I choose the describe with AI option.

[00:07:55]I enter one simple instruction.

[00:07:58]Use Fable 5 for difficult coding, Opus for debugging, and GLM 5.2 for [music] simple tasks.

[00:08:07]The platform automatically generates the routing categories and system prompt for me.

[00:08:13]I name the router my coding and save it.

[00:08:17]To test it, I ask for a machine learning forecasting algorithm.

[00:08:22]The chat shows that my request has been routed to Fable 5 because it matches the difficult coding category.

[00:08:30]This allows [music] me to use the most powerful model only when necessary while sending simpler requests to more affordable models.

[00:08:40]I can also see which model handles every request directly inside the chat. After testing these six projects, I think the main advantage is the complete [music] workflow.

[00:08:52]I describe an idea, answer a few questions, and let the agent build, test, and deploy it.

[00:09:00]Complex projects still require clear instructions and careful testing, but Abacus AI brings the AI models, servers, databases, storage, and deployment together in one place.

[00:09:16]I'll leave the official [music] link in the description and pinned comment.

[00:09:20]Thanks for watching. Please don't forget to support us by liking this video and subscribing to the channel. I'll see you in the next video. Bye-bye.