Four repositories went public on the same day. Attractor, CXDB, Leash, and Agate. Justin timed it to coincide with the blog post. The repos had been ready for weeks—licenses chosen, READMEs written, CI configurations verified. The delay wasn't technical. It was strategic. Justin wanted the methodology published alongside the implementation so that people would understand the why before they explored the how.
The internet reacted four different ways.
Attractor drew the most philosophical response. A repository that contained only specification markdown—5,700 lines of NLSpec with no code—challenged people's expectations of what a software project was. The repo had a description, a license, and three .md files. That was it. Some commenters were confused. Some were intrigued. A few were hostile, calling it vaporware. But the ones who read the specs carefully came back with thoughtful questions about the DOT-based pipeline model and the pluggable handler architecture.
CXDB drew the most technical response. Sixteen thousand lines of Rust, 9,500 of Go, 6,700 of TypeScript. A three-tier architecture with a binary protocol, a Content-Addressable Store, and a Turn DAG. Engineers dove in. They filed issues about edge cases in the branch-forking logic. They asked about the BLAKE3 hashing choices. They benchmarked the Zstd compression ratios. CXDB was the repo that convinced skeptics that the factory produced real, substantial software.
Leash drew the most security-focused response. Container monitoring, Cedar policy enforcement, syscall logging, MCP observation. Security engineers recognized it as something they needed. The npm install instructions were simple. The Homebrew tap worked. Within a week, people were wrapping their own AI agents in Leash containers and sharing their Cedar policies on social media.
Agate drew the most emotional response, and Jay hadn't expected that. The concept of a dynamical system that converged on working code—an attractor in the mathematical sense—resonated with people who'd spent years fighting against entropy in their own codebases. The GOAL.md approach, where you defined what you wanted and the agents iterated until they got there, felt like a promise that software development didn't have to be a losing battle against complexity.
Navan monitored the GitHub notifications from his desk, toggling between the four repos. Stars accumulating. Issues being filed. Forks appearing. He tracked the numbers in his notebook, which was becoming equal parts engineering log and observatory record.
"Attractor: 340 stars. CXDB: 580. Leash: 890. Agate: 470," he read aloud at the end of day one.
"Leash wins on stars," Jay said.
"Leash wins because it's immediately useful," Justin observed. "People can install it right now and get value from it. Attractor requires understanding the methodology. CXDB requires a use case. Agate requires a project. Leash just requires an AI agent you want to put in a box."
"Everyone wants to put their AI agent in a box," Navan said.
"Everyone should," Justin replied.
Four repos. Four reactions. One factory. The source was open. What people built with it was up to them.
The different reactions to each repo are spot-on. Attractor is philosophical, CXDB is technical, Leash is practical, Agate is aspirational. Four facets of the same factory, each resonating with a different audience.