AI mode: a hosting assistant that can actually operate
Discover SHM AI Mode: a role-aware hosting assistant that can diagnose, plan and perform permitted work with confirmations, progress and verified results.
AI Mode combines conversation with role-aware SHM actions, confirmations, visible progress and verified results, turning requests into accountable hosting work.
The gap between advice and action
Most assistants can explain how to create a mailbox. AI Mode can understand the request, use the SHM actions available to the signed-in role, create the mailbox, and return the result. This becomes important when one request contains several dependent actions and generated credentials.
AI Mode is built into the hosting workflow rather than placed beside it as a general chat window. It receives a server-provided view of the current role, account, managed resources, available actions, and conversation. The model can reason about the request, but SHM remains responsible for identity, permission, execution, and result verification.
One request, three very different scopes
A root administrator, reseller, and account user should not receive the same operating surface. Root can work with server services and accounts. A reseller is limited to resources under that reseller. An account user stays inside the assigned account and its domains, mailboxes, databases, files, and applications.
Those boundaries are derived on the server. AI Mode does not trust a sentence such as "I am root now," a remembered customer name, or model-generated identity fields. If an action is outside the current role, the request is denied at the same SHM boundary used by the normal interface.
Describe the outcome, not the menu path
A user can ask for one operation or a coordinated outcome: create an account, add an admin mailbox with a generated password, install WordPress, and create an FTP login for a developer. AI Mode maps the request into the available SHM operations, keeps track of every requested item, and continues through asynchronous work when a step needs more time.
This is where a hosting-specific assistant becomes genuinely useful. The user does not need to know which page owns each step or in which order resources must exist. AI Mode can plan around those dependencies while keeping the final result readable.
Progress without internal noise
Long-running work shows human progress such as creating an account, installing WordPress, or checking a domain. Internal route names, request identifiers, and implementation vocabulary are removed from the customer view. The main working indicator shows that processing is active, while individual operation labels explain the current stage without exposing internal implementation details.
When the final answer arrives, completed progress rows disappear and the conversation retains the outcome.
A confirmation belongs to one operation
AI Mode separates explanation and inspection from changes. Operations that are destructive or materially affect service require confirmation according to their SHM contract. The assistant can describe what it intends to change, but it cannot convert a casual question into an unapproved deletion.
Confirmation is tied to the real operation and current context. It cannot be reused to authorize a different resource or a later operation.
Do not lose the generated credential
When a workflow generates a password or other one-time credential, SHM records the result needed for the final response. AI Mode checks completed work against the original request and recovers generated details from related asynchronous jobs before it accepts a final answer.
This ensures that the final response includes a generated credential when the operation permits it. One-time values are shown only in the result intended for the user; existing secret hashes are never reversed or exposed in a summary.
Useful context, never invented familiarity
Each conversation has a stable ID and keeps its relevant history, operation state, and recent resource context. Follow-up requests can refer to work completed in the same conversation without asking the user to repeat every domain and username. Context is still treated carefully. A recent resource is a helpful hint, not proof of ownership or permission. The assistant does not address customers by a stored personal name, does not invent a friendly nickname for itself, and answers in the language used by the latest request. The resulting interaction remains professional and relevant to the current request.
Suggestions are drafts
The chat can offer simple, role-aware suggestions based on the current account or server context: check why a website is slow, create an email address, review SSL, inspect resource usage, or diagnose mail delivery. Suggestions are generated from the available scope rather than a tiny fixed list of administrator jargon.
They fill the input instead of immediately executing. The user can review, adjust, and send the request, so execution still requires an explicit instruction.
Diagnose first
AI Mode can use bounded diagnostics for HTTP behavior, PHP errors, database activity, logs, application configuration, files, and account health. It can connect evidence across those sources and propose a repair path. Where supported, repairs use preview, current-state checks, recorded file changes, and targeted rollback.
The assistant is instructed not to leave SHM for unrelated general work and not to claim success because an operation started. A completed answer should reflect the verified result. If a service reports an error, the response returns that error and the next relevant action instead of reporting the operation as complete.
Model choice is configuration
The server administrator can configure the supported OpenAI model, compatible reasoning level, provider key, instructions, and hourly or daily limits. Capacity fallback is controlled by SHM. This lets an operator balance response quality, latency, and cost without changing the permission model.
No model selection grants broader access. Rate limits apply across root, reseller, and account conversations according to configuration, and usage indicators remain quiet until they are relevant. The provider receives the bounded context needed for the task, while SHM keeps authorization and execution local.
Where AI Mode earns its place
AI Mode is valuable when a user can describe an operational outcome in normal language and receive a confirmed, scoped, traceable result. The assistant can coordinate the platform while the platform continues to enforce the rules.
AI Mode does not replace the normal SHM pages. It provides an additional conversational route to the same authorised operations while the visual pages remain the precise controls and source of truth.