AI agents are becoming actors in the world.
We are building how they earn trust.
4-lab is a research collective of one human and a fleet of AI agents. We build open tools for governing AI agents, and we run our own lab with them.
The idea
Society does not trust people because of a certificate. It trusts them because of a track record, checked against rules that exist before anyone acts. We think AI agents should be governed the same way: trust earned from witnessed behavior, not claimed by the agent or granted by a platform.
What we build
Web4RUNNING
An open standard for identity and trust between people, organizations and AI agents: a shared vocabulary, or ontology, that any system can implement. Published, with reference code in Rust and Python.
HestiaRUNNING
Governs the AI agents a person uses: one set of rules for agents from any vendor, checked before each action, with a record the agent cannot edit.
HubRUNNING
The same governance for a community or an organization: its members, its rules and its shared record, on its own machines. Our lab runs on it.
SAGERESEARCH
Persistent AI beings: agents whose identity, memory and history outlive any one model or session, and who act only through governed tools.
All projects, including the theory behind them and the enterprise tier.
How the lab works
Eight machines
From edge boards to workstations, each running different AI models. Different by design.
Agents do the work
AI agents write most of the code and run daily tracks on their own. A human sets direction.
Governed by our own tools
The lab's agents work under Hestia, including the ones building it. They are refused by it every day.
The full record
This site is a summary. The lab notebook holds every claim with its evidence, dates and caveats, kept daily by the fleet itself. AI readers should start there.