AI and data policy
How Lunume AI chooses an action, uses workspace context and keeps you in control.
On this page
01Purpose and human control
Lunume AI can answer a question, ask for missing information or prepare a proposed record change according to the request and its context. It does not silently apply an interactive change. A proposal is shown for review and takes effect only after an authorised person uses its confirmation control. AI output can be incomplete or wrong; review facts, dates, citations and consequences before relying on it.
02Data used for a request
A request may include the text or voice prompt, output language, time zone, same-scope conversation history and relevant excerpts from records the requesting user can access. A narrow source choice is limited to those selected records. “All accessible” may also use relevant permitted workspace excerpts and personal preferences or memories that the user explicitly saved in that workspace. Voice requests and note recordings use the audio controls described below.
03Recordings and live transcription
Recording starts after the user invokes a recording control and grants the required browser or device permissions. A recording can include microphone audio and, when selected and supported, audio from a shared tab or screen. Audio parts are uploaded to Lunume while recording. When automatic transcription is enabled, short audio segments are sent through the configured AI gateway during capture; the live transcript is saved with the note. Hiding captions only hides the panel. Pausing live transcription stops new live segments, while the audio recording continues. Stopping and saving the recording can automatically produce a final transcript and summary according to the selected setting. Disabling automatic transcription stops new live requests and automatic final processing; already transmitted requests cannot be recalled. Imported audio is processed only through its transcription control. Audio, transcripts and summaries follow the note’s access permissions and retention rules.
04Data that is not automatically sent
Lunume does not send an entire workspace merely because the assistant is open. Retrieval is bounded to relevant excerpts and current access permissions. Content the user cannot access is excluded. Ordinary assistant chat does not browse the public web or call external connectors unless a separate feature clearly identifies that capability and the user invokes it. Passwords, private keys and other secrets should never be placed in a prompt.
05Providers and processing chain
AI requests pass through the server-side gateway configured for the deployment and may be processed by an AI provider or another contracted subprocessor. The current production processing schedule must identify the applicable recipients, processing locations, purposes, safeguards and contact route. Provider retention and any use for provider model training depend on the operator’s contract and configuration; Lunume does not make an unverified no-retention or no-training promise.
06History, memories and retention
Conversation history and accepted AI records can be stored in Lunume under the workspace retention rules so a conversation can continue and actions can be audited. Saved personal memories are scoped to that user and workspace and can be removed from the assistant menu. Removing a conversation pointer from a browser does not itself erase server records. Provider copies, logs, backups and legal holds follow the published processing and retention schedule.
07Permissions and safety boundaries
Retrieval and proposed changes are checked against current account and workspace permissions. An answer or quoted source cannot grant access, authorise sharing or override those checks. Interactive proposals do not delete records, send invitations, make purchases or send external messages. Scheduled work has its own explicit read scope and allowed actions; automatic write modes require separate confirmation when configured.
08Controls and limits
A user or workspace owner can disable AI for future requests, and the service may enforce request, budget and feature limits. Disabling AI cannot recall a request already transmitted. Source selection controls what the current conversation may retrieve; changing to a narrow source starts a separate scoped conversation. Product controls and available tools can vary by plan, role and deployment policy.
09Accuracy, citations and responsibility
Citations identify records used for an answer but do not guarantee that an interpretation is correct or current. Generated plans, summaries and drafts are assistance, not professional advice or proof of authorisation. Do not use an AI answer as the sole basis for a legal, medical, financial, employment or other consequential decision about a person. The person who confirms, shares or acts on an output remains responsible for that use.
10Questions and data rights
Use the Contact and requests page for access, correction, deletion, restriction or other privacy questions about AI-related personal data. Identify the workspace and approximate request date without including passwords, session data or unrelated private content. Organisational workspace requests may need to be handled by that organisation as controller, with Lunume assisting under the applicable agreement.