Services

Work with us

Miniaturize. We take the behaviour your team needs and make it small enough to own: your weights, your hardware, tests you can run.

Each engagement has a fixed scope and a fixed price. We quote the price once we know the work. There is no hourly billing.

Start here

Style Transfer

Length
About 1 week
Makes smaller
A voice.
For
A team whose agents sound generic, or a person or brand whose writing is the product.
You get
A LoRA adapter on a 1.7B–4B model that writes in the voice. You own the adapter. Tests agreed before training: voice gain, and accuracy bounds on tool use and arithmetic. A short report with the results.

Agent Ops Audit

Length
2 weeks
Makes smaller
Where your agents could run small.
For
A team that knows agents could help, but not where. Or whose pilot stalled.
You get
Workflows ranked by how well a small model could handle them. One working loop on your data. A plan with costs and risks.

Agent Build Sprint

Length
3–4 weeks
Makes smaller
One production agent.
For
A team with one workflow chosen and a mandate to ship it.
You get
Harness, tools, an evaluation suite with pass bars agreed up front, guardrails, deploy and handover documents.

How the steps fit

Style Transfer leads to the Audit: what else could run this small? The Audit leads to the Sprint. Book a Sprint within 30 days of an Audit and we credit half the Audit fee.

The current Style Transfer method can make the model invent memories it does not have. The tests include a check for that, and the report says how it scored.

Next step

Write to us

Tell us the behaviour you want to own, what writing you have the rights to use, and when you need it.

Email info@moonstripe.com →

We reply with a scope and a price.