Where your data sits is your call.
liuma runs on top of the systems you already have; it does not replace them. There are two deployment shapes, and which one fits depends on whether your data may leave the building.
On your own hardware: data never leaves
The whole of liuma installs on your machines and runs inside your network. Business data, conversations, deliverables and audit records all stay in your storage. Nothing is sent back.
Connections to external systems are authorized by you one at a time, with the credentials held on your side. A system it has not been given is a system it cannot reach — and it will not route around that.
Cloud: Singapore instance
If you would rather not operate it yourself, use our Singapore instance. Each customer gets an isolated workspace; data is never co-mingled.
Cloud and on-premise are the same codebase — there is no reduced edition. You can move from the cloud onto your own machines at any point.
Model endpoints
It can use an inference endpoint you already run, including open-weight models you host yourself, or one we provide. Switching models does not invalidate the conventions and criteria you have taught it.
Permissions and audit trail
Every tool call and every verdict stays on file, traceable to the result file the conclusion rested on. The audit reads the trail itself, not an explanation written afterwards.
Writing back into business systems is off by default. Turning it on is done system by system, action by action.
