Services
Programs for your team, custom models you own, and the services that keep either running.
What we build
Workflows and automation
The machine reads the forms, matches the invoices and drafts the monthly report from the systems you already run; a person signs off where it matters.
Answers from your own documents
Staff ask a policy or procedure question in plain words and get the answer with its source shown.
Custom models
A model trained on your material, so it knows your terms, your documents and your way of working.
Video and media
We produce and edit video, and generate pictures, music and other media alongside it.
Every offer by name
Fixed fees, inclusive. Nothing per seat.
Programs
Every program includes an operations review, staff training, a written operating manual and an acceptance benchmark agreed in writing before work starts.
- Pilot
- Ninety days, one team: one workflow in production, two more chosen from an operations review.
- Annual Program Recommended
- Twelve months: five to six workflows in daily use. Recommended because one workflow rarely pays for itself; five or six, found quarter by quarter, do.
- On Premises
- The Annual Program on a machine built, installed and tuned in your building, yours outright.
Pilot → Annual Program. The first workflow is live inside 30 days. At ninety days you decide; the Pilot fee is credited against an Annual Program started within 90 days of the Pilot’s end. In the Annual Program, stop at any quarter boundary.
Inside the programs
- Integration and automation
- Agents, retrieval and tooling wired into the systems you already run.
- Software engineering
- Production systems. Often the AI is the easy part.
Custom models
Most of our projects start from an existing open-source model. We train it on your own material so it handles your terminology, documents and tasks the way your people would, then slim it down so it runs quickly and at a sensible cost. We also correct or refresh what a model knows, test it against your real work before anyone relies on it, and, where the job calls for it, train one from the ground up. Tell us what you need it to do and we will tell you what is feasible.
Trained on your data, on hardware we own, and handed over: the weights, the evaluation report, and a signed assignment.
- Feasibility check
- The first stage of every project. Three to five days: whether your data can support what you want, a written go or no-go.
- Proof of concept
- Four to six weeks: a working model on a slice of your data, an evaluation report, a demo.
- Production model
- Ten to sixteen weeks: trained, evaluated, tested adversarially, packaged to run anywhere, assigned to you.
Feasibility check → Proof of concept → Production model. Each stage is paid on its own; stop after any. The feasibility check is credited against a project within 90 days.
- Keeping it current
- Drift monitoring, scheduled retraining, updates when better base models appear.
The model leaves with you. Where it runs afterwards is a separate choice.
Ongoing services
- Private AI endpoint
- Hosted on our hardware. Single tenant, United States, nothing retained, unlimited users.
- Installed at your site
- A machine without a program: sized for your work, built, installed and tuned in your building, yours outright on delivery.
- Managed operation of your hardware
- Monitoring, model updates, quarterly retuning, critical response around the clock.
- Monitoring only
- Telemetry, alerting, a monthly health report, quarterly patching.
- Advisory retainer
- Vendor decisions, governance, roadmap.
Talk to us
The first conversation is to find out whether this suits us both. It is free and needs none of your data.