Centralux is an AI-native engineering agency for trades and field-service operators. We take the spreadsheets, whiteboards, text threads and paper tickets that actually hold your company together, draw them as one system, and build it — quote to cash, on your own data, in an environment nobody else shares.
Nothing here gets thrown away. The estimate spreadsheet, the whiteboard schedule, the customer text thread — each one is a real interface your crew already knows how to use. We route them into a single system of record, then route clean, checkable outputs back out. The wiring is the work.
They stack in this order for a reason: you cannot automate a pipeline that has no single place to write to, and you cannot safely ask questions of data whose provenance nobody can name.
We inventory every place your operation currently stores state — spreadsheets, whiteboards, texts, email, paper, and the software you already pay for — then design one system of record the rest feed into or retire against.
Detail →Estimate, approval, schedule, invoice, payment — rebuilt as one continuous chain where every handoff leaves a record and every outward action passes a gate. This is where the leaks close.
Detail →Ask your own business questions in plain language and get an answer that shows its work — the query, the rows, the source. When the data cannot answer, the system says so instead of guessing.
Detail →Isolation, provenance and auditability built as foundation rather than retrofit. Also available on its own, as a review of a system you already run and no longer fully trust.
Detail →Most AI work fails quietly. It produces an answer that looks right, from a number nobody can trace, and by the time you discover the number was a default, you have already sent the invoice. Everything we build is designed so that class of failure is visible rather than silent.
Your data lives in an environment provisioned for you alone. No shared database, no pooled search index, no other company’s records one query away from yours. Multi-tenant is cheaper to run and impossible to fully reason about; we take the other trade.
Every value an automation writes carries where it came from — which document, which message, which system, which run. A value with no source is stored as missing, never quietly as zero. Missing is a fact worth recording; a fabricated zero corrupts every calculation downstream of it.
History is append-only. Corrections are new entries, not overwrites, so the question “why does this say that?” is answerable months later — by your bookkeeper, your accountant, or an auditor — without calling us.
Actions that leave your building — an invoice, a customer email, a payment, a document sent to a supplier — pass a human gate before they go. Internal record-keeping runs unattended; anything a customer will see is proposed, reviewed and released.
Systems built on this basis are live today for field-service, retail and e-commerce operators. We do not publish client names, logos or numbers on this site — the ones you would want to see belong to people who did not agree to be marketing material. Ask on a call and we will talk through architecture in detail.
Money is rarely lost inside a stage. It is lost in the joints between them — the estimate that never became an invoice, the change order agreed by text and never billed, the deposit collected and never applied. Each joint below is a real handoff with a gate, a record, and a tap into the audit log.
Scope and cost basis assembled from your own numbers, then versioned so a later change is a new version rather than a silent edit.
RECORDS: line items · cost source · versionThe customer accepts one specific version of the quote, through a channel that produces a record instead of a memory.
RECORDS: accepted version · timestamp · channelCrew and date are committed against the accepted scope, so what is booked and what was sold cannot drift apart unnoticed.
RECORDS: assignment · materials · calendarBuilt from the approved scope plus what the field actually recorded — every line traceable to its source, nothing retyped.
RECORDS: line items · source trace · releasePayment posted and reconciled against the invoice it belongs to. Mismatches surface as exceptions rather than disappearing into a balance.
RECORDS: payment · reconciliation stateThree phases, each with a flat fee agreed in writing before the work starts, and each one a real stopping point. You are never buying the next phase to make the last one useful.
We map how your operation actually runs — not how the org chart says it does. On site where possible, because the whiteboard tells the truth.
The system gets built in your own tenant, reviewed against real jobs rather than demo data, and run in parallel with your current process before anything is switched off.
Optional. A monthly retainer for monitoring, changes and the next build — or a clean handoff. Cancel with notice; the system and the data stay yours either way.
A discovery call is a working session, not a pitch. Come with the spreadsheet you are afraid to change and the process that only works because one person remembers it. You will leave with a drawn diagram of how that could run as one system — whether or not you ever hire us.