A new AI app called 'FORGE' quickly creates a team of AI agents to research any question, mimicking how human teams work. It's an impressive example of rapid development, moving from concept to a functional system in just three days.
Get ready to meet 'FORGE', a new AI application that can assemble a full research team for you in a snap! Imagine having a dedicated group of AI agents working together, just like human experts, to answer your toughest questions. What's truly amazing is that this clever app went from a simple idea to a fully functional system in just three days. This isn't just a technical marvel; it means you could soon have access to incredibly efficient research assistance right from your home server.
'FORGE' is designed to mimic a small human research team. One AI agent plans the work, another scours the web and reads sources, a third writes up the findings, and a final agent meticulously checks the write-up before it even reaches your eyes. You even get to watch the whole process unfold live, seeing which agent is active, what they're reading, what information they're passing along, and even how much it's costing you. This transparency is key to understanding and trusting the AI's work.
A big part of 'FORGE''s speedy development was a smart design choice from day one: every action, every step, is recorded as a log of events. This meant that the first version of the app, complete with its fancy dashboard showing workflows and live updates, was actually powered by a simulator – fake agents producing realistic events. When real AI agents finally replaced the simulator, the user interface didn't need to change at all because they emitted the exact same event logs. This foresight allowed the creator to build a beautiful and intuitive interface without waiting for the complex backend to be fully operational.
The real agents in 'FORGE' use a mix of powerful yet affordable technologies. Most agents rely on GLM-5.3 Flash models, processed through OpenRouter, keeping costs low at about $0.15 per million input tokens. The critical reviewing agent uses Claude, ensuring high-quality checks. For web searching, 'FORGE' leverages DuckDuckGo via a free Python package, and a small script efficiently pulls the main text from web pages for reading. This blend of accessible and effective tools makes 'FORGE' a powerful example of what's possible with thoughtful AI implementation.
'FORGE' is designed to mimic a small human research team. One AI agent plans the work, another scours the web and reads sources, a third writes up the findings, and a final agent meticulously checks the write-up before it even reaches your eyes. You even get to watch the whole process unfold live, seeing which agent is active, what they're reading, what information they're passing along, and even how much it's costing you. This transparency is key to understanding and trusting the AI's work.
A big part of 'FORGE''s speedy development was a smart design choice from day one: every action, every step, is recorded as a log of events. This meant that the first version of the app, complete with its fancy dashboard showing workflows and live updates, was actually powered by a simulator – fake agents producing realistic events. When real AI agents finally replaced the simulator, the user interface didn't need to change at all because they emitted the exact same event logs. This foresight allowed the creator to build a beautiful and intuitive interface without waiting for the complex backend to be fully operational.
The real agents in 'FORGE' use a mix of powerful yet affordable technologies. Most agents rely on GLM-5.3 Flash models, processed through OpenRouter, keeping costs low at about $0.15 per million input tokens. The critical reviewing agent uses Claude, ensuring high-quality checks. For web searching, 'FORGE' leverages DuckDuckGo via a free Python package, and a small script efficiently pulls the main text from web pages for reading. This blend of accessible and effective tools makes 'FORGE' a powerful example of what's possible with thoughtful AI implementation.
