Skip to content
Verified

On-Prem

Run the Xuda platform stack on infrastructure you control, installed and updated by Xuda engineers under contract.

On-Prem is a delivery model, not a module you switch on. Xuda engineers install the same platform stack that runs Xuda's own production regions onto a machine you provide, then keep it updated with the same tooling Xuda uses on its own fleet. There is no self-serve purchase, no installer to download, and no license key anywhere in the platform: the arrangement is enforced by the contract, not by the software.

The public product page is at xuda.ai/on-prem-lp. Inside the product, the entry point is the Enterprise App card in the dashboard Create sheet.

Where On-Prem sits among the other models

External VPS registers a server, it does not install Xuda on it. On-Prem is the opposite: a full platform install, with everything that implies about packages, ports and ongoing updates.

What the standard install produces

Your machine runs the full platform stack and also joins Xuda's private control mesh, keeping a set of control databases (accounts, sessions, API keys, billing, logs, tickets) continuously replicated with Xuda's master. A fully isolated install with no peering back to Xuda, or a fully air-gapped install with no outbound internet, changes the engagement rather than being a setting. Say so in the first conversation if either is a requirement.

Start the conversation

Open the Enterprise App card

Press + Create anywhere in the dashboard, find Enterprise App, badged On-prem, and press Talk to us. You land on Support with the ticket panel open.

Pick a topic

The link does not preselect a valid topic, so submit stays disabled until you choose one. Inquiries is the closest fit for an On-Prem conversation.

Describe the deployment

Put five things in the ticket: what you want to run and for how many users, where the machine lives and who owns it, your isolation requirement stated plainly, any compliance requirement (single sign-on, provisioning, audit export, data residency), and your outbound network position at install time and afterward.

Choose a support plan

Support is the one part of this you can act on yourself, at xuda.ai/dashboard/plans. The plan controls which channels the support form lets you open a case on:

A greyed-out channel on the support form is this gate, not a bug.

Prepare the machine

Xuda installs onto Debian or Ubuntu, bare metal or a Proxmox virtual machine. The bootstrap brings its own dependencies, so your team does not prepare them, but should know they will be there: Node v22, pm2, Apache CouchDB 3.5.2 (bound to the private mesh only, behind an nftables guard), RabbitMQ, memcached, WireGuard, nftables, imagemagick/poppler-utils/tesseract-ocr, and Google Chrome. The install also writes environment files under /etc, a secrets file at /root/secrets.json, and a private registry credential at /root/.npmrc, each root-only.

Open the network paths

At install time and every update, the machine must reach the private npm registry, NodeSource, the CouchDB package repository, your distro's package mirrors, and Google's package host. While running, it also needs Xuda's control mesh over WireGuard, Cloudflare and Cloudflare R2, OpenAI (every AI model in the catalog is an OpenAI model, with no setting for an alternative), and Stripe for billing. Hand this list to your network team before an install window is booked.

What the install does

A Xuda engineer runs it over SSH, from Xuda's own tooling; you do not run it yourself. A single archive, the platform build, both pm2 configuration files, the platform config and the installer script, is unpacked, the private registry credential is written, every module in the platform configuration is installed, and the stack starts under pm2. Onboarding a brand-new machine needs one SSH password, once, for the first hop; it is passed only in the environment of the process that uses it, never written to disk, a log or a notification, and you should rotate it afterward.

How updates reach you

Updates are pushed from a single orchestrator when a Xuda engineer queues them, not pulled on a schedule you set.

A failed deploy leaves the installation drained out of rotation on purpose, rather than serving traffic from a half-updated machine, and it rejoins automatically once healthy, roughly a quarter hour later. Every operator action is logged as a security event you can read and export yourself.

Keep your own access

Add your public keys under Settings, then assign them to the server from its Advanced tab. Assigning keys to a non-VPS deployment needs Enterprise T1 or above.

Membership tiers that matter here

Next

Read Create and run a VPS or External VPS first if you have not confirmed the lighter, self-serve models do not already cover what you need.