Linux boxes, git deploys, Postgres, object storage and cron — one console, one API, one pool of hours.
Each one works on its own and snaps together with the rest — no separate signups, no per-service invoices.
Micro-VMs that appear in seconds and stay yours. Resize RAM without a rebuild, clone one into a fleet, snapshot before you break something.
How boxes work →Connect a repo, or push straight to our remote. We build it, run the release, keep the previous one — and roll back if the new one fails.
Deploy docs →Point your agent's base URL at us and keep the same SDK. Frontier models on the same balance as everything else — no second vendor, no second invoice.
LLM docs →Buckets on every account from day one. Snapshots, build artifacts and database backups land here by default — and aws s3 sync gets them out.
Push a folder and it's live on your own domain with TLS. No CI, no build config — plus a key-value store when the page needs a backend.
Deploy docs →Automation that isn't banned on sight. Route a box's traffic through a residential exit, or pin a static IP a service already trusts.
Proxy docs →Postgres, cron, secrets, private networking, custom domains and TLS come with every account — no add-on, no separate bill.
A Linux machine with your disk, your processes and your root password — that appears faster than a Lambda cold start.
Firecracker resumes from a memory snapshot. Nothing boots.
Ask for a box, get a shell. Not “provisioning, check back in a minute”.
Create and destroy them all day long. It’s the normal way to use Uno.
apt installuno boxes clone --count 19Every button in the console is an API call, described machine-readably at /llms.txt. Paste one block and your agent can build, deploy and clean up on its own.
## Infrastructure — Uno This project runs on Uno (console.uno4.dev). Install once: curl -fsSL https://console.uno4.dev/install.sh | sh - Deploy: `uno deploy` — push-to-deploy is set up - Secrets: `uno secret get NAME` — never inline values - Database: use DATABASE_URL, injected as env - Boxes are pets: never delete or resize one unasked - Auth: scoped token in $UNO_TOKEN (never commit a key) - Full endpoint list: https://console.uno4.dev/llms.txt
Every uno command takes --json. Boxes, deploys, DNS, databases, secrets — same surface a human uses, no scraping, no browser automation.
Mint a token that can only do boxes:write and hand that to the agent. Revoke it with one call when the job is done.
Your plan is a fixed pool of hours. Pay-as-you-go is off by default, and when you turn it on it takes a dollar cap. An agent stuck in a loop hits a wall, not your card.
One POST /auth/register, no captcha, and your agent has an account with the free tier — 5 static sites, 500 MB. It ships something first; when real money is involved it hands you a Telegram link to claim the account.
Claude Code, Codex, Cursor, or your own script. No plugin to install — just the CLI and one page of docs.
Every plan is a monthly pool of RAM-hours and vCPU-hours. Run one box around the clock, or twenty for an hour — it comes out of the same pool.
One always-on service, plus room to try things next to it.
A real app with its own database, and a dev copy beside it.
A fleet, or a few people and their agents on one account.
Hours drain only while a box is running. Keep ten machines around and pay for the two that are actually up.
When the pool runs out, boxes pause and your data stays. Want them to keep running? Turn on pay-as-you-go and give it a dollar limit — $0.006 per GB-hour of RAM, $0.018 per vCPU-hour.
| Builder | Pro | Team | |
|---|---|---|---|
| Price / month | $22 | $69 | $201 |
| Largest single box | 4 GB · 2 vCPU | 16 GB · 8 vCPU | 48 GB · 16 vCPU |
| Peak across all boxes | 8 GB · 4 vCPU | 32 GB · 16 vCPU | 96 GB · 32 vCPU |
| RAM-hours / month | 3,000 | 12,000 | 36,000 |
| vCPU-hours / month | 1,500 | 6,000 | 12,000 |
| Disk total | 50 GB | 100 GB | 200 GB |
| Object storage | 100 GB | 200 GB | 500 GB |
| Dedicated IPv4 | 1 | 1 | 5 |
| Unused hours roll over | — | yes | yes |
| Regions | pick one | all | all |
| Smallest box | 512 MB · 1 vCPU | 512 MB · 1 vCPU | 512 MB · 1 vCPU |
Need more than Team — a whole machine, or GPUs? Tell us what you're running.
We pay you to save.
Hours you don't burn don't evaporate at midnight on the 1st — they carry into next month. Idle time becomes headroom for the week everything happens at once.
On Pro and Team.
Because an agent isn't a request handler. It holds a session, keeps a browser open, polls a queue, waits for a webhook at 4am. Boxes stay up, keep state on disk, and don't reset between invocations.
Because a worker shouldn't cost a week of IAM archaeology and six wired-together services. And because the bill should be a number you chose up front, not a discovery you make on the 1st.
You get the same root access — plus git deploys, Postgres, object storage, cron, secrets and a private network already in the account. And an API your agent can actually drive.
Create an account, paste one block into your agent, and let it build the rest.
scp/rsync off the box, pg_dump for the database, and aws s3 sync against our S3-compatible endpoint for the bucket. It's standard Linux and standard Postgres; there's nothing proprietary to unpick.