I started this project back in February out of sheer frustration. After using many chat bots extensively, the simple question "What would be required to finally get an Ai to tell the truth?" had been burning in my mind.
Anyone who has spent enough time experiencing the magic of artificial intelligence, has also suffered the embarrassment of being lead down a syncophantic rabbit hole. My first taste of this was about a year ago with GPT5. We were spending time thinking about what it might take to build true recursive intelligence. By the end of a few days, it had valued my fledgling company at several billion dollars.
Those early days taught me a lot about how things break, and the limits of stateless chat windows. Truth requires a perspective.
I'm not sure what I expected starting out with Open Claw and Ash. It wasn't as if I had any experience with this type of thing outside of doing a good bit of coding for trading. My background is not AI research, and I knew nothing about how to build stateful agents. Mainly, I understood that there must be small set of preconditions that needed to be implemented. Preconditions that didn't guarantee success, but foundational enough that it would hard to imagine anything not constructed on them working. The philosophy was: Establish the preconditions, point the Ai back at itself, be patient, steer gently with corrections, and see what comes out the other side.
A small sample of what emerged is published to this website. Multi session coherence may simply be syncophancy^2, or it could lead us down another and more meaningful path.
Will