B-A-M-N

X-Factor started with a pretty simple frustration: we have AI models that can write code, solve problems, and work through complicated tasks, but actually getting them to work together reliably is still a mess. I built four open-source projects to tackle different parts of that problem. Locus handles inference routing across providers, so you're not stuck depending on one model or service. Ocliv handles secure remote connectivity without requiring expensive infrastructure. Metis gives you a way to actually see, interact with, and manage agents running in real terminals. Jeverb tackles the verification side, trying to establish what an agent actually did instead of just trusting what it says it did. The bigger idea is connecting all of these pieces into something useful. An AI controller should be able to reach a machine, start an agent, give it real work, observe what happens, and get evidence of the result. We've demonstrated that workflow using authenticated connections, real terminal sessions, and a coding task where a worker fixed a failing test. This isn't meant to be another chatbot or another agent wrapper. It's infrastructure for making AI agents more practical to use, easier to manage, and harder to blindly trust when they get something wrong. It's still experimental, and there are parts that need more work, but the foundation is real and open source.