Every manufacturing company starts the same way: a founder's spreadsheet, a shared WhatsApp group, and a growing pile of PDFs nobody can find. Triox Mobility was no exception — until we decided that running a full-stack EV ODM on disconnected tools was a bigger risk than building our own. The result is admin.triox.in: a unified operations platform that functions as ERP, PLM, CRM, and manufacturing execution system — with an AI copilot (TX Assistant) woven through every workflow.
The problem: operational fragmentation in EV manufacturing
An ODM operation spans disciplines that traditional software silos apart. Sales talks to customers about configurations that engineering has not yet released. Procurement orders parts against BOMs that PLM has not approved. Manufacturing builds against work orders that accounting cannot invoice. Quality inspects vehicles whose warranty data lives in a different system. Each handoff is a chance for human error, delay, and data loss.
- Leads arrive from the website, WhatsApp, and trade shows — but follow-up lives in personal inboxes.
- Quotations are drafted in Word, priced in Excel, and approved over email with no version control.
- PLM drawings sit in folders; vendors receive outdated revisions because share links are manual.
- Manufacturing orders are created on the shop floor whiteboard; inventory counts drift from reality.
- Accounting reconciles bank statements against invoices that do not match what was actually shipped.
The insight
The cost of fragmentation is not the software licence you did not buy — it is the quotation that went out with last month's BOM, the part that was ordered twice, and the customer who waited three extra weeks because nobody saw the engineering change order.
The Triox admin architecture: one platform, every function
We built admin.triox.in as a Next.js static application on Firebase Hosting, backed by Supabase (PostgreSQL with Row Level Security) and Supabase Edge Functions for privileged operations. Every module shares the same database, the same auth model, and the same audit trail.
| Module | ERP Function | What It Replaces |
|---|---|---|
| Leads | CRM / pipeline | Spreadsheets, WhatsApp follow-up lists |
| Quotations & Sales Orders | Quote-to-cash | Word docs, email approvals |
| Customers & Portal | Customer self-service | Email status updates |
| PLM | Product lifecycle | Folder shares, manual revision control |
| Work Orders | Manufacturing execution | Whiteboard, paper travellers |
| Manufacturing & Assembly | Shop floor MES | Verbal instructions |
| Inventory | Stock management | Periodic manual counts |
| Vendors & Purchase Orders | Procurement | Email POs, phone confirmations |
| PDI & QA | Quality inspection | Paper checklists |
| Post-Market Surveillance | Field compliance | Ad-hoc complaint logs |
| Accounting | Finance & GST | Separate Tally/manual books |
| HR & Onboarding | People ops | Email onboarding packets |
| TX Assistant | AI copilot | Asking colleagues "where is…?" |
PLM: engineering data that manufacturing can actually use
Product Lifecycle Management is where most mid-size manufacturers fail digitally. Drawings exist, but they are not linked to BOMs. BOMs exist, but they are not linked to purchase orders. Purchase orders exist, but they reference part numbers that engineering renamed last quarter. Triox PLM ingests assembly structures, indexes documents, and exposes approved revisions to vendors through secure share links — so the part on the shop floor is always the part engineering released.
- Engineering uploads assemblies and drawings; PLM indexes folder structure and BOM lineage.
- Revisions are approved before release — manufacturing cannot build against draft drawings.
- Vendor share links expose only the latest approved revision for their scope.
- Work orders reference PLM BOMs directly — no manual transcription of part numbers.
- Cascade lineage tracks which engineering change affects which open work orders and purchase orders.
TX Assistant: AI that operates the system, not just chats
Most "AI for business" products are chatbots bolted onto existing software. TX Assistant is different: it is a goal-directed agent with server-side tools that can query products, customers, work orders, and quotations; draft documents; and render interactive UI (forms, cards, carousels) inside the conversation. It follows the same agent architecture we use for vehicle autonomy — specialised tools, structured memory, deterministic routing for critical workflows, and a safety layer that prevents unauthorised actions.
Example workflows TX Assistant handles today
"Create a quotation for Alpha 2.2 for Acme Logistics" → agent pulls product config, applies pricing rules, renders a review form. "What is the status of work order WO-2026-0142?" → agent queries live data, returns a status card. "Send onboarding documents to the new vendor" → agent triggers the document edge function with correct templates.
Quantified impact: what changed when we unified
| Metric | Before (fragmented) | After (unified admin) |
|---|---|---|
| Quotation turnaround | 3–5 business days | Same day (AI-assisted) |
| BOM-to-PO errors | ~8% mismatch rate | <1% (PLM-linked) |
| Engineering change propagation | 1–2 weeks manual | Hours (cascade alerts) |
| Inventory accuracy | ~85% (periodic counts) | >97% (transaction-linked) |
| Customer status inquiries | 30+ min staff time each | Self-service portal + AI |
| Onboarding new staff | 2–3 weeks to learn systems | 1 week (one UI, AI guide) |
Why we chose build over buy
Off-the-shelf ERP (SAP, Odoo, Zoho) and PLM (Windchill, Arena) are powerful — for companies whose processes match the software's assumptions. An EV ODM with software-defined vehicles, open ECU architectures, Vehicle Twin telemetry, and AI copilots does not fit a template designed for discrete manufacturing in 2010. Building on Supabase gave us relational integrity, RLS security, edge functions for privileged ops, and the flexibility to add AI-native workflows that no ERP vendor ships today.
- Supabase PostgreSQL: one schema linking leads → quotations → work orders → inventory → accounting.
- Row Level Security: every table gated by staff role — no accidental data leaks.
- Edge Functions: service-role operations (document upload, email, AI inference) never exposed to browser.
- Static export: admin frontend deploys to Firebase in seconds; no server to maintain.
- AI-native: TX Assistant is not a plugin — it is a first-class module with tool access to every domain.
Lessons for every manufacturing company
You do not need to be an EV ODM to benefit from this approach. Any company that designs, procures, builds, and ships physical products faces the same fragmentation. The fix is not buying more software — it is unifying data around a single source of truth and adding AI agents that operate that data on behalf of your team.
“The admin app is not overhead — it is the product that lets us ship vehicles. Every hour saved on operations is an hour invested in engineering.”
— Triox Mobility Operations
Building an EV programme? Let's talk about how platform + operations technology accelerate your path to production.
Partner With Us