Custom AI agents
Agents and model training for real workAgents built around your systems, documents and decisions, with a person approving anything that touches judgment, money or risk.
Learn about custom AI agentsWe train your team, automate real workflows, and build private AI systems you actually own.
Sometimes the right answer is a product that already exists, set up properly around your work. Sometimes it's something built for you. We'll tell you which, and start with whichever is smaller.
Agents built around your systems, documents and decisions, with a person approving anything that touches judgment, money or risk.
Learn about custom AI agentsMake AI useful to the people already doing the work. Hands-on sessions built on your roles, documents and recurring tasks.
Learn about AI Cowork trainingLink your tools and data so information flows on its own — on dedicated GPU infrastructure when control matters.
Learn about private GPU hostingPut Grok Bot to work on the repetitive web and app chores, with dedicated accounts and a person approving anything that matters.
Learn about Grok Bot setupAnswer the calls you miss, take down the job in your system, and send emergencies and hard calls to a person.
Learn about AI voice agentsOne or two weeks mapping how your work runs and the hours it takes, ending with a ranked list and the first step quoted.
Learn about discovery sprintsA plain screen where your managers see what the AI drafted and the evidence behind it, and approve, edit or send it back.
Learn about approval dashboardsA quote request lands as a drawing package and a scanned PO. The system reads it, fills the quoting sheet, finds the closest job you've run, and flags what a person needs to look at.
That's the kind of AI we build.
Examples of what we can build, not client case studies. Pick the one nearest you.
It goes below zero on a Friday night and forty people call inside an hour, all of them saying the same three words. Your on-call tech is already driving. Whoever picks up is writing addresses on the back of an envelope, and the calls that can wait until Monday are burying the two that can't.
A line that answers every one of those calls. It takes the address, the system, and whether there is any heat in the house at all, checks it against tonight's schedule, and texts the on-call tech only the jobs that genuinely can't wait. Everything else is in Monday's queue by morning, already written up.
Between foliage and February vacation week you answer the same handful of questions several hundred times. Is the Kancamagus open. Do you take dogs. How far to the tram. Can we check in early. Is there room for a trailer. They arrive by email, by Booking.com message and by text, most of them at eleven at night, and a good half are from people who haven't booked yet.
An assistant trained on your own house rules, your rates, your local knowledge and the replies you've already written. It drafts the answer in the guest's language, sends the straightforward ones itself, and puts anything touching money, cancellations or a complaint in front of a person before it goes anywhere.
Every customer wants their boat in the water the same weekend in May and out of it the same weekend in October. In between, half the work is looking up what you did to hull 214 last season — which wrap it had, whether the impeller was done, what the owner was told about the outdrive — and the answer is in a folder in the back office.
Your service history, paper files included, put behind something you can just ask. "What did we do to the Regal last fall?" comes back with the work and the invoice. Then the scheduling: winterization and commissioning slots offered out by text in the order that suits the yard, rather than the order people happened to call.
Intake forms, referral letters, prior authorizations, notes written after hours because the day was full of patients. It is precisely the work AI is good at, and precisely the work you cannot put through somebody else's cloud and sleep well.
The same assistant, running on a machine in your own office. It summarizes intake, drafts the prior-auth letter, and pulls the relevant history into a note for you to correct and sign. Nothing is sent anywhere. There is no per-seat bill, and no vendor holding your records hostage at renewal.
The sap runs for five weeks; the orders arrive all year. Wholesale accounts email their POs in whatever format suits them, the farm-stand regulars text, mail-order customers want to know where their gallon has got to — and the one person who knows the answer to all of it is out in the sugarbush.
Order emails and PDFs read into a single list — customer, grade, size, quantity, ship date — so the pack list writes itself. The weekly note to your mailing list drafted from what actually happened that week. And a shop assistant that handles the grade, allergen and shipping questions before they reach your inbox at all.
Quote requests come in as emailed drawing packages, spreadsheets and scanned POs. Somebody has to read each one, pull the part number, quantity, material, tolerance and due date, work out whether you've cut it before, and put a number on it. The ones you don't answer inside two days are generally the ones you don't win.
AI that reads the drawing package and the PO, drops the line items into your quoting sheet or ERP, finds the closest job you have actually run, and drafts a first-pass number for your estimator to correct. Because a good deal of this work is ITAR- or CUI-controlled, it runs on a machine in your building — no drawing is ever sent to a public model.
A Portsmouth summer and a Portsmouth February are two different businesses in one lease. You order and staff for one and get handed the other. Meanwhile the inbox fills with private-event inquiries, allergen questions and "are you open Thanksgiving", and the answers live in three people's heads, none of them on shift.
Covers forecast from your own booking history, the weather and the local event calendar, so prep and the schedule start from a number rather than a feeling. An assistant on the inbox answering the allergen, hours, parking and private-dining questions out of your real menus and policies — and handing the event inquiries to a human with the date, headcount and budget already collected.
Half an hour on the phone or over coffee. You describe your week; we say plainly which parts AI is good at, and which it isn't.
One useful starting point. Often that's setting up a product you can use right away; sometimes it's a small custom build. You see it working before anything larger is agreed.
You get the system, the code and the notes on how it runs. Nothing we build stops working the day you stop paying us.
25 years of data engineering and systems integration taught us something: most AI projects fail on the plumbing, not the model.
Two of our assistants, ServiceMate AI and WorkMate AI, sit on top of ServiceNow and Workday, answering staff questions and handling tickets that would otherwise wait for someone to open them. Same care for permissions, integration and reliability; none of the enterprise overhead.
Read the case studiesNo. Our work is usually the less visible part: connecting your documents and systems, setting reliable boundaries, deciding when a person must review an answer, and leaving you with something that works outside a demo.
Usually not. Often the answer is an existing product, such as Grok Bot or AI tools you already pay for, set up properly around one job. When nothing off the shelf fits, we build something small and custom. Either way you see it working before anything bigger is agreed.
Repetitive jobs with recognizable inputs and a clear human decision at the end: reading order emails, drafting routine replies, extracting quote details, finding past service records, moving approved information between systems.
We map where the data goes before we build. Sensitive work can stay on equipment you control; lower-risk work can use carefully selected cloud services. You see the design and the tradeoffs first.
The first 30-minute planning call is free. If there is a worthwhile next step, we define a focused engagement and quote it before work begins.
Yes. T10 is based in New Boston and works across New Hampshire. We can start by video or phone, and travel to you when seeing the actual workflow would help.
Tell us about it. We'll spend 30 minutes figuring out whether AI can actually help.
No pitch. No deck. No obligation.
Run a New Hampshire nonprofit? We check, fix and back up nonprofit websites for free. See how it works