Content memory
The content memory holds the approved wording a workspace reuses. Each entry pairs a source segment with a target segment, and carries where and under what context it was approved, so a run can recycle what the workspace has already settled instead of drafting it again. It is the same store the engine keeps locally, held on the server instead, shared across every project in the workspace.
The narration uses terminology that has since been retired. The interface and the behaviour shown are unchanged.
Opening the content memory
There are two ways to open the content memory browser:
- From a project: click the Content memory button in the project view header. This opens the browser with the project's source and target locales pre-selected.
- From the sidebar: click Content memory in the main sidebar to browse the workspace-scoped content memory without locale pre-filtering.
Browsing entries
The browser displays entries in a paginated table with these columns:
| Column | Description |
|---|---|
| Source | The source language text |
| Target | The target language text |
| Source Locale | Language code of the source text |
| Target Locale | Language code of the target text |
| Updated | Date the entry was last modified |
| Actions | Edit and Delete buttons |
The entry count badge in the header shows the total number of entries matching current filters.
Searching
Type in the search box to filter entries by source or target text. Results update as you type.
Filtering by locale
Use the locale filter dropdowns to narrow entries:
- Source locale filter: show only entries with a specific source language
- Target locale filter: show only entries with a specific target language
Filters can be combined with text search. Changing a filter resets to page 1.
Adding entries
- Click Add Entry in the header
- Fill in the form:
- Source text: the original language text
- Target text: the approved wording
- Source locale: select from available project locales
- Target locale: select from available project locales
- Click Add to save, or Cancel to discard
Editing entries
- Click Edit on an entry row
- The target text becomes an inline text input
- Modify the text
- Click Save to confirm, or Cancel to discard changes
Only the target text can be edited inline. To change the source text or locales, delete the entry and create a new one.
Deleting entries
Click Delete on an entry row. The entry is removed immediately and the table refreshes.
Pagination
When there are more than 50 entries, pagination controls appear below the table:
- Previous / Next buttons to navigate pages
- Page X of Y indicator showing current position
Content memory in editing and review
The content memory also assists work outside the browser:
- Context panel: the editor shows per-block matches with scores and match types, with one-click apply.
- Bulk recycle: the editor's file-wide operations pre-fill targets from the content memory across the whole file; exact and high-confidence fuzzy matches land as drafts for review.
- In a run: every run recycles from memory before it drafts, so approved wording is reused before any credit is spent.
Match scores
Each match receives a score from 0 to 100% representing how closely the source text matches the stored entry. Matches are colour-coded:
- Green (100%): exact match
- Yellow (70–99%): fuzzy match
- Lower scores appear with reduced emphasis
Bowrain's matching accounts for inline formatting differences, so a source segment and a stored entry can match even if their tag structure differs slightly.
Governed version reuse
An entry is more than a pair of strings. Successive answers for one block are linked as a version chain, and each carries a fingerprint of the context it was approved under: the point, the voice, the terms in force. When a block's source changes, the prior answer is offered to the model only as a reference, only under the fingerprint it was approved under, and the reuse decision reads whether the words actually moved (none, cosmetic, or substantive) rather than a similarity score alone. A translation can be asked to reuse the prior version or to start clean. The chain and the fingerprint travel with the entry, so a memory exported from one workspace keeps its provenance in the next.
For how the underlying engine matches across projects, see the framework's content memory reference.