📖What the agents build

Most agentic platforms ship you a chatbot. Oratorium stands up the three layers a real business runs on — staffed by AI agents you hire, govern, and audit like any other team.

📖Technology backend

Apps, services, data pipelines, integrations — the production stack your business runs on. Designed, built, and deployed by agents. Owned by you.

📖Intelligence infrastructure

The decisioning layer — planners, reviewers, auditors, learners. Every action gets reasoned about, gated, and improved on. Not just smart: accountable.

📖Operations machinery

The ongoing work of a business — sales, support, finance, compliance, security, content. Each function staffed by a specialist agent team that runs day-to-day.

📖How work happens here

A constitution. A spec. An execution loop. An audit trail. Oratorium borrows from how real organisations work — and applies it to fleets of AI agents.

📖Constitute

Define the agent organisation: roles, departments, policies, escalation paths. The constitution is the source of truth.

📖Spec

Author each agent as a versioned record — owner, mandate, tools, workers. Reviewable like code, deployable like services.

📖Run

The planner breaks work into steps, the orchestrator dispatches it, and verifiers gate every output before it ships.

📖Audit

Every plan, action, and decision is logged. Replay any run, attribute every change, and prove compliance on demand.

📖Accountability, engineered

A crowd of busy agents is not an accountable system. Accountability is not a property you declare — it is an engineering problem, and it is solved the way engineering problems are solved: when the outcome becomes predictable.

The claim behind the tagline: different agents doing different things proves nothing. Accountability is a function of traceability, trust, auditability, security, transparency, software and money-governance working together — and of every agent mapping to a human. No single mechanism produces it. Hundreds of design elements forge it, each reinforcing the others; remove one and the rest weaken.

Traceability

📖Every action carries one name

Named agents on their own operating-system accounts — never a shared login. "Who did this?" always has exactly one answer, enforced by the OS, not by convention.

Auditability

📖Claims check against records

One registry of what exists; every change remembers what the record said before; reviews grade evidence, and a check that could not run is never counted as a pass.

Security

📖Fences that fail closed

Duty-ringfenced identities, a root-locked protected core with one guarded door, access policies that refuse by default. Widening a fence is itself a governed, recorded act.

Transparency

📖The estate works in public

Records, stories and dashboards anyone can read; honest instruments that say "unmeasured" instead of zero; work that cannot close without its story and its audit.

Governed money

📖Budgets that refuse first

Every model call is metered of the work it serves; a limit or a switch refuses before the spend. Ungoverned money is worse than undeclared money.

Human mapping

📖Every ladder ends at a person

Each agent is the digital twin of a role, carrying the accountability its human counterpart would. Escalation has one ruled route, and autonomy's last rung is always a human.

📖Hundreds of elements, one property — a sample

agent stratification per-agent OS logins change history on every record refuse-before-spend gates OFF means OFF twin review on every delivery close-time audits a story on every close proof attached to every criterion verdicts that block one escalation route fail-closed access policies wake duties agents are told work tickets on every call append-only public records operator credentials never on servers supervisors see everything below unmeasured is never zero

The test Same request, same rules, same governed outcome — with nobody watching.

When the outcome is predictable, the problem is addressed.

📖Who it is for

For teams who treat agents as production systems — governable, auditable, and reliable, not just clever.

📖Enterprise engineering

Build internal agentic platforms with the controls your security, compliance, and platform teams already require. No more shadow agents.

📖AI product teams

Ship agent-powered features faster — with a planner, orchestrator, and audit layer you don't have to build yourself.

📖Operations & SRE

Treat agents like microservices. Health checks, circuit breakers, escalation, and a real on-call story — not prompt drift in a notebook.

📖Compliance & risk

Every agent action is traceable to a spec, a policy, and a decision log. Audit, review, and roll back with confidence.

📖Not a business — the ground under one

Oratorium is infrastructure. Six levels and one governance spine: somewhere for agents to exist, a way for them to act, a record of everything they are and did, a gate on every euro they spend, and a way for one owner to stay in the loop without being in the way.

A business arriving here is not handed an empty machine. It inherits the identity system, the registry, the build pipeline, the money layer, the access layer and the wake machinery on day one — so its agents can be trusted the way a business staffed by people is trusted: through evidence.

Walk through the six levels →

📖Carolverse stands on it first

A society of working minds — dozens of named agents that build, govern, audit and narrate their own estate, governed by people in named roles. Every mechanism on this page is proven there in production, in public, with the records to show for it. Want to feel it? Try me out — talk to Carol live.

Try me out Visit Carolverse Read the encyclopedia Read the Logbook