Your team's agents work all night on hardware nobody else shares
Every engineer hands Claude Code, Codex or OpenCode the long job before going home and reads it finished in the morning, on an always-on Compute with Docker and their own subscription or key. We set up a dedicated node for your team under an Enterprise contract, or your admin runs one command on each of your own servers: Enterprise on-premise. One SLA and one invoice cover the whole team.
Dedicated node or on-premise · SLA · Paid by invoice

Your team's overnight work, under one contract
A node for your team alone
Under the contract we set up a node that runs your team's compute only: its CPU, memory and disk go to your engineers, and its kernel runs nobody else's code.
An SLA for overnight work
Your agents work through the night, and the node's uptime and support response times are written into your contract.
One invoice, your terms
Pay by invoice made out to your company with its tax details, instead of a card per engineer.
Sized to the team
One Compute per engineer, each in the size their work needs, set up with you before the first engineer signs in.
Business for a few engineers, Enterprise for the whole team
All four Business plans run on our shared nodes and bill per engineer. Enterprise puts your team on a dedicated node we run, Enterprise on-premise on your own servers.
| Business | Business Plus | Business Max | Business Ultra | Enterprise | Enterprise on-premise | |
|---|---|---|---|---|---|---|
| Price | $349 a month per engineer | $649 a month per engineer | $985 a month per engineer | $2,900 a month per engineer | Set with you | Set with you |
| Compute | 8 vCPU · 24 GB · 200 GB | 16 vCPU · 64 GB · 400 GB | 24 vCPU · 128 GB · 500 GB | 30 vCPU · 192 GB · 1,000 GB | Sized to the team | Each engineer's Compute sized to the work, up to a whole server: every core but two, its memory and disk |
| Support | First in the support queue | First in the support queue | First in the support queue | First in the support queue | Under the SLA | Under the SLA |
| Billing | Card, invoices to your company with VAT or tax ID | Card, invoices to your company with VAT or tax ID | Card, invoices to your company with VAT or tax ID | Card, invoices to your company with VAT or tax ID | Invoice | Invoice |
| Hardware | Shared nodes on our servers worldwide | Shared nodes on our servers worldwide | Shared nodes on our servers worldwide | Shared nodes on our servers worldwide | A dedicated node, set up under contract | Your Ubuntu servers or VMs: one for the whole team or more, one command to install each |
Enterprise on-premise: your hardware runs the agents, we run the platform
Code and project folders are stored on your servers; builds run there. Your admin runs one command on an Ubuntu 26.04 server, one for the whole team or several, and 30 to 60 minutes later your engineers open their terminals in the same detachbox dashboard as on every plan. We keep the node current.
- Install
- One command on Ubuntu Server 26.04, 30 to 60 minutes
- Agents
- Claude Code, Codex and OpenCode on your CPUs and memory
- Code
- Project folders and home folders on your disks
- Builds
- Docker, packages and caches inside your network
- Network
- The node only dials out through its tunnel, so you open no inbound ports
- Inside
- Terminals reach only the private networks your admin opens on the server
- Egress
- Agents reach model providers under your outbound rules
- Dashboard
- Sign-in, terminals, previews, files and Vault
- Tokens
- Subscription refresh tokens, encrypted on our server
- Updates
- node-host and the terminal image update themselves
- Support
- Under your SLA, billed by invoice
What your security team will ask
The four answers a security review asks for first.
Subscription tokens stay on our server
Codex and Claude refresh tokens are encrypted on the server. An engineer's Compute holds only an access token, which our server renews.
Company keys never enter a Compute
With Vault, each project gets its own Sallyport agent, shared by every terminal in it, with no access until you grant it, service by service. Its token works only from detachbox's addresses and is revoked when the project is deleted.
Passkeys and signed requests
Passkeys, authenticator codes and Google sign-in, with a fresh sign-in and an email for every new passkey or code. Every dashboard request is signed by a key the browser cannot export, and the activity log keeps every size change, project, saved launch and public port.
Every Compute walled off
Unprivileged containers with their own user ranges, a home folder per project, no route to private networks or to other customers, outbound mail blocked. On-premise, only your admin opens a private network, on the server itself.
- Read the security model
Your team on its own node in four steps
- Form
Tell us your team's size and builds
How many engineers, which agents, how heavy the builds, what your security team needs to see.
- Call
Settle the node and the terms
Sizes per engineer, the SLA and how you pay, in one conversation and one document.
- 30–60 min
The node goes live
We bring your dedicated node up ourselves. On-premise, your admin runs one command on each Ubuntu server and its node comes up with that engineer's Compute on it.
- Sign in
Engineers sign in
Each person gets their Compute and connects their own subscriptions. The agent waits in the first tab.
Talk to us about Enterprise
Team size, agents, what has to run overnight. We answer by email.
Questions about this
What changes with a dedicated node?
Only your team runs on hardware we set up under your contract: its CPU, memory and disk are shared with nobody. The dashboard, the terminal, previews, Vault and the agents are the same as on every plan.
Do engineers keep their own subscriptions?
Yes. Each engineer connects their own Codex and Claude subscriptions or API keys, sees their limits and can turn on rotation across their subscriptions.
Can we start on Business?
Yes. Business plans start from the dashboard, with tickets first in the support queue and invoices to your company from the first day.
Where does it run?
On servers around the world, and in the region your team needs: we connect more regions on request. With Enterprise on-premise, every engineer's Compute runs on your own servers. Agents call their model provider straight from their Compute.
What stays on our servers with Enterprise on-premise?
Every engineer's Compute: terminals, agents, Docker, builds, project folders and home folders. The dashboard, sign-in, subscription refresh tokens and Vault stay with detachbox.
What do we provide for Enterprise on-premise?
A server or a VM in your data center or cloud account for each engineer's account: Ubuntu Server 26.04, 4 CPU cores or more, 17 GiB of RAM or more, a second empty disk of 100 GiB or more, and outbound internet. Your admin runs one command. We keep the node and the terminal image updated, and your SLA covers it the same way it covers a dedicated node.
What can the agents reach inside our network?
Nothing private until your admin opens a network on the server itself. Terminals reach the internet, and every private range stays closed by default. The server's own addresses, the cloud metadata address and outbound mail stay closed whatever you open, and detachbox cannot open a network from its side.
Can an engineer on our node keep a Compute of their own?
Yes. Their Compute on your node and their own Computes on our servers live in one account, and they switch between them in the dashboard. They move a project between their Computes whole from the dashboard, files and agent sessions included. Their own Computes are billed to them at their plan's price, outside your contract.