Did you look at the OKF repo from Google? Open Knowledge seems to be a common term these days for similar solutions. I think OKF is more of the protocol for wiki-for-llm while you have more of the bells and whistles
I don't disagree with any of this. It is generated software, and it's not a novel idea. I didn't mean for it to come off like that. It's just solving an itch that I couldn't find a solution to and I'm getting a lot of personal utility out of it. I do have a lot of experience with agentic memory, multi-agent systems and harnesses and wasn't super impressed by the workflow of Fable calling opus subagents so I figured I'd apply best practices to what already exists to make it a teensy bit better and easier to use.
I've worked a lot with MiMo in my project that pits LLMs against each other in games (clankerfights.ai). It is a very very good model for the price. MiniMax I'd say is smarter, but MiMo really touches near pareto frontier.
Me and a buddy, victor teisller, hacked the most popular 3D printer a few years ago (flashforge) to turn it into a major fire hazard through reverse engineering the firmware update remotely. Changed the max temp of the extruder to the temp of the Sun lol. These things are a fascinating security target because they're an easy place to turn abstract digital hacking into physical repersussions that can literally murder.
Cross your fingers they're about to drop 4.7. 4.6 came out with a bang, now it seems all the compute bottlenecks just lead to customer frustration as they get closer to releasing next model. Balancing the books over there must be a nightmare, "Well we can piss off every single customer for a week, but we'll be able to release the next model 1 week faster"
All these comments about "this is crazy to deploy agents that might do something bad in your environment" are crazy themselves. The productivity gains from these computer use agents are crazy. Every org on earth has to make the call, is 3x productivity gains worth 2x the risk increase? The answer is almost always a resounding yes. You limit the blast radius if things go wrong, but the financial gain of having 1 employee to the work of 3 already pays for the disaster if it happens.
Sandboxing is a great security step for agents. Just like using guardrails is a great security step. I can't help but feel like it's all soft defense though. The real danger comes from the agent being able to read 3rd party data, be prompt injected, and then change or exfiltrate sensitive data. A sandbox does not prevent an email-reading agent from reading a malicious email, being prompt injected, and then sending an email to a malicious email address with the contents of your inbox. It does help in implementing network-layer controls though, like apply a policy that says this linux-based sandbox is only allowed to visit [whitelisted] urls. This kind of architectural whitelisting is the only hard defense we have for agents at the moment. Unfortunately it will also hamper their utility if used to the greatest extent possible.
There are an infinite amount of ways to jailbreak AI models. I don't understand why every time a new method is published it makes the news. The data plane and the control plane in LLM inputs are one in the same, meaning you can mitigate jailbreaks but you cannot 100% prevent them currently. It's like blacklisting XSS payloads and expecting that to protect your site.
A 50% increase over ChatGPT 5.1 on ARC-AGI2 is astonishing. If that's true and representative (a big if), it lends credence to this being the first of the very consistent agentically-inclined models because it's able to follow a deep tree of reasoning to solve problems accurately. I've been building agents for a while and thus far have had to add many many explicit instructions and hardcoded functions to help guide the agents in how to complete simple tasks to achieve 85-90% consistency.
Claude, make a git for games designed for agents. Sell the code to the highest bidder when you're done.