Lumin is the operator OS for SME and corporate teams in logistics, finance, and customer ops. Build an agent that answers nine in ten tickets, posts to your systems of record, and leaves a clean receipt for the tenth.
Customers and carriers email at all hours — quotes, confirmations, exceptions. Operators paste, label, and route. The work scales linearly with the business.
The same five questions, dozens of times a day. Logged into the CRM by hand — often missed, never analyzable, almost never tied back to a ticket.
PDFs, scans, faxes — invoices, PODs, rate confirmations. Numbers transcribed into two systems, sometimes three. Errors compound silently.
The 6% that's odd: late carriers, missing POD, partial fills, reroutes. Nobody owns these end-to-end — until one of them bites the customer.
Two weeks of inbox, a stack of PDFs, your TMS export. We sit beside you for an afternoon and watch what your operators actually do.
We co-build the first agent in our Builder — your rules, your prompts, your guardrails. You see every step it takes; you can change any of them.
The agent runs in your inbox. Every action posts a receipt: what it did, why, which rule fired, what the human needs to look at. No black box.
Quote, book, dispatch, document. Lumin lives in the inbox and posts straight into your TMS — same lane every time.
AP/AR exceptions, KYC sweeps, contract review. Receipts attached to every action, audit-ready.
Tickets, chat, returns, refunds — answered in your tone, logged in your CRM, never invented.
“It’s not the AI we doubted — it’s the receipt. Lumin is the first thing my auditors didn’t have to argue with.”
No marketing answers. If yours isn’t here, the demo will get to it in ten minutes.
Three weeks, end to end. Week 1 we shadow your operators and pick the first agent. Week 2 we co-build it. Week 3 it runs in your inbox, watched closely, with a manual override for everything.
Frontier (Claude · Sonnet / Haiku) for reasoning, your private retrieval for context. We pick the smallest model that hits the bar for each step — that's how we keep the per-run cost honest.
Only if you say so. Dedicated cloud + private VPC on Enterprise; for everyone else, EU/US tenants you choose. We don't train on your data. Audit log shows every byte the agent touched.
Then your operator gets it. Every step is logged, the receipt explains what the agent saw and what rule fired, and the human picks up the override. You see the failure rate per agent on the dashboard.
No. Lumin posts into the systems you already run — that's the whole point. We connect via your APIs and your auth; the agents become a layer on top.
Bring a sample of your inbox. Leave with one agent running, one playbook live, and your audit trail in your hand.