Conversation data, exports and removal¶

Conversations, spoken instructions, generated audio and usage records have different lifetimes. Understanding those differences helps you decide what to submit, interpret missing history and handle a member's departure without promising more than the system has completed.

Availability: These backend lifecycle paths are implemented. The current frontend is still being developed, and live export, cleanup and removal outcomes have not been verified for this draft. This page explains background behavior; it does not introduce an unverified delete-history or export button. See the conversation guide and the speech guide for the intended user journeys.

What is kept, and why¶

The ten-minute audio lifetime starts before preparation finishes. It is not ten minutes of guaranteed listening after a voice becomes ready. A written answer may still be available after its audio expires.

Audio expiry is also separate from physical cleanup. The system checks expired artifacts through scheduled maintenance, and failures or backlog can delay their removal. Do not interpret an expired player as a deletion receipt.

No general conversation-history deletion period is established by these paths. The recent-history list and the amount of history used to answer a new question are both bounded, but those bounds do not themselves delete older records.

When history or audio becomes unavailable¶

First identify what is missing: a source, a written message, a recording, spoken playback, or the progress of an unfinished request. They require different explanations.

1. Return to the original conversation or request, where your access still permits it, and check its current saved status.
2. For expired audio, use the available written content. Creating another recording or spoken answer is a new deliberate processing request that may use credits.
3. If access or retention has changed, follow the displayed guidance or contact your administrator. An old link does not restore access.
4. If work or cost is still uncertain, give support the original reference rather than submitting the same instruction again.

Stopping recording, stopping playback, cancelling remaining work, removing a member and deleting an organization are separate actions. None should be used as a substitute for another. Stopping a request does not reverse a completed business change or erase its usage record.

What an organization export contains¶

The HiveAI contribution to an authorized organization export includes saved conversation records and written messages, alongside other tenant data. Some interaction and speech records use narrower summaries.

Audio files, the original/corrected recognition transcript copies, replay payloads, private choice descriptions and navigation credentials are not included in the speech and interaction summaries. Words that were actually submitted remain part of the separately exported conversation messages, where those messages still exist.

This means “speech transcripts omitted” does not mean “all spoken words absent from the export.” For example, if you correct a recording and send “Summarize tomorrow's agenda,” the submitted message can remain in conversation history and its export even though the original recording and recognition transcript copies are omitted.

An export is a data contribution assembled for an authorized job. It is not a complete restorable image of the platform, a guarantee that all source products are included, or proof of one perfectly simultaneous snapshot. It does not grant its recipient access to source systems.

Before sharing an export, administrators should confirm the recipient, purpose, scope and approved storage location. A minimized interaction summary does not make the entire bundle safe for public distribution.

What happens when a member leaves¶

Removing a member from an organization triggers a separate HiveAI data-removal request. Processing is asynchronous; removal of membership does not prove that every associated record and audio object has already been cleaned up.

The request uses the organization's applicable hold setting. If that policy information is unavailable, the current producer chooses the hold path. Authorized staff should resolve policy questions through the organization's governance process, rather than infer the answer from an empty conversation list.

When no hold applies, the HiveAI handler removes the departing member's owned conversation history, associated conversation runs, citations and personal memories. Conversation deletion also removes its related interaction and speech rows. Draft proposals owned by that member are removed; other relevant proposals and shared work records can remain with the member's direct participation identity removed.

Affected briefing schedules are disabled and the departing member is removed from their ownership or recipient information. Produced briefings and reusable-workflow records may remain with adjusted identity information. This process does not transfer every responsibility to a new person, reverse prior source-system actions, or remove another member's independent history.

Managers should therefore review outstanding approvals, shared workflows and scheduled briefings as part of the departure process. Confirm the responsible owner for ongoing work; a data-removal event is not a complete operational handover.

How the hold path treats content¶

The implemented hold path keeps selected records and accounting while minimizing personal content. It does not preserve the original conversation and recording as an unchanged readable archive.

Conversation text, titles, citation details and personal-memory statements are minimized. Saved interaction content, private choice labels and prepared navigation credentials are cleared or replaced with unavailable content. Original and corrected speech transcripts and recognition evidence are removed, voice preferences and audio objects are deleted, and selected status and accounting information remains.

Some identity, classification, authority, reference and timing information remains in the retained records. Minimized or pseudonymized does not mean anonymous. For example, retained classified-memory authority can still identify the original subject, and retained request records preserve ownership and accounting relationships.

A held conversation must not be presented as an invitation to recover erased content or continue a stale decision. If the history is unavailable under retention, ask the responsible administrator about the permitted next step. The current path does not reconstruct the removed words.

Why cleanup and charging may finish at different times¶

The system attempts to reconcile already-authorized speech work before member or organization cleanup. Where a speech authorization has no recorded settlement time, cleanup is blocked so that the outstanding obligation is not silently discarded.

There is a current implementation limit: an unknown spoken-answer outcome can already have a settlement timestamp while its final amount remains unknown. That record can pass the cleanup guard. Therefore, a cleanup result or a “recovery complete” label alone is not proof that all final charges are known.

Support should check the original speech job, its current outcome, the known or unknown amount and the cleanup evidence separately. Creating a new speech job does not settle the old one. A late cost receipt can clarify the original charge without recovering usable audio.

Physical audio deletion and database updates also happen in separate systems. Some audio may have been removed even if a later database step fails. An interrupted cleanup can therefore leave a mixed result that needs reconciliation; it is not necessarily a complete rollback.

Whole-organization deletion¶

Whole-organization deletion is a coordinated administrative operation. HiveAI contributes deletion of its tenant-owned database tables after its speech cleanup preparation. That contribution is distinct from member removal and does not apply the same per-member hold branch.

The overall operation must establish its own authorization, review and policy eligibility. A successful HiveAI contribution describes its own scope. It does not certify erasure of exported files, downloaded copies, external provider records, source-product data, backups or every related storage and index owner.

Counts in a deletion result describe affected database records. They are not counts of people, conversations, documents or successfully erased external files. Repeating an already-completed contribution may produce different counts; the original job and its evidence remain important.

What administrators and support should verify¶

For a removal or missing-history enquiry, keep a concise record of the organization, affected person, approximate time and original conversation, speech or data-operation reference. Avoid copying confidential content into an unrestricted ticket.

Check these outcomes separately:

• Membership and source access changed as intended.
• The correct member or organization lifecycle job was processed.
• The applicable hold decision is understood.
• Remaining shared work has an accountable owner.
• Speech processing outcomes and charges are known, or explicitly still unresolved.
• Audio cleanup succeeded within its actual scope.
• The coordinated operation has the required contributions from other owners.

Executives should use these distinctions when reviewing governance readiness: a green component status, a small export or a zero-row repeat is not an organization-wide erasure guarantee. Current environment evidence and the authorized review process are still needed.

Return to conversations, review speech recovery, or check conversation access and outcomes.