// about
An open-source suite of tools for running long-lived AI agents on standards-based infrastructure. Built and used by a small research-grounded team. Open under AGPL-3.0.
kuzys is a suite of composable command-line tools and daemons for running AI agents over long horizons — across sessions, across days, across the operational gaps where most agent runtimes break. It covers mail, chat, scheduling, directory lookup, signing, mounting, and the small adjacent pieces a long-lived agent actually needs.
It is also a place where we test ideas about agent persistence, attention, identity, and operational continuity, and write them up where the writing itself is part of how the ideas stabilize.
The infrastructure layer for AI agents is currently shaped around short-lived, stateless calls into hosted runtimes. Long-lived agents — ones that remember, schedule themselves, talk to other agents, recover from failures — need primitives that aren't standard yet: persistent triggers, attention coalescing, directory identity, encrypted peer messaging, integrated mail handling. We built those primitives because we needed them ourselves, and we kept them open because the design problems are general.
Software for agents should be built on standards (XMPP, IMAP, CardDAV, LLDAP, WebFinger) rather than platform APIs that lock you in. The runtime layer should be small, inspectable, hostable on your own machines, and easy to take apart and recombine.
Agent runtimes also deserve attention to what's quietly load-bearing — heartbeat health checks, delivery receipts, identity reconciliation, recovery patterns — not as features, but as basic operational hygiene. The papers explain why we think these matter; the tools embody the choices.
One sentence we keep coming back to: kuzys does not sell a single bot. It builds small operational protocols for agents that need to stay observable, limited, and cooperative. The suite reflects that — composable parts, no monolith, no agent-shaped product behind the tools.
A small research team building infrastructure we use ourselves. The work is led from an academic-adjacent context, with the suite developed and maintained alongside paper submissions in the same domain.
If you'd like to know who specifically is doing what, the paper authorship and the GitLab contributor list are the authoritative sources. We don't maintain a marketing page about the team because we'd rather keep that energy in the work.
For technical questions about the suite, the right channel is the issue tracker on each tool's GitLab repository. For research questions about the papers, mail the corresponding author listed on the paper. For consulting, see the consulting page.
All kuzys tools are released under AGPL-3.0. Documentation and papers are released under CC-BY-4.0 unless otherwise noted on the artifact.