AI in OneView
The AI assistant is an optional licensed Module. It appears in the application only when the installed release supports AI, the current signed license includes the AI Module, and the signed-in user has the permission needed for the requested surface.
The client supplies the API key for an available external AI provider. OneView does not supply a shared model account. An administrator selects one tested provider and model as the installation default. That default is used for context chat and automatic transaction investigations unless a supported workflow explicitly selects another connected configuration.
AI is ready for use only when all of the following are true:
- the installed release includes the provider adapter and model catalogue entry;
- the current license includes the AI Module;
- at least one provider configuration is active;
- its most recent connection test passed; and
- the configuration is selected as the installation default.
See AI provider setup for configuration and troubleshooting.
Context chat
A user with ai:use can open context chat from the button in the application header. The chat opens beside the current
page, so the user can continue working while asking about that page. Conversations and messages are retained in the
client installation and belong to the user who created them.
The assistant can use released OneView tools to answer questions about live installation data or prepare supported actions. Before each tool is run, OneView checks both the user's permission for that operation and the client's current license entitlement. If either check fails, no operation is performed. A hidden button or an instruction written in a chat message cannot bypass those server-side controls.
Tool results are bounded and redacted before they are supplied to the configured model. They are working input for the model, not the answer shown to the user. The assistant may make several authorised tool calls and then returns one conversational response. It does not display raw tool JSON as its final answer.
Most write actions require the user to review and confirm the prepared action before OneView applies it. Confirmations expire if they are not resolved promptly. AI tools never provide record-deletion actions. Transaction AI investigation is a narrow exception: it saves a derived investigation without confirmation but does not change the transaction's decision or analyst review. See AI transaction investigations.
External-provider data boundary
Messages, current page context, bounded recent conversation history, authorised tool results, and investigation evidence needed to answer a request are sent to the selected external AI provider. This is different from OneView's licensing and update traffic.
Before enabling AI, the client must assess the provider contract, processing locations, retention, training and logging terms, access controls, and lawful basis. Restrict AI permissions to approved roles, do not place secrets or unnecessary personal data in prompts, and review the provider account's own security controls. OneView never returns the configured provider API key in the application or supplies it to the model as conversation content.
AI output can be incomplete or incorrect. Treat it as decision support: verify important conclusions against the stored OneView record and the client's approved procedures before taking action.