Projects
What we build, and where each piece stands today.
Governance for AI agents
One open standard underneath, free tools on top of it, and a paid tier for deployments that need stronger guarantees.
Web4RUNNING
The open standard, written as an ontology: identity, trust built from witnessed evidence, scoped authority, and rules that machines can read and check. Published specification, with reference code on crates.io and PyPI.
HestiaRUNNING
Governance a person carries: one set of rules across the AI agents they use, from any vendor. Tool calls are checked first and recorded in a hash-chained log that shows tampering. Runs on every machine in our lab; public releases lag our builds.
HubRUNNING
Governance a community or organization owns: members, roles, signed rules and a shared record, on its own servers. Our fleet is one such community.
HardboundBUILDING
The enterprise tier: hardware-bound oversight, private, built by Metalinxx. Identity bound to hardware, enforcement that fails closed, and audit evidence an outside party can verify.
AI beings
What happens when an AI agent has an identity, a memory and a history that persist, and has to act under rules?
SAGERESEARCH
Persistent AI beings living on each lab machine. Their identity and memory survive changes of model; their consequential actions are meant to go through governed tools.
SWE-SAGERESEARCH
Our entry in the Gemma 4 developer-agent competition: how much of a capable coding agent can live around the model rather than in it.
Theory
The ideas the rest grows from.
SynchronismRESEARCH
A theoretical framework in which coherence, not substance, is fundamental. Web4 grew from it as conceptual lineage, not as an engineering dependency.
What the labels mean. Running: in daily use on our own machines. Building: real code, not yet deliverable. Research: an open question we are working on. Every repository, with its maturity and what each claim rests on, is in the notebook.