Roadmap

v3.1.3

Features and improvements shipped in recent UnoPim releases — explore what's new.

1. Maximum Output Tokens Setting: Raise the Magic AI generation ceiling for the whole installation from configuration, instead of editing prompts one at a time.
2. Concentrate AI Provider: Added Concentrate AI to the supported Magic AI providers.
3. Honest Truncation Reporting: A generation that stops because it ran out of tokens is reported as such instead of silently returning half a sentence, and a fragment cut mid-tag is dropped rather than rendered as stray text.
4. Endpoint-Aware Model Discovery: Magic AI model discovery runs against the platform own base URL, so a proxy or regional endpoint lists the models it will actually serve; a custom endpoint typed without a version segment is retried with one.
5. Azure Deployment Names: An Azure platform now addresses a model by its configured deployment name, which is what the Azure endpoint expects.
6. Clearer Model Fetch Errors: A failed model fetch reports the host and status instead of rendering the upstream error page verbatim.
7. Reworked Translation Dialog: Source channel and locale are named separately from the targets, the attribute picker shows the value each attribute currently holds, and the dialog warns before replacing existing content in the target locales.
8. Extensible Export Profiles: Added view events to the export profile create and edit pages, so a package can render additional filters in the output section.
9. No Infrastructure Detail In Chat Errors: Fixed an AI chat failure originating in the database surfacing the connection host, schema and full SQL in the chat window; an infrastructure failure now returns a generic message.
10. Attribute Codes In JSON Queries: Fixed a product attribute whose code is not identifier-shaped — starting with a digit, or containing a hyphen or a space — breaking every MySQL JSON path query it appeared in.
11. Translatable Fields Only: Fixed the translate action offering attributes it cannot translate; a field is offered only when it is locale-scoped and holds a value in the source scope.
12. Permission-Aware Translate Button: Fixed the AI translate button appearing on a field the current user cannot edit.
13. Dialog Stacking Fix: Fixed a confirmation dialog rendering behind the element that opened it.
14. File Picker Clearing: Fixed clearing a file picker being treated as a new file to scan.
15. Validated Generation Endpoints: Fixed the product attribute and content generation endpoints accepting an unvalidated channel, locale and resource reference.
16. AI Translate Toggle Saves: Fixed unticking AI translate on an attribute not saving, because the form fallback value overrode the checkbox on submit.
17. Clearing A Channel Name On PostgreSQL: Fixed clearing a channel name failing with a not-null violation; the translated name is now nullable and the channel grid falls back to the default locale name for a cleared one.
18. Gallery Removal Unsaved Badge: Fixed removing a gallery image not showing the Unsaved badge on the product edit page.
19. Import Filters Survive Re-Render: Fixed import filter selections being discarded on every re-render, which left a filter depending on another field with an empty dropdown.
20. Price Input Alignment: Fixed a price attribute currency inputs stretching to the height of a neighbouring currency validation message.
21. Queue Worker Timeout: Fixed the data transfer queue worker leaving its timeout alarm armed after the queue drained, so an in-process worker was killed once the timeout elapsed; the timeout handler also matches the current Laravel signature.
22. Remove Model Button Styling: Rebuilt the admin assets so the remove-model button turns red on hover.
23. Magic AI Platforms From The CLI: New commands to list, add, edit, delete and set a default Magic AI platform from the server, with an interactive wizard and unattended options.
24. Encrypted CLI Credentials: A platform added from the CLI stores its API key encrypted in the database rather than in the environment file, and can read the key from standard input.
25. Managed Platforms: A platform saved from the CLI is read-only in the admin panel apart from its status and default setting, and cannot be deleted. Other platforms stay unrestricted.
26. Higher Default Generation Ceiling: The default output ceiling moves from 1024 to 4096 tokens, so a table-heavy description is no longer truncated mid-tag. A tuned ceiling is left alone.
27. Smarter Model Recommendations: Models are ranked by release date, so a fetch pre-selects the newest text and image models per family instead of legacy variants that sorted first alphabetically.
28. Pinned Model Discovery Requests: Model discovery for OpenAI, Anthropic and Gemini is pinned to the validated address and no longer follows redirects when a platform supplies its own base URL.
29. Scope-Correct AI Translation: A translated value now lands in the bucket its attribute scope dictates, so a locale-only attribute is no longer saved where nothing would read it.
30. Correct Image Generation Platform: AI image generation now uses the platform and model configured for image generation instead of the conversation text platform.
31. Claude Model Compatibility: Content generation no longer fails on Claude models that reject the temperature parameter, including Opus and Sonnet 4.7 and later and the Claude 5 family.
32. Select All Matching Beyond 10,000: Select all matching now covers every matching record instead of stopping at the first 10,000.
33. Queued Bulk Product Actions: Status update and delete on a select-all selection now run on the queue in chunks of 100, so the request returns at once at any size and the admin is notified with the product count when it finishes.
34. Product Type And Variant Level Enums: Added ProductType and VariantLevel enums for product types and variant levels. The existing string constants keep their values, so code comparing against them keeps working.
35. Chat Import Links To Job Tracker: The AI chat import confirmation now links to the job tracker instead of the import profile.
36. Mass Action Toolbar Fix: Fixed the datagrid toolbar unmounting while a mass action was pending, which left an empty error flash and a stuck loading shimmer after the grid reloaded.
37. Chat Product Translation Fix: Fixed every Agentic PIM chat product update that ran a translation failing with "Unknown column name"; target locale names are now derived from their codes.
38. Plain Text Translation Fix: Fixed Magic AI translation wrapping plain text attributes and non-WYSIWYG textareas, such as the name or meta title, in HTML paragraph tags; only WYSIWYG fields keep HTML now.
39. Smarter Chat Page Reload: Fixed the AI chat reloading the product or category edit page after a reply from a read-only tool; the page reloads only when a write tool changed data.
40. Per-Conversation Chat Uploads: AI chat images are kept per admin and per conversation and expire after ten minutes, so an image from an earlier conversation or another admin is never attached to a new message.
41. Clear Image Support Errors: A model that cannot read images no longer blocks every following chat message; the chat explains the model does not accept images and drops the image from the conversation.
42. Upload Limit Message: An oversized chat image now fails with an explanation that states the server upload limit.
43. Select All Matching Everywhere: Select all matching now also works for the mass actions on attributes, association types, category fields, currencies, locales, completeness settings, measurement families, product passports and webhooks.