Questions, answered straight.

A named specialist that owns one control function end to end. Vera reviews loan files against your credit policy. Ira works the reconciliation exceptions your matching engine can't close. Each one reads, checks, and drafts; a named human approves. You hire the role, not the tooling.

A copilot helps your team go faster at the work they already do. An AI employee takes over a queue: every file, every exception, every alert, oldest first. Copilots assist a person. Our employees own a workload and report to one.

RPA replays clicks on structured screens and breaks the day a form changes. An AI employee reads unstructured documents, cross-checks them against your policy, and writes up findings with citations. RPA automates keystrokes. This does the review work between the keystrokes.

A builder hands you a canvas and says 'now go build.' That is a new engineering project, not less work. Our employees ship as finished specialists on a production harness: scope, boundary, approval gate, and audit trail already built. You evaluate their output, not their architecture.

No. Not as a setting you toggle, as an architecture. An AI employee never touches a system of record. It reads, checks, and drafts. Every finding, memo, and closure is a draft until a named human approves it, and the approval is logged with the evidence.

The hard boundary around what the employee may do. Inside it: read documents, run checks, draft findings. Outside it: posting entries, closing alerts, releasing batches, moving money. Everything outside the envelope requires a named human's approval, and that requirement does not relax over time.

A complete trail per item: the documents read, the policy clauses checked, the finding with page-level citations, the reasoning, and the named human who approved it with a timestamp. It is more trail than most manual processes produce today. The trail is exportable; you do not need us in the room to show it.

The reviewer rejects the finding, and the rejection is logged like everything else. Because a human approves every item, a wrong draft costs minutes, not a regulatory incident. We measure and report the disagreement rate; you see it, not just us.

Two grades. Managed SaaS for small teams: one shared multi-tenant deployment, run by us. Sign up and go, with your data never mixed with another customer's, never trained on, and exportable. The custom grade for enterprises: deployed on your own infrastructure, or as a dedicated isolated unit we host exclusively for you. Single-tenant isolation lives there: your own database, your own storage, your own domain.

Model access is provisioned and governed as part of the deployment. Your model risk team reviews how it is governed once, and every employee works under the same arrangement. Your data is never used to train models, ours or a provider's.

No. That is what the custom grade is for. In the on-your-infrastructure arrangement everything runs inside your perimeter: documents, findings, audit trail. Many risk and compliance teams can only buy this way, which is why we built it.

On the custom grade, you choose the region. If your regulator requires data to stay in-country, it stays in-country. Residency is a configuration, not a negotiation.

We run the employee on a historical corpus your team already reviewed, for example 300 closed loan files. It reviews all of them; you compare its findings against what your reviewers found and what later surfaced. No production access, no integration project, no risk to live operations. On the custom grade, this is the Deploy narrow phase of the engagement.

Weeks, not quarters. The corpus already exists and the foundation is already built, so most of the elapsed time is your team scoring the output. You leave with numbers: catch rate, coverage, and disagreement rate on your own files.

Three things. Catch rate: findings your sampled process missed. Coverage: 100% of files read versus the 2 to 5% a sampling process reviews. Money: the value of what was caught, in currency, not activity metrics. If the numbers are not there, you have spent a few weeks learning that cheaply.

Because both sides should have skin in it. A free pilot gets deprioritized by everyone and proves nothing. A paid pilot gets a named owner, a deadline, and an honest verdict. The pilot fee credits against the first year if you proceed.

Two grades. Managed SaaS: a monthly fee per AI employee with a usage limit included, and per-unit billing beyond it. The custom arrangements, on your infrastructure or a dedicated isolated unit: priced in the contract. In every arrangement a unit of work is something your team approved: a reviewed file, a closed exception, an audited invoice. Never per seat.

The loaded cost of the review capacity you would otherwise hire, and the cost of what sampling misses. A second-line team sampling 3% of files leaves 97% unread. Price the employee against reading all of it, then check the pilot numbers against your own portfolio.

Your volumes are not a mystery to you. You know your monthly file count, exception count, and invoice count. Multiply. We cap monthly spend, and the meter is visible to you at all times, so the bill is boring on purpose.

No. Your documents run your employees; they do not train anyone's model, ours or a provider's. Model access is provisioned and governed as part of the deployment, with data-protection commitments in the contract.

No. Each employee is scoped to the document sources its job requires and nothing else. Vera sees loan files, not HR records. Access is permission-aware and every read is logged, so you can answer 'what did it look at' with a query, not a meeting.

Yes. Findings, memos, and the full audit trail are exportable in standard formats at any time, including on exit. Your review record is yours. A vendor holding your audit trail hostage would be a strange thing to accept from a governance product.

Early-stage. Small team, live product, first deployments in progress. We will not pretend to be a 500-person vendor. What you get instead: the people who built the foundation are the people on your pilot, and the roadmap moves at the speed of your feedback.

The employees run on Platos, an open-source runtime we build in the open and operate in every deployment. The scheduling, approval gating, and audit plumbing are inspectable, which matters when your model risk team asks how the system actually works. The specialists and their harness are ours on top.

Because the design assumes you shouldn't have to. Nothing acts without a named human's approval, every action is logged, the trail is exportable, and on the custom grade it can run entirely inside your perimeter. The pilot is on historical data. Trust arrives through the numbers, not the pitch.