Plenum

Our own product: an AI native construction ERP

Plenum is a product we designed without a client: an ERP for electrical contractors, with projects, procurement and inventory in one product. Eight AI agents plan, buy, chase and receive inside the limits people set, and every decision that commits money or moves stock waits for a person. We sell it as a Figma kit for teams building construction, procurement and inventory software.

Construction is where an ERP earns or loses its keep. A crew that turns up to a floor without its cable costs the day, and a switchgear order signed a week late can move an energization date by a month. The question the product has to answer, every morning, is whether the material will be there when the work is.

The job report leads with material readiness

The home screen is a job report for one project, a two level hospital in the example. It opens on readiness: how ready each electrical system is for the next 14 days, and a glass digital twin of the building with the readiness of each area pinned to it. Committed spend sits beside it against the buyout plan.

Below are the decisions waiting, with the deadline that makes them urgent, such as a switchgear order that has to be signed by a date to hold the energization, then the shortages, the deliveries by day and how often each supplier arrives on time.

Ask in plain language, then watch the run

A project manager can ask for work in a sentence: review material requirements across my active projects for the next two weeks, identify shortages, compare suppliers and prepare recommendations. The agents turn that into a run with a visible plan, nine steps across seven agents, each with its time and the agent that did it.

The run shows its results as it goes: seven shortages with a proposed fix for each, whether that is to buy, transfer or expedite. When a step fails, it says so and offers a way forward, here a supplier portal that is not connected, with a call or a retry. Approvals are routed by a written policy, and a run on sample data is labelled as a simulation rather than passed off as real.

Quotes compared against the date that matters

Three quotes for 6,000 feet of cable look like a price question until they are plotted against the date the cable is needed. The cheapest one misses the need date by a week, and the chart shades the window where it is too late for the floor. The sourcing agent recommends the only quote that lands before the need date and within 4% of the lowest price, and says so in one sentence.

Approve with a change

An approval is rarely a clean yes or no, so Plenum lets people approve with a change. A budget increase can be approved in part, with the rest held until the next quotes arrive, or approved in full on a condition. The budget line is drawn as it will look after the decision, the note goes to the people and agents involved, and the decision is logged. Edits people make are tracked against what the agents proposed.

The job, day by day

The project screens follow the job: a six week lookahead coloured by material readiness, where a package short of cable shows the shortage on the bar itself; daily site reports filed from the field with photos, with the issues they raise sent to the right agent; and inventory with a warehouse twin, days of cover for every item and a shortage watchlist the agents track until each one closes.

What the agents are worth

The agent value report answers the question every buyer of an agent product eventually asks. It counts the hours saved each week and where they came from, and the outcomes this quarter from the decision log checked against receipts: stockouts avoided, purchase orders chased before their promise date, overruns caught before the order went out, and stock moved between jobs instead of bought. Trust is measured by agent, as how often people accepted a proposal as it was, edited it or rejected it.

21 charts for jobs, stock and suppliers

Operational data needs more than bars and lines, so the file includes a chart library of 21 components, from capsule bars and a readiness radar to a Gantt lookahead, a cost waterfall, a rack heatmap, a purchase order funnel and a risk matrix, with four chart states and a visualization matrix for choosing between them.

A system for live operational data

The file runs on 66 variables, 54 colour, 7 spacing and 5 radius tokens, with 17 component sets and 63 variants plus 36 standalone components, and 16 text styles in Onest. The 35 screens span six modules, global, construction, procurement, inventory, AI agents and reports, with 93% of frames in auto layout, three clickable prototype flows, the information architecture, the user flows and a design QA page. The kit is on our products page.