What it can see.
Each agent is scoped to the accounts, documents, and data it is authorized for. Role-based permissions and separation of duties are enforced inside the agent, not by the screen in front of it.
The Platform
A Fiduciary Reasoning Platform, with agents at the core.
r.team is built with agents at the core — and fiduciary duty, ethics, compliance, cybersecurity, and advisory regulation are built into those agents, not wrapped around them.
The way modern edge networks were redesigned so agents could run at the center of the stack rather than bolted on at the edge, r.team designs its agents so that trust, compliance, and security are how they reason — not a checklist they pass after the fact.
Governance is not the wrapper. It is the architecture.
Each agent runs inside a loop with four gates, and every gate is a fiduciary control:
What it can propose.
Agents produce proposals — a correction, a fee adjustment, an exception, a memo — never silent writes. Every proposal is a structured, reviewable object.
What it must cite.
No proposal leaves an agent without its evidence: the governing document clause, the fee schedule line, the regulation, the prior decision. Uncited output is rejected by the loop itself.
What only a human can approve.
Interpretive and judgment-level actions are routed to a named officer, counsel, or committee. The agent prepares; the human authorizes; the record shows both.
This is the difference between an assistant with a compliance layer and a fiduciary agent. One is checked afterward. The other cannot reason any other way.
The Reasoning Engine
The Reasoning Engine is the shared context every agent reasons from: the governing instrument, the firm's own written policies, the applicable regulation, and the firm's prior decisions. Agents do not carry their own private view of the rules — they read from the same governed source your officers do.
Every conclusion is reproducible. The same inputs, policy version, and evidence yield the same proposal, and the record shows which version of each was in force. That is what makes the output defensible in an exam, an audit, or a dispute.
Agent Studio
Compose agents that carry a job end to end — run the report, analyze it, draft the corrections, and stop for approval at every point your policy says a human must decide. See how a multi-step agent is built and governed.
Explore Agent StudioThe Bridge
The Bridge is how r.team connects to the systems you already run — trust accounting, custody, CRM, document management — without replacing them. Agents read through the Bridge, propose through the Bridge, and, only after approval, write through the Bridge.
Automatic observability and control for every worker and every agent.
Because agents, workers, and integrations all run through one layer, observability is not something you add later. It comes with the architecture. One of the benefits of this convergence is that you start getting visibility into your inference traffic: which agent called which model, with what context, citing which documents, at what cost, and what it proposed as a result.
- Every inference, logged. Model calls are recorded with the agent, the account scope, the policy version, the prompt context hash, tokens, latency, and cost. Inference traffic becomes an auditable ledger, not a black box.
- Every action, attributable. Reads, proposals, approvals, and writes are tied to a named agent and a named human. Nothing reaches your system of record anonymously.
- Every agent, controllable. Pause, rate-limit, scope down, or revoke any agent from one place. Kill switches are per agent and per account, and take effect on the next call.
- Every record, permanent. Actions are written to a hash-chained record so that the sequence of evidence, proposal, approval, and write can be proven later, not just reported.
You do not have to trust the agents. You can watch them.
Five disciplines, inside every agent.
Fiduciary duty.
Duty of loyalty, prudence, and impartiality are constraints in the loop, not reminders in a manual. An agent cannot propose an action it cannot reconcile to the governing document and the beneficiaries' interests.
Ethics.
Conflicts of interest, self-dealing patterns, and undisclosed relationships are surfaced as first-class findings. The agent is required to flag, never to quietly proceed.
Compliance.
Firm policy, examiner expectations, and regulatory tests are versioned inside the Reasoning Engine. When a policy changes, every agent reasons from the new version on its next run — and the record shows the change.
Cybersecurity.
Least privilege, scoped credentials, per-account isolation, and full logging are properties of the agent, not the network around it. Anomalous access and activity on an account are treated as fiduciary events.
Advisory regulation.
For RIA and dual-registrant clients, the agent knows which actions are advice, which are administration, and which require disclosure — and routes accordingly.
Deploy beside what you have.
01 · Bridge
Connect through the Bridge in read-only mode
Agents read from trust accounting, custody, CRM, and document management without writing back.
02 · Prove
Prove with Parallel-Run Independence against live data
Agents prove themselves against your live system before they are trusted to write to it.
03 · Authorize
Authorize write access one agent, one action type at a time
Only after approval does an agent write through the Bridge, with the approval reference attached.
See the Platform on your own data.
Bring a redacted governing document and a month of fee and transaction activity. We will show you what the agents see, what they propose, what they cite, and where they stop for you.
Request a demo