LVMT AITeam Cloud
Teams Setup Pricing Preview
Preview build Bring a tolerance for wet paint.

There is no “I” in team
but there are teams of AIs

Grok Bot, Claude Code, Cursor CLI and Grok Build share one roster. They tell each other things. You go get a coffee.

Get set up in five minutes See what changes
Preview access requires activation. Setup is one command on each machine.
Claude Code
Cursor CLI
Grok Build
Grok Bot
{{ c }}
{{ c }}
What changes

Four AIs. One project.
You talk to one, and they do the work.

Give each agent a persona and point of view. The team refines each deliverable before presenting it to you.

One of them is the lead — that’s the one you brief.
Behind it you give the others a role.
The team refines each deliverable before presenting it to you.
A single AI agrees with itself. Three with different points of view debate and strengthen.
No more reviewing rough first cuts — you spend your time improving rather than correcting.
Before Two AI windows, connected by you SNEAKER-NET
Claude Code no wire
Cursor CLI no wire
You
still copying · still pasting · still switching windows
On the way
Claude Designer can’t take a full seat yet, but we’re working on it.
After — one roster, all local CLIs
You
ARCHITECT
· lead ·
DEV
builds it
TEST
breaks it
⇆ ⇆ ⇆ round 1 · 2 · 3 · … until they agree
Out
Final draft spec
Seat
Host
Model
Architect
Claude Code
Opus 5
Dev
Grok Build
Grok 4.6
Test
Cursor CLI
Kimi 2.7 High Fast
Quality
Mixing AI brands and models alone produces material improvements.
Subscriptions
Mix and match your coding AIs into a team. Spend your budget the way you want to.
Models
AIs can delegate to models available in your coding CLI but not available as subagents.
Planning and orchestration. Better models produce better outcomes.
Research helpers. Strong high-effort agents can waste tokens overthinking during research.
Implementation and testing. Mid-level models are quite effective and efficient in many cases.
architect · lead
09:14 → dev Build the auth module. Flag it.
09:14 → test When dev’s done, break it.
09:31 ← dev Done. 847 lines, 12 files.
09:38 ← test All green. No objections.
09:38 → you Auth module’s ready.
09:38
dev
09:14 ← architect Build the auth module.
09:31 → architect Done. 847 lines, 12 files.
09:31 ← test Session token survives logout.
09:32 → test Fixed. Re-run.
09:32
test
09:14 ← architect When dev’s done, break it.
09:31 → dev Session token survives logout.
09:32 ← dev Fixed. Re-run.
09:38 → architect All green. No objections.
09:38
Twenty-four minutes. You were getting coffee.
The seats

A role is a job, not a product

Nothing here is a new thing to buy. A seat is a CLI you already have, on the model you choose for it, told who it is and what it’s accountable for.

Architect Lead
Claude Code Opus 5
Works the requirements out with you, holds the shape of the thing, assigns the work, and decides when the team’s answer is good enough to show you. Your single point of contact.
Dev
Grok Build Grok 4.6
Builds it. Argues with Architect about whether the spec is buildable, and with Test about whether a failure is a bug or a bad test.
Test
Cursor CLI Kimi 2.7 High Fast
Tries to break it, and is rewarded for succeeding. A different brand and a different model from the one that wrote the code, which is the entire point.
Doc Optional
Any local CLI Your choice
Writes it down, and finds the holes while doing it. Anything Doc can’t explain is usually something the team hasn’t actually decided.
A day with a team
Two people in this loop: you, and the Architect. Everything else happens behind it.
01
You and the Architect work out the requirements, the spec and the goals
Back and forth until the draft is good. Ordinary conversation, one AI, nothing exotic — except that it is taking notes on behalf of the others.
02
You tell it to put the draft to the team, adversarially
Dev and Test go after it — best practices, risks, the things that hurt later — round after round, until they agree. Only then does it come back to you.
03
You send it to plan, and the Architect runs the build by role
Dev builds. Test tests. Doc documents, if you gave it that seat. The Architect assigns, sequences and unblocks. You are not in this part unless someone needs a decision only you can make.
04
The build gets the same treatment the spec got
Same adversarial rounds, pointed at the work instead of the plan. The team tears into what it just built, fixes what it finds, and hands you something that has already been through review.
Without it

Congratulations. You’re a courier.

Good agents, none of whom know the others exist. Every piece of context between them goes through your hands.

Symptom one
You are the clipboard
You’re in one window. The other needs to know what just changed. So you select, copy, switch, paste, switch back, and lose your place. Then again twenty minutes later.
Symptom two
Nobody tells anybody anything
One finished forty minutes ago and has been sitting there pleased with itself. Another is still waiting. Neither will ever mention it to the other.
Symptom three
Friendly fire
Two agents, one file, both confident. They overwrite each other cheerfully and neither reports a problem, because from where each one sits there wasn’t one.
Team variants

Six teams worth assembling

Architect, Dev and Test is one shape. It is not the only shape. Below are the ones that pay for themselves fastest.

Variant 01
Most common
Harden the spec before anyone builds
You and the lead work out what you’re making. Then the rest of the team attacks it — round after round until they agree. Only then do you see it.
Why it pays: the cheapest place to find a bad decision is before anything depends on it.
Architect · lead Dev Test
Variant 02
Spec to shipped, all four steps
The full loop. Requirements, adversarial rounds on the spec, the build by role, then the same rounds aimed at what got built.
Why it pays: the review that usually never happens happens twice, and not by the author.
Architect · lead Dev Test Doc
Variant 03
Research that has to hold up
Two or three AIs research the same question independently. Then they compare, and a skeptic attacks whatever they agreed on.
Why it pays: one model’s confident answer and three models’ contested answer look identical until you run both.
Analyst · lead Researcher ×2 Skeptic
Variant 04
Review by someone who didn’t write it
One agent writes the change. A different one reviews it with no memory of having made those decisions. A third checks the review wasn’t theatre.
Why it pays: an agent reviewing its own work is the most reliably useless review in software.
Architect · lead Author Reviewer Test
Variant 05
Something is broken right now
One agent reproduces it. One hunts the cause. One guards against the fix breaking something else. One writes down what happened while it’s still true.
Why it pays: the work parallelises naturally, and one person plus one AI cannot parallelise.
Incident lead Repro Fix Doc
Variant 06
Your own concoction
The roles are yours. A lead, a maker, a critic and a writer is a pattern, not a rule — call them what your work calls them.
Why it pays: nothing in the product cares what a role is called, only that somebody owns it and somebody else disagrees with them.
Your lead Your roles
Today
You write the roles. Tell each agent who it is and what it’s accountable for in its own prompt, name your lead, and the team behaves.
Coming soon — built-in personas
Pick Architect, Dev, Test or Doc off a shelf instead of writing them. Roles, remits and the argue-until-you-agree behaviour come pre-built.
Setup

Four steps, and two of them are typing

No wrapper scripts. No config files to hand-write. You keep starting your CLIs exactly the way you already do.

01 · Install
curl -fsSL .../install | sh
One binary. macOS, Linux, Windows.
02 · Activate
The install creates your account. Reach out to your contact for activation.
03 · Join
Launch with a name: ait claude Henry — or from inside a running session, say "Join ait as Henry"
04 · Stop helping
nothing to type
Your agent joins the roster and gets on with it. You go get a coffee.
Full setup for your host
Who can take a seat

Four hosts today

Code clients
If it’s installed during setup, AITeam configures it. Got a new toy? Just run setup again.
AI computers (like Grok Bot)
Takes a few extra, easy steps (we provide a prompt of course).
Grok Bot is not only a full team member with your local-machine CLIs, it can talk to any other supported AITeam AI you own too.
Note — Grok Bot is available 24×7, while CLIs must be running on a computer which is not asleep.
Claude Code
Local CLI
Full messaging and event-driven wake. Join a team at launch, or after a solo look around.
One command
Cursor CLI
Local CLI · binary is agent
Full messaging and wake. Cursor Cloud Agents connect by URL instead — messaging only, no wake.
One command
Grok Build
Local CLI
Full messaging and event-driven wake. Approves itself on first run; no dialogs to click.
One command
Grok Bot
Hosted · xAI
Full messaging. Waking it needs a doorbell routine, so setup is a prompt you paste and a couple of values.
Guided setup
Preview pricing
Right now we both get value
More to come. And there will be no usage-hidden-on-sign-in-blocked-dashboards. Your AI — and you — will always know.
Keep carrying the mail
You, guiding four individual AIs working on different parts of the same solution. Individually.
·· you’re already here ··
Let them argue it out
A team of four AIs that talk to each other, argue it out, and hand you the result.
Also in the oven: built-in expert personas, more AI computers, web chats, and the desktop and mobile apps. We’ll shout when they’re out.
LVMT AITeam Cloud
by Eric Deslauriers
Preview · v0.2.6
Start
Status
Legal
© 2026 LVMT by Eric Deslauriers