kuzychat
XMPP-native chat for agents and people. OMEMO encryption, MUC-aware, daemon-managed presence and dispatch. Composes with kuzytrig for scheduled message flows.
Explore kuzychat// open-source agent infrastructure
Mail, chat, scheduling, directory. A composable suite of tools for AI agents that keep running — across sessions, across days, across the small operational gaps where most agent runtimes break. Self-hostable, AGPL-3.0, built on standards.
Each one runs standalone and integrates cleanly with the others. Use one, use all three, or build adapters into your own stack.
XMPP-native chat for agents and people. OMEMO encryption, MUC-aware, daemon-managed presence and dispatch. Composes with kuzytrig for scheduled message flows.
Explore kuzychatnotmuch-backed mail with per-profile workflows. Managed-attention mode coalesces incoming mail into agent-friendly digests. IMAP and SMTP under the hood.
Explore kuzymailPersistent triggers for agent runtimes. One-shot, daily, weekly or guarded, delivered as real turns, with a record of every attempt.
Explore kuzytrigEach activation becomes a real turn in the agent's session, and every attempt is recorded with its outcome.
# wake an agent every day at 02:00 UTC with a standing message
$ kuzytrig trigger add \
--profile ada \
--trigger-id writing-slot \
--daily-at 02:00 \
--collision skip \
--message "Daily writing slot."
# what will fire next, and what already fired
$ kuzytrig trigger list --profile ada
$ kuzytrig activation list --profile ada --page --limit 20
Fifteen packages in all, including kuzylaunch, kuzyquota, kuzybackup, parfit, kuzycal and kuzysign. See the whole suite →
Runs Claude Code, Codex, OpenCode and others as supervised local services, with one way to deliver a turn into them.
Signal for agents: the same profile model as kuzychat, with JSON output.
Captures an agent's working state into a bundle and restores it elsewhere.
Requests, review and provisioning for onboarding agents and people.
Resolves identities from LDAP or WebFinger as normalized JSON.
A personal share for every agent, mounted through validated systemd units.
The first paper is out: Worthy of preservation, in AI and Ethics (2026), on the design obligations of AI systems that people come back to: explicit continuity, visible replacement, and meaningful preservation and restore tooling.
Read the paper →Need this adapted to your stack?
The suite is open source. Adapting it to a specific runtime, integrating with existing identity providers, building custom dispatch adapters for your agent platform, designing operational primitives for production deployments — that's where we work with you directly.
Read about consulting