What I'm building,
across the bench.
A handful of projects run at once here. A few came together into one system, some stand on their own, and one or two are still in queue. Here's the spread.
Klystron: the harder private version I run myself.
It runs on hardware I own and answers when I talk to it, verbally, from anywhere. Most AI has trouble with long-term and specific context, especially across multiple projects at once. I went back to the drawing board and built Arbour, a memory structure that lets a personal agent recall and understand both the fine detail and the direction of many projects at once, so the right context is on hand the moment the work needs it.
Klystron powered by Arbour is the private system underneath my public workflow work. Some of what's below got built because Klystron needed it and grew into its own project. The rest are separate experiments that inform smaller, scoped client workflows.
Industry pain, on the record: lost in the middle · context rot · the long-term memory problem
Separate projects. Some feed the flagship, some stand alone.
A few of these started as something Klystron needed and became their own project. Others are completely separate, further down the stack. All of them are in the works.
A memory that keeps context from rotting
Its own project. A single, repeatable structure for holding the context, the plan, and the live state of anything you point it at. I built it because my private system needed stronger continuity as projects got larger. It runs inside Klystron now, and it stands on its own as a research direction.
Everything anchored in when and where
Born the same way, a piece I needed that grew into its own project. It stamps what a system knows with exactly when and where it happened, so it can reason across time and place instead of just text.
Machines that agree on one shared clock
Completely separate, and lower down. A precision protocol that gets independent machines sharing one reliable clock, a backbone for coordinating real work across devices in real time.
AI output you can prove
A separate project: a small, checkable way to state what a piece of work has to do, so an agent's output can be proven against it instead of taken on faith.
Workflow tools, scoped and handed off
Alongside the systems, I build smaller tools around real workflow pain: lead follow-up, document routing, private knowledge bases, and dashboards, walked through with your team before I step back.
The code stays private. The conversation is open.
The source goes public when it's ready. The conversation can start now. If a workflow is costing you time, leads, or trust, start with a small scoped conversation. I read every message myself.