A governed agent workforce for the routine work of named roles — the daily performance pack, the weekly management report, the monthly board pack, the case-readiness check, the campaign compliance gate, the digital funnel review, the vendor SLA reconciliation. A deterministic mind computes every figure and decides whether a run may proceed at all; the model only drafts prose, and is shown the names of figures rather than their values. Reports arrive the way executives actually read them — a PowerPoint pack in the inbox with a short brief in the email body — and every deck ends on a provenance slide naming the run, the rules pack and the model that drafted the prose. People are interrupted at judgement forks and nowhere else, autonomy is earned one decision at a time on evidence, and every hour returned is measured against a declared baseline rather than claimed. Every run is a row: what was calculated, which rules applied and from where, what the model was shown, and who resolved what.
A governed agent workforce for the routine work of named roles — the daily performance pack, the weekly management report, the monthly board pack, the case-readiness check, the campaign compliance gate, the digital funnel review, the vendor SLA reconciliation. A deterministic mind computes every figure and decides whether a run may proceed at all; the model only drafts prose, and is shown the names of figures rather than their values. Reports arrive the way executives actually read them — a PowerPoint pack in the inbox with a short brief in the email body — and every deck ends on a provenance slide naming the run, the rules pack and the model that drafted the prose. People are interrupted at judgement forks and nowhere else, autonomy is earned one decision at a time on evidence, and every hour returned is measured against a declared baseline rather than claimed. Every run is a row: what was calculated, which rules applied and from where, what the model was shown, and who resolved what.
Any organisation where senior people spend their mornings assembling the same pack, reading the same file against the same rules, or reconciling the same report — and where “the model said so” is not an acceptable answer to an auditor. WorkForceOS is sector-neutral by construction: the agents serve roles, and the rules they are judged against come from a GraphIQ pack curated for your jurisdiction and industry. It is delivered under Outcome-as-a-Service: Xamun invests the build, deploys it on your infrastructure, and you pay for what the workforce actually runs.
Each agent serves one role and one task, on a schedule or when a record arrives. There is no box to type a request into, by design. The seven below are the functions the platform ships with; the same machinery takes new ones, because an agent is a configuration, not a codebase.
Each reads a table. A real extract — your warehouse, your service desk, your contract register — replaces the demonstration data without touching an agent. The three marked .PPTX deliver a PowerPoint pack by email; the rest produce a verdict that lands in a queue.
As a PowerPoint pack in the inbox — because that is how executives read a report, and a system that makes them log in to a dashboard to find one has moved the work rather than removed it. The email body carries a short brief of what happened; the deck is attached. The body deliberately does not restate the report, because an email and a deck that both claim to be the report are two documents to keep in step.
The period, the workspace, and one line stating that every figure was computed by code and the narrative drafted by a named model.
The period’s headline figures as tiles, and what moved — compared against the prior period, or week on week for a daily pack. Never both, and never a zero standing in for the other.
A chart and a table of the split, with each line’s share.
The series that moved beyond the rolling range, with the value, the mean and the standard deviations.
The provenance slide: the run ID, the agent, the timestamp; the obligations pack and version with how many rules applied and how many blocked; the model that drafted the prose and the statement that no figure came from it; and any rule that blocked, with its citation.
Each report has its own distribution list, set per decision in the workspace. The executives who read a pack are rarely the people who work in the system, so they are named in a directory rather than given accounts. With no list configured, the workspace’s supervisors receive it.
One send per run per person, keyed on the run — so a retry after a transient failure cannot deliver the same pack twice. The pack carries your workspace’s own sender name; absent one, the deployment signs itself rather than guessing at yours.
If no mail transport is configured the delivery is skipped, not faked: the run still happened, the pack is still on the run record to download, and the ledger still counts it. A missing credential costs you the delivery, never the work.
Because the model never computes, decides or approves. That separation is enforced by code and by tests, not by a prompt — and it travels with the document, on the provenance slide.
Loads the data. Computes every figure. Resolves the rules in force on the date being judged. Decides whether the run may go on at all. If a rule blocks, the model is never called.
Shown the names of the figures, never their values. Every number it writes is substituted from the calculation; a figure it invents fails the run. No free-text prompt from a person ever reaches it — a guard test fails the build if one does.
Whichever applies to you — because no rule is hard-coded. Every rule an agent applies is resolved from GraphIQ and carries its instrument, its article and its pack version, and every run says whether GraphIQ answered live or an offline copy did. A run is judged by the law in force on the date of the event, so a case from June resolves June’s instrument and says so. Obligations may block a run; entries still awaiting curation may only warn, and every screen says which is which. No customer identifiers leave for the knowledge base — it receives amounts, flags and risk tiers, never names, ID numbers or nationality.
Draft regulatory substrate — compliance support, not a legal opinion.
As much as the record has earned, one decision at a time, paced by how reversible the work is. The stage governs the release of what a run produced — never the gate. A blocking rule stops a run at every stage.
The system proposes. A person acts.
The system acts on sign-off.
The system acts. Audits are sampled.
Real screens from the live demo you can open above. Your own deployment is a separate instance, branded and configured to your operation.
Under Outcome-as-a-Service, with no upfront CAPEX. What you open above is a demo instance; what you buy is your own separate deployment — on your infrastructure, under your brand, source-available, and configured to how your operation actually works. Xamun invests the build, and you pay on what the workforce actually runs. A quiet month is a small invoice; a run stopped at the gate produced nothing, and is billed as nothing.
Bring us one workflow and we will show you WorkForceOS running it.
