The operator keeps the software they already trust. MalvaOS reads from those systems, lets the agents reason across them together, and writes back as recommendations — never a forced action.
Equipment loggers, PLCs and satellite feeds all land as spreadsheets: mixed formats, mixed units, blank cells, free-text notes. The LLM reads all of it and returns one normalized table the site systems can use.
| ts | unit_id | hub_tmp | odo_km | note |
|---|---|---|---|---|
| 07:42 | ADT-02 | 103.4 | 48,112 | — |
| 07:44:12 | ADT 02 | 104.9 | (blank) | rising |
| 07:46 | adt-02 | 221.3°F | 48,117 | (blank) |
| 07:48 | ADT-02 | (blank) | 48,120 | (no reading) |
| 07:51 | ADT-02 | 106.2 | 48,126 | smell? |
"Volvo truck 2 has a hot smell after loaded uphill runs. No warning lamp."
Without this, someone re-keys it by hand — or the export sits unread until something breaks.
Ready to structure
The export will come back as clean rows, with the issue called out.
EPC means Engineering, Procurement and Construction. This check compares what engineering approved with what procurement is about to buy.
Item 8 · 500 kVA Cummins diesel generator plant · multiple units · Deep Sea sync + switchgear
Line 1 · Diesel generator set 400 kVA prime · Qty 2 · ex works N'Djamena · 26 wk lead
Ref GEN-QUO-0187 Rev B · 2 × genset 400 kVA · terms 30% deposit on order
Deposit 30% · PO-ETC-0412 · 2 × 400 kVA diesel genset · due on receipt
Generator bus → main switching panel → crusher, screen and bag packing line
Energisation precedes dry commissioning of the 30 tph processing plant
Normally checked by hand across engineering, procurement and finance — often after the deposit is paid.
Ready to reconcile
The check compares engineering requirements, vendor scope and commissioning dependencies.
Existing systems stay in place. MalvaOS reads from them through APIs and exports — it doesn't try to replace any vendor.
One question can pull answers from FMS + SCADA + LIMS at once. That cross-system reasoning is where the gains live.
Add Mineflow → Geology sharpens. Add Strayos → Blast sharpens. Plant and Fleet inherit the better data automatically.
Every recommendation is logged with its actual outcome (head grade, throughput, downtime, $/t) and fed back into a reward model. The orchestrator learns which agent combinations and parameter choices actually move the dollar metric — not just the theoretical one.
Fleet dispatches trucks to the nearest high-grade stockpile by GPS distance alone. Head grade lands inside spec 72% of shifts.
After ~200 shifts of feedback, the orchestrator couples Fleet with Geology. Dispatch is conditioned on the block model — it knows which face blends with which ROM stockpile to hold the 2.0% Sb target.