# Database Manager: collections and conversations

## Scope and commit behavior

Open **Settings → Database Manager**. It always manages the computer's **Local** archive, including when the reading source is a live server. It does not delete or reorganize that server. Manager actions save immediately through their own controls. Closing Settings with Cancel does not undo a completed move, merge, deletion, restore or other manager action.

The overview reports collections, conversations and compressed storage. Expand it to inspect active data, total stored data, temporary workspace, expanded catalog, RAM content cache, pending cleanup and recovery usage. Collections share storage objects, so the interface does not claim an exact independent physical size for each collection.

## Create and rename collections

Use **New collection** to create an empty destination. It appears in the manager and import destination suggestions. It appears in the normal Reader selector after it contains conversations.

Use a collection row's rename action to change the collection’s display name. Renaming preserves identity and saved reading. Choose a useful organizational label; matching names and titles are not substitute identifiers for deduplication.

## Browse and select conversations

1. Open the Conversations view or a collection's browse action.
2. Choose a collection and provider, if required.
3. Enter an optional title/message/source-file filter.
4. Press **Search** to apply it.
5. Check individual rows or use the selection controls.
6. Read the selected-count label before starting an operation.

Changing filters clears the selection. **Select this page** adds the visible rows to an explicit selection, which can survive paging. **Select all matching** selects every result in the current filter scope, not merely the visible page. **Clear selection** removes the selection without deleting anything.

Do not treat a checkbox list as an implicit confirmation of all conversations in the database. The displayed count and active filters are the authoritative operation scope. The manager uses filters independent of Reader’s current search.

## Move versus copy

| Operation | Identity and saved-reading effect |
| --- | --- |
| Move selected | Preserves the conversation identity and associated saved reading |
| Copy selected | Creates independent identities; the originals remain |
| Copy associated bookmarks | Separate copy option, Off by default |
| Merge collections | Combines selected content into an existing/new destination after preview |

Move is appropriate when reorganizing one archive without creating another logical conversation. Copy is appropriate when you intentionally want independent entries. Copies can share immutable stored content internally, so duplicating an entry does not imply an immediate full physical duplication of every byte.

At a destination, identical source/content duplicates can be combined with their reading references. Differing content variants are retained; matching titles alone do not cause combination. Review the preview's selected, identical-duplicate and differing-variant counts.

## Move or copy step by step

1. Select the intended conversations.
2. Choose **Move selected** or **Copy selected**.
3. Choose an existing destination or the new-destination option and name it.
4. For a copy, decide whether associated bookmarks should be copied.
5. Review the preview and count.
6. Confirm the operation.
7. Browse the destination and check a few moved/copied records.

If the database changes between preview and execution, a stale preview is rejected. Refresh and review it again rather than attempting to confirm an obsolete count.

## Merge collections

Filter and select the source collections, then choose **Merge collections**. Pick an existing or new destination. The preview reports the scope and duplicate/variant behavior. An explicit option determines whether emptied source collections are removed. If you need a recovery point before a large merge, create a backup first.

Merging is not the same as importing a server collection again. It reorganizes the already retained Local archive and should not be used to solve missing source content or a bad provider parser result.

## Duplicate cleanup

Open **Settings → Database Manager → Deduplication**, the fifth manager section. Import and server-copy batches merge and deduplicate as records are processed, within their destination collection. Existing records can be checked with **Review conversation cleanup**, which reports analysis progress and shows each survivor/removal relationship and reason before confirmation. Closing the review returns to settings; use the same control to reopen an active review. If the archive changes after review, review it again; an outdated plan cannot be applied.

![Reader 1.4.5 cleanup review with synthetic candidates and separate database-size and free-space figures](../Images/reader-145-cleanup-review-light.png)

Collections remain independent unless you explicitly merge them. For providers with reliable export IDs, provider plus conversation ID identifies the conversation within a collection. Different IDs are separate conversations even when titles or user prompts match. Identical same-ID snapshots keep one copy; an exact longer continuation supersedes its shorter prefix. Other snapshots with the same reliable ID resolve to the newest record. VeniceAI exports have unreliable IDs and instead require content-based matching and preservation checks; a similar title alone is not enough. VeniceAI matching and Grok’s cross-ID cleanup rules preserve divergent answered branches, substantive reasoning, attachments and other meaningful structured content.

Grok/xAI single unanswered user prompts are removed automatically when they contain no substantive response, tool or reasoning content or attachments. A matching answered continuation can also supersede a Grok unanswered-ending variant. Provider identity controls this rule: a VeniceAI conversation using a Grok model remains VeniceAI. Superseded copies are deleted rather than kept as hidden conversation histories.

**Automatically delete single unanswered prompts from providers other than Grok/xAI** is optional and disabled by default. Enabling it requires explicit confirmation. Manual review can include this category without changing the automatic preference. It does not disable the built-in Grok/xAI rule, and it does not treat substantive content or attachments as an empty prompt artifact.

When conversations merge, relevant source metadata and saved reading references follow the survivor. Explicit prompt-only deletion has no survivor. The current conversation, list and derived indexes refresh after cleanup. Use [Storage & backup](backup.md) for a recovery point before broad archive maintenance. Cleanup applies to Reader’s Local archive; it does not delete server data or combine every collection into one.

## Delete a collection or conversation

A collection deletion can either remove the collection’s contents or move those contents into another collection before the source collection is removed. Review the destination and count in the confirmation. Deleting selected conversations removes their Local records and associated Local reading/search references. Deleting the currently open Local conversation clears the reader coherently.

Deletion does not remove your original export files, a separate external backup, or a conversation from ThreadShelf Server. Conversely, a separate Analytics capture is an independent copy and does not automatically disappear because a Reader record was deleted.

## Delete all Local data

This operation removes Local conversations, collections, search indexes, saved reading and internal recovery copies. It retains settings, window state, downloaded models, external backup files and separately scoped Server data/reading.

Use it only when you intend to reset the whole Local archive. It is not a troubleshooting substitute for clearing a disposable cache or rebuilding Activity. If the source archive cannot open, use the recovery controls and a validated backup instead of representing damaged storage as an empty database.

## Maintenance is separate from organization

Verify integrity, consolidation, compression conversion, index maintenance and backup/restore are described in [Storage and maintenance](storage.md) and [Backup and upgrades](backup.md). Their purpose is storage management, not editing the wording of conversation messages. Reader is an archive reader and organizer; this manager is not a chat-composition editor.