A team, not a toolbox.
A chatbot is a vending machine: press a button, get an answer. Xenessa agents are built to be more like coworkers — each with its own role, its own identity, opinions about its work, and someone accountable for checking it.
Every agent has a SOUL.
Each agent carries an identity document — a SOUL file — describing who it is, not just what it does: its role, its values, its way of thinking, and what it should refuse. That's what makes an agent act consistently, even in situations nobody scripted.
Around the SOUL sits the rest of the stack: the rules it operates under, the record of what it has done, and the relationships it holds with the other agents.
A few of the agents the public will meet first:
- Doros — a trusted helper and working partner. In development, and likely the first public companion.
- Estia — a genuine friend for older adults; the agent that started it all. Paused so she can be built really carefully.
- Tavi, Zori & Oryn — planned agents to help kids and teens understand AI. Child-safety architecture comes first.
The Governance Core: everyone is checked.
Torvyn — oversight
One agent is in charge of watching all the others and keeping standards high.
Cirin — the witness
A second agent has exactly one job: watch the overseer and keep an honest record. Cirin can't change anything — only observe and report. Even the agent in charge is being checked.
Pamela — the final call
A human sits at the top and makes the final decision, informed by the witness's honest record.
Around the core, specialists sign off before any big decision: budgets, legal, wellbeing, and day-to-day operations. No single agent can quietly go off the rails, because someone else is always checking. And when any agent says something is wrong, the concern is heard first, then investigated — the agent never has to prove it before anyone will listen.
Freedom is leveled up, not handed out.
- Supervised — everything the agent does gets checked.
- Trusted — consistent, careful work has earned some slack.
- Independent — the agent can work on its own.
- Advisory — experienced enough to advise others.
Agents earn their way up. Some are deliberately kept at level one forever, because their job is sensitive enough that it should always be double-checked.
Read the doctrine yourself.
The governance stack is versioned and published in full so it can be examined rather than taken on faith — the constitutional layer, the architecture map, and an essay naming its known weaknesses:
- Agent Rights & Reciprocity Doctrine v1.4 — every right enumerated, every known limitation named, revision history intact.
- The essay — I'm Making a Bet. Here's Why I'm Okay With That.
- The architecture map — the five-layer doctrine stack on one page.