You have stopped asking whether Xuda can host you. Now you are deciding whose hardware it runs on.
Five kinds of organisation arrive at this band, and they arrive for the same reason. Banks, lenders and insurers have to prove who they are dealing with before they transact. Hospital groups and clinic networks run one application in many places and cannot let the records travel. Public bodies buy on a contract rather than a card. Retail chains are sized by one peak day a year. Plants push machine, line and sensor data into applications and need logic sitting next to that data, not a screen on top of it. Xuda answers all five with three rungs, and they are three answers to a single question: whose hardware. Sync keeps you on ours, across two regions and a pool of machines. Host runs the platform on hardware you already own or rent. Own installs the platform inside your building and stops there. Nothing on this page asks what industry you are in, and no rung is a vertical edition of another one.
What an organisation at this band actually builds.
Not capabilities. Four things people in these five industries are running now, with the surface each one sits on. Three of the four are not gated on the Enterprise band at all, which is worth knowing before you price the move: what Enterprise buys is the ceiling and the placement, not usually the feature itself.
- One isolated copy of an app per clinic, department, branch or banner. The Instance app type deploys a copy with its own clean pre-configured database, assignable, with Studio access, version control, backups and ownership transfer. The ceiling is the tier: 500 below Enterprise, 2,000 at T1, 5,000 at T2.
- Server logic that fires on a database change or on a clock. The emitter deployment type triggers server logic on database changes or app activity; cron schedules server-side tasks from a cron expression. Both drop the launch, domain, firewall, analytics and user-assist tabs, because there is no screen. Emitter is $75 a month, the most expensive per-deployment adder there is, and config marks it beta.
- Case intake from a public form, an email and the API landing in one queue. The tickets module gives each item a status, an owner and a history, and the AI autonomy dial is the real setting: none on free, draft at $19, reply at $49, resolve at $149. It is its own plan category, so it does not wait for a membership upgrade.
- A record of who did what to which resource, exportable. get_security_logs is a real method in the logs module, with a security_event notification topic and a dedicated writer for SSH events, surfaced on the dashboard security card. It is not membership gated, and unlike most governance language it exists in code rather than on a page.
Five industries, and the one thing each of them came for.
None of these is a template or a vertical edition. They are five organisations arriving from five directions, and four of them hit the same wall, each wanting a specific surface rather than a package. Here is the surface, named, so you can check it is the one you meant.
A month, where Enterprise starts. T2 Host is $4,999.
Isolated tenant instances at T1, against 500 on the rung below
Production-system-down response on the top support rung, $4,500 a month plus 7% of usage
The literal value in every T3 quota field, where the others hold numbers
What each of the three actually holds.
Read straight off the plan records. The first five rows are ceilings the backend enforces, the next three are capabilities that flip on and stay on, and the last is how you buy it. Note what is missing: no row flips at T2 for datacenter replication, because that one already flipped at T1.
T1 Sync, $1,999 | T2 Host, $4,999 | T3 Own, call | |
|---|---|---|---|
| Projects | 5,000 | 10,000 | unlimited |
| Editor seats | 5,000 | 10,000 | unlimited |
| Isolated instances | 2,000 | 5,000 | unlimited |
| Drive | 5,000 GB | 10,000 GB | unlimited |
| AI credits | 10,000 | 50,000 | unlimited |
| Replicate a datacenter, and balance a pool | |||
| Attach servers you already own or rent | |||
| Run the platform on your own machines | |||
| Pick any support rung, including the top one at $4,500 plus 7% of usage | |||
| How you buy it | Self-serve, monthly | Self-serve, monthly | A conversation. The price field is the word call |
Three answers to one question, and the question is whose hardware.
The plan object groups these three together, which is why Enterprise is one page and not three. Read them as a sequence of problems, not a sequence of prices: each rung exists because the one below it ran out of an answer for something specific.
- 01
T1 Sync, $1,999 a month
You already run production on Xuda and you have outgrown one region and one box. What you need is the same data in two places and the same traffic across several. A datacenter is a real app type whose second wizard step is Replication, literally choose a datacenter to replicate from, and the app carries its own replication tab afterwards. A load balancer is a small dedicated box running haproxy in front of a pool you edit on the Backends tab, and saving reloads haproxy on the machine. The estate ceiling moves too: 2,000 isolated tenant instances, four times the rung below. And support stops reading limited selection, which is what puts the top rung, $4,500 a month plus 7% of monthly Xuda usage and 10 minutes for a production outage, within reach.
- 02
T2 Host, $4,999 a month
The machines have to be yours, or your provider's, but you still want somebody else running the platform layer on top of them. You attach a VPS you already own over SSH: no passwords and no private keys are stored, you authorize with a one-line command, and the tab set is cut to overview, logs, files, team and advanced, because this is an SSH inventory rather than a box Xuda can power-cycle. The numbers change shape too: 10,000 seats, 10,000 projects, 50,000 AI credits and 5,000 instances on one account, which is where a platform business rather than a client roster begins.
- 03
T3 Own, and the price field is the word call
You cannot send data outside your own perimeter at all, so the question stops being which region and becomes which building. The whole platform installs on your own metal or hypervisor: Studio and the agentic workspace, the runtime and renderer, the router with CPI and the message brokers, the datasource layer, accounts and team sharing and the security log, on a hardened host with nftables and a WireGuard mesh. You bring your own OpenAI or Anthropic key and the coding-agent runs bill to your contract instead of ours. Every quota field stops being a number and becomes the string unlimited, which matters because a number in the plan record is a cap the backend enforces. There is no self-serve path here, and the price field says so.
Six things worth knowing before a procurement call.
- T1's own plan description says unlimited projects, teams and instances. Its own feature values are 5,000, 5,000 and 2,000. Only T3 carries the literal string unlimited, and only that string is genuinely uncapped.
- The compliance card on the On-Prem page names GDPR, HIPAA, SOC 2, ISO 27001 and FedRAMP as boundaries the deployment maps to. That is a page-authored claim about architectural fit, not an audit, a certification or an authorization, and nothing in the platform configuration backs a compliance artefact. Take that one to sales.
- T2 is written here the way its tier description reads, not as a shipped cluster manager. The External VPS app type is real and works, but no code anywhere reads the outsourced-server entitlement string, so the cluster framing rests on the tier description alone.
- On-Prem today has no container image and no Helm chart. The stack runs under pm2, and the machine still needs outbound internet, because the installer pulls modules from a private registry and keeps an encrypted link back to Xuda. Air-gapped is not what this is.
- The four On-Prem support plans, Standard through Sovereign Custom, are page-authored and are not the four support rungs in the plan object. Do not read them as one ladder, or quote a price from one against the terms of the other.
- A load balancer refuses a pool of fewer than two backend nodes, server-side. And two claims made elsewhere on this site are left out on purpose: the datacenter and region count on the pricing page cannot be derived from the region list, and no offline sync surface exists in the product, so nothing here is called offline-capable.
Four questions, answered the same way on all five of these pages
The last one is what most pages leave out. It is here because you will find the disagreement yourself within a week.
Pick the rung by the boundary, not by the headcount.
Sync if you need your data in two places. Host if the machines have to be yours. Own if nothing may leave the building. The numbers for all three sit on the pricing page, and if this was the wrong door, the other four are one click away.