Keep your CLI.Stop paying per token.
We install an open-weight model on GPUs rented for your team and hand you the one-line connect for Claude Code. Unlimited use, billed per seat. The opening cohort is capped.
01
The problem is tokens.
Weekly caps, a “20×” plan that is not 20× tokens, API meters on every laptop. Most terminal work does not need a frontier lab model. A seat contract ends the meter.
02
It is your own CLI, and it feels like it.
Nothing new to install. Claude Code, Codex or any client that sets a base URL. Prompts go to GPUs rented for your team, not to a US lab by default.
03
It costs a fraction.
One RTX 5090 serves about eight people at once. You pay a flat seat, we rent the cards. New hires are a line item, not another Max subscription.
01
Pick a model
Like choosing an OS on a VPS. Qwen, DeepSeek, or another open-weight model. Or we pick one that fits the repo.
02
We provision it
Automatic install on rented GPU cloud. No cluster to own, no hardware to buy. Large teams get a dedicated card.
03
Connect your CLI
Claude Code, Codex, or another compatible client. One click, then the exact environment variables.
04
Add seats
Each person gets the same endpoint. Unlimited use under the contract. New hires are a line item.
Everything Claude Max gives you. At a fixed price.
Keep Claude for the few tasks that need it. Point the daily work at the endpoint so the expensive subscription is not on every chair.
−73%
Same Claude Code, fixed price, no token cap. $131 less per person every month — $15,720 a year for a team of ten.
| Claude Max | Anomalia | |
|---|---|---|
| Per person | ~$180 / month | $49 / month, fixed |
| Token limit | Weekly cap, then you wait | None. Unlimited under the contract |
| Tool | Claude Code | Claude Code — the same CLI, unchanged |
| Where prompts go | Vendor cloud | GPUs rented for your team |
| Savings per person | — | $131 / month · $1,572 / year |
| Team of 10 | $21,600 / year | $5,880 / year — save $15,720 |
Do I have to stop using Claude?
No. Keep Claude for the few tasks that need it. Point the daily work at the open-weight endpoint so the expensive subscription is not on every chair.
Is this a new CLI?
No. You keep Claude Code, Codex, or another client that can set a base URL. We are the model behind it, not a replacement for it.
Do you need our source code in your cloud?
The model runs on GPUs we rent for your team. Prompts go to that endpoint, not to a US lab by default.
Why a waitlist?
Shared GPUs serve about eight people per card. We open a fixed number of seats, fill them in list order, and add cards for the next cohort.
34 seats left.Then the list closes.
Opening cohort of 200. We reply with a short contract, in order of the list.