05 / Private Hosted AI

AI on NurexNode servers, wired into a workflow you define.

Your team or application connects to a model environment hosted on NurexNode servers — configured for a defined task, evaluated against representative work, and bounded by hosting, access, and data-handling terms agreed up front.

How a request becomes a record.

Working diagramHow a request becomes a business recordPrompts cross one agreed boundary; drafts come back for review before anything is written into your systems.
How a request becomes a business recordA coordinator or application sends approved fields to a NurexNode-hosted model environment that retrieves only approved sources. A draft returns beside the original for review; after confirmation the validated record is created in the support system. A timeout is checked before retrying, never silently duplicated.YOUR SIDECOORDINATOR / APPLICATIONsends approved fieldsREVIEW SCREENdraft beside the originalSUPPORT SYSTEMtask created after sign-offNUREXNODE-HOSTED ENVIRONMENTMODEL ENVIRONMENTconfigured for one taskAPPROVED SOURCESretrieval — current documents onlyapproved excerpts1 · REQUEST —APPROVED FIELDS + TASK2 · DRAFT FIELDS —MISSING INFO FLAGGED3 · VALIDATED RECORDAFTER CONFIRMATIONTIMEOUT → CHECK BEFORE RETRYAGREED INFORMATION BOUNDARY — PROMPTS IN · DRAFTS OUTtenancy, retention, and access fixed in the hosting terms
  1. RequestOnly approved fields and task instructions cross to the hosted environment.
  2. Hosted modelThe model environment is configured for one task and retrieves approved sources only.
  3. ReviewThe draft returns beside the original, with missing information flagged instead of invented.
  4. Record creationYour application validates and creates the record only after a person confirms.
  5. Timeout pathAn unconfirmed request is checked before retry — never silently duplicated.

Connect AI to a business task.

Consider this service when you want to connect to a NurexNode-hosted model environment rather than operate the model hardware in your own environment. The starting point is a repeatable task and a decision about where its output belongs: a screen for review, a draft record, or a step inside an existing application.

Bring representative inputs, the intended users, the applications involved, and any restrictions on sending information to the hosted environment. These establish whether the proposed connection and processing arrangements suit the work before integration begins.

From response to business record.

Illustrative example

Turn a support note into a draft that someone can approve

Suppose a coordinator receives a note about a replacement laptop. An application sends the approved request fields to the hosted model and asks it to prepare a structured draft: equipment needed, issue described, and information still missing. The coordinator sees the original note beside the proposed record.

If the note omits the asset identifier, the draft asks for it instead of inventing one. After review, the application creates the task in the existing support system. A connection timeout leaves that creation unconfirmed; the application checks the destination before retrying. The model prepares content, while application rules control validation and record creation.

Working excerpt01 /

Example request contract

Sent to the model
Approved request text and output instructions
Returned for review
Draft fields, missing information, original context
Required before creation
Validated fields and coordinator confirmation
On an uncertain connection
Check for the existing task before retrying

Define the output. Check the behavior.

Representative deliverables

  • A hosting and connection plan identifying access, information flows, and operating responsibilities.
  • A configured model environment and the agreed interface or workflow integration.
  • Task evaluation results, integration documentation, and a record of limitations.

Acceptance to agree

  • Check output quality against representative tasks and the expected review standard.
  • Test missing information, invalid output, denied access, timeouts, and repeated requests.
  • Measure response time and workload behavior; verify the agreed fallback when the connection is unavailable.

Prepare for operation.

Hosted operation still needs a shared responsibility plan. Identify who administers user access, maintains the client application, updates source documents, reviews model changes, and investigates failures. Agree service and support terms instead of treating the initial integration as an undefined ongoing commitment.

Document configuration, credential rotation, evaluation, and changes to the model or prompts. The operating plan should also address ending the arrangement: application dependencies, information export where applicable, and the agreed handling of stored data.

Access and operating terms.

Choose how people connect

A user-facing interface and an application connection have different access needs. Define authentication, credentials, permitted actions, request limits, and revocation. Network dependencies and the user experience during an unavailable connection belong in the design.

Map what crosses the boundary

List prompts, attachments, retrieved passages, outputs, and diagnostic information separately. Agree which are processed or stored, where, for how long, and who may access them. Avoid sending fields the task does not need.

Keep the model inside its role

Prompts describe the task and expected response. Retrieval adds approved source material without changing model weights. Integration validates the response and connects it to a workflow. A model suggestion should not silently become permission to change a business record.

Make operating terms explicit

Define the hosting arrangement, model choice, capacity assumptions, and responsibilities for updates and incidents. Discuss tenancy, location, retention, and any specific assurance requirements directly; do not infer them from the word private. Test the configured environment against the agreed use.

Before we begin.

Is the model running in our office?

No. This service provides access to a model running on NurexNode servers. Local AI is the relevant option when the model must run on hardware in your environment.

Will our requests be used to train a model?

Data use must be specified in the engagement. Sending a prompt or retrieving a document is technically distinct from training. Any proposed training or fine-tuning needs its own purpose, data permissions, and evaluation.

Can it connect to an existing application?

The application’s supported interfaces, access permissions, and data rules determine the integration. Establish how requests are sent, responses are validated, and failures are handled before making it part of an operational workflow.

Next / Your project

Which workflow would you like hosted AI to support?

Tell us what the hosted model should help users do, which application it would connect to, and what information it would process. Include any requirements that affect access or hosting arrangements.