Staging draft Work in progress · For review; coverage and acceptance incomplete

Save and manage a personal preference

Use personal memory for a small, deliberate preference that may help future HiveAI conversations. Review the exact wording before saving. The fact that something appeared in a conversation does not mean you agreed to save it as personal memory.

Availability: The save sequence below follows the inspected 06 Final Screens. The current frontend does not yet deliver that HiveAI journey. The backend supports additional owner controls, but their complete screen steps have not been verified. Use those sections to understand the behavior expected from a compatible client, without assuming a particular menu or button is already available.

Before saving

Use your own account in the correct organization with current HiveAI access. Keep the preference short and useful. For example, “I prefer concise summaries with the next action stated first” describes presentation; “approve every request I submit” attempts to change authority and is not an appropriate preference.

Do not include credentials, sensitive personal information or protected material merely to make the assistant remember it. A memory is not a secure substitute for the correct governed source. Current source access still matters.

A statement must contain 3–1,024 characters after surrounding spaces are removed. The current service allowance is 50 non-deleted items; paused items still count. The service checks for an existing matching statement when creating a new item. If it reports that a preference already exists, review the existing item instead of repeatedly attempting to add it.

Review and save the proposed wording

  1. In a supported HiveAI conversation, ask it to remember the specific preference. The inspected example asks to remember a preference for weekly summaries on Friday afternoons.
  2. Read the Preference to save text in the review state. Confirm that it expresses what you intended, without extra assumptions or sensitive details.
  3. Read the reminder that nothing has been saved yet. If the wording is wrong, do not confirm it; ask for the wording to be corrected. An inline editing control was not established in the inspected review frames.
  4. When the proposed wording is right, choose Save memory.
  5. Check for Preference saved and compare the saved wording with your intended preference. A review card or a request to save is not a saved outcome.

The mobile design keeps the preference and confirmation within the conversation, with wider navigation behind Menu. The desktop and mobile review states show Waiting for confirmation; the saved companions show Saved. These are the intended visible states, not proof that a deployed client persisted the item.

The example about Friday summaries records a preference only. It does not schedule a briefing or promise delivery on Friday. A scheduled action requires its own supported setup and confirmation.

Understand when a saved preference is used

Only active, explicitly confirmed items qualify for personal-memory prompt selection. The implementation considers a bounded recent set, then applies the character budget and any required current source checks. A saved preference may therefore be absent from a particular request.

The current settings consider at most eight items within 2,048 characters, including explanatory text added by the service. Recent updates affect ordering. The service can stop before using every eligible item when the budget is reached; an omitted item does not necessarily mean it was deleted.

Preferences provide context rather than higher-priority instructions. They cannot grant access, override a source's rules or make an answer correct. Continue checking important answers and their supporting sources.

Pausing or deleting affects later selection. It does not rewrite an answer already delivered or reliably recall information already included in a request that is under way.

Correct, pause or resume a preference

The backend supports these controls for the owner. The inspected Figma save screens say preferences can be changed or deleted later, but do not show the complete management journey. Exact navigation and control labels remain pending.

Your intention Supported behavior Check afterward
Change the wording Correct the existing item using its current version The new text and the item's actual current status
Stop using it temporarily Pause an active item It is paused and retained for later review
Use it again Resume a paused item It is active again, subject to normal selection and access limits
Remove it from saved use Delete the item It no longer appears in the normal memory list or export

Correcting a paused preference leaves it paused. Resume it separately if you want it eligible again. The current action-summary path can incorrectly label a correction as active, so verify the current saved state in a compatible client rather than relying on that summary alone.

When another change occurs before your edit is saved, refresh the current item and review the change again. Do not blindly overwrite an updated version or assume that retrying with an old view is harmless.

Delete with the right expectation

Deletion clears the saved readable statement, removes its explicit-confirmation flag and excludes it from ordinary memory reads and later prompt selection. The inspected owner controls provide no restore operation.

Some identifiers, statement hashes, timestamps, classification and provenance evidence remain. Deleting a preference does not delete the conversation in which you asked to save it, erase prior answers, remove a previously exported copy or prove erasure from every related system.

If you need broader account or organization data handling, use the applicable governed process described in data lifecycle guidance. A personal-memory deletion is one part of that wider lifecycle.

Understand an export

The owner export returns a snapshot of currently readable, non-deleted preferences, including active and paused items and associated provenance. It includes an export time and the owning subject. It is not an organization-wide employee export or a complete copy of conversations, audit records, deleted preferences and all underlying source documents.

Currently inaccessible classified records can be omitted. An empty export is not proof that no historical or retained metadata exists. The backend supplies structured data; a download format, file-saving flow and import operation have not been established by the inspected screens.

Treat an exported copy according to your organization's information-handling rules. Later changes or deletion in HiveAI do not automatically update or erase a copy already delivered elsewhere.

Recover without duplicating a change

Situation What to do
Still waiting for confirmation Review the wording; save only if you want that exact preference
The saved outcome did not arrive Check the original request and current owned items before submitting a new save
The preference already exists Review the existing item and correct it if needed
The allowance is full Review existing preferences; pausing does not free a slot
Wording is rejected Remove unsuitable content and keep a short, ordinary preference; ask support if valid wording remains blocked
The item changed during editing Refresh its current text and status, then decide whether the change is still needed
A correction appears to resume a paused item Check the current saved state; the action summary can be misleading
A protected preference is missing Ask about legitimate current access; do not infer that a hidden source was deleted
Memory or HiveAI access is unavailable Confirm the correct organization and contact the administrator or support
A deleted preference still appears in old chat text Treat that chat as a separate record with its own lifecycle

After a timeout, an operation may have been committed even if the response was lost or a later read failed. Repeating the same intended operation and starting a different operation are not interchangeable. A compatible client should recover the original request and refresh state; avoid rapid repeated saves or deletions.

Give support the time, organization, visible message and a safe request reference. Do not send authentication tokens or the sensitive content of a preference merely to identify the problem.

Review with users and business owners

The initial walkthrough should cover one successful save, an unconfirmed proposal, a correction while paused, resume, deletion, an export, a stale edit and an uncertain outcome. Ask the reviewer to explain what saving does, when a preference may be omitted, what deletion retains and why a preference does not schedule work or change permissions.

These are proposed review checks. They have not been recorded as user acceptance, a live service exercise or a completed frontend delivery.

Zerohive documentationStaging draft · 22 Oct 2026

Search documentation

Search guides, platform concepts and tasks.

↑ ↓ to move Esc to close