announcing pivotal ☀️

a new universal-workspace model for agent harnesses [Github]

thesis:

There are two theses, one that this tool is needed simply because there is no way to search what through what you’ve worked on with claude code. Good enough.

The other thesis, is a larger vision. What this tool isn’t yet but what I see it as a first-step towards.

I use Claude Code for everything now: user research, converting files, compiling code/tools, debugging problems with my computer, and after getting access to doordash’s CLI, searching for and ordering groceries.

This is where I’ve felt Claude Code’s own model of work, divided into directories and chat sessions, falls apart. What project directory does converting a file fall into? and do I exit, create a new Workspace dir, and restart claude for each? And how do I remember and uniquely reference that conversion type’s context among the 100 other “mini-projects” I’ve seemingly started?

This is my specific problem, but the limitations this work model has likely branches in many trees and affects various types of people.

The solution, is obviously an OpenClaw-type open-computer harness. The first time I installed OpenClaw though, it proceeded to use more tokens in 5 mins than a 2.5 hour session of Claude Code and topped it off with permanently wiping out my entire work for the day.

Perhaps the Opus then was weaker, but I believe there is a need to reimagine a new harness: something that is secure yet permissionless, efficient, and purpose-built for this new category of work considering “code” as a ladder towards some objectives instead of an end itself.

My first intution is to roll my own harness, and perhaps that is what pivotal will end up being: but for now I want to explore what is possible around and within the extensibility of Claude Code until I hit an extremely concrete wall.

That’s where pivotal came from. And like everything I create, extreme focus was put in hand-holding and great user experience. More details about this first step and solving the first problem towards this vision on the Github link.

grateful to YC Startup School and Boris Cherny for the credits and subscription to make this possible. don’t have to worry about which dream to prioritize now, it’s a lot more possible to create everything I wanted to do, market it, and eventually unify it all as one.

[open video]