Skip to content

Managed AI infrastructure

The AI infra team you don't have.

A managed layer that gives your whole team access to the AI tools they already use - Claude, ChatGPT, and coding assistants - with per-person budgets, spend visibility, and one invoice. So running it is nobody's second job.

Read the API docs
Acme IncMA

Spend / mo

$335.55

Requests

4,812

Budgets

1

budget alert

Daily spend, June

sample data

the problem

AI adoption without an owner is a liability.

For teams that adopted AI faster than anyone could govern it - several tools, several vendors, and no one whose second job is to own the spend, the keys, and the policy.

ownership

Everyone wants AI. Nobody owns it.

Your team is already pasting company data into personal chatbots on their own logins. No visibility, no policy, no owner.

spend

Spend is unbounded by default.

Usage pricing has no ceiling. Finance finds out at the end of the month, after the money is gone.

fragmentation

Every tool is its own island.

Claude Code, Cursor, ChatGPT, a chat workspace - different vendors, each with its own account, bill, and keys, and no budget or policy spanning any of them.

setup

Set up in two API calls.

Everything an admin can do is a documented endpoint - the dashboard is one client of the same API. Two calls provision the org and the team; each member claims their own key.

1Create the org

POST /v1/orgs
{ "name": "Acme Inc",
  "admin": { "email": "admin@acme.com" }, ... }
201 { "org": { "id": "org_01abc123", "status": "active" }, ... }

2Invite a member

POST /v1/orgs/org_01abc123/members
{ "email": "alice@acme.com",
  "tools": ["claude_code", "open_webui"],
  "budget": { "limit_cents": 500000, "reset": "monthly" } }
201 { "status": "pending", "allowed_models": ["claude-sonnet", "gpt-5"], ... }

3Employee claims their key via cubed login

$ cubed login
# CLI coming soon
# today: POST /v1/me/keys/claim
✓ key claimed - secret shown once

governance

Safe means enforced, not promised.

Every tool, every vendor, one gateway. Each employee credential is scoped to the tools you grant and bound to a budget - a request past the limit is blocked before it costs anything.

virtual keys

Per-person credentials, scoped to the tools you grant - issued, rotated, and revoked in one click.

hard caps

Budgets that block at the gateway, with soft warnings before the limit.

metering

Usage by member, model, and day - token counts and cost, never prompt content.

audit log

Every admin change is recorded - who changed what, and when.

What blocked looks like

POST /v1/chat/completions

429 budget_exceeded

"Budget has been exceeded. Current cost: $50.00, Max budget: $50.00"

The cap is infrastructure, not a policy document. Overruns are blocked, not discovered.

how it works

Running in a day. Not a quarter.

01

We set it up

Tell us who's on the team. We provision the org, members, budgets, and keys - no provider accounts, nothing to install.

02

Your team just works

One credential for the tools they already use - Claude Code, with Codex CLI, Cursor, and Gemini CLI integrations being verified - plus a chat workspace. You keep your own Cursor subscription: chat, inline edit, and completions route through Cubed, while Composer and Tab stay on Cursor's backend.

03

You stay in control

See who spent what, on which model, against which budget. Billing is one invoice: a per-member platform fee plus usage at cost.

Give your team AI. Keep your week.

Onboarding is concierge - tell us about your team and we'll have you running the same week.