Prodgy Overview

Current Version current In development

Changelog (unreleased) History

PBIs

Legacy Assistant portal removed after migration completion

New The legacy built-in dashboard (PBI 79865) has been removed from the Assistant application. All requests to /dashboard and its sub-paths are now redirected to the new integrated administration panel at /admin/queues. Existing bookmarks and external links continue to work automatically. The deployment pipeline has been updated to correctly inject PRODGY_APP_URL into the Windows Service, ensuring the redirect works in all environments.

Reliable Prodgy Assistant meeting queue (never lose a scheduled meeting)

New The meetings and recurring patterns the Prodgy Assistant must join are now stored in the database as the source of truth, so they are never lost after a restart or instability. The superadmin Queues screen reflects the real state in real time — scheduled meetings, calls in progress and recurring patterns, including calls started on the fly via @Prodgy join. When an item gets out of sync with the queue, it is flagged on the screen and a superadmin can re-send that item or run a full reconciliation, all without restarting the service.

Bugs

Multiple fixes in Knowledge Storage metadata and interface

Bugs Fixed several issues in the Knowledge Storage upload form and document info modal: the metadata accordion no longer collapses automatically during Magic Fill; metadata fields (source, subject, language, tags) are editable after Magic Fill fills them; the filename in the Details tab now correctly renders special characters (e.g. “Ações” instead of “Ações”); the Quality tab now shows the model, provider and processed_at fields in the Processing block; Magic Fill now detects the document’s predominant language from the actual file content instead of using the user’s profile language; and the quick filter labels (“All”, “Type”, “Facets”) are now correctly translated when the interface is set to English.

Magic Fill facets reliability improvements

Bug Fixed two issues affecting the Magic Fill facets field on the document import form: (1) the auto-trigger no longer fires before the facets vocabulary has finished loading, preventing empty or missing facet suggestions on the first upload; (2) facet values returned by the AI are now matched case-insensitively against the controlled vocabulary, so valid suggestions are no longer silently discarded when the AI returns a casing that differs from the vocabulary.

Overview magic generation now uses existing Knowledge Base content

Bug Fixed an issue where the magic (✨ Sparkles) button on the Knowledge Base Overview tab generated product content without considering the documents already indexed in the base. The generation now retrieves relevant content from the active Knowledge Base (semantic search) and uses it as context, producing summaries that reflect the product’s real knowledge instead of generic content based only on name, description, and website. When there is no active base or retrieval fails, generation keeps working normally, and any failures are now logged for diagnosis.

AI section generation dialog now shows templates correctly

Bug Fixed an issue where the “Generate Section with AI” dialog in the Knowledge Base Overview did not display available prompt templates. The dialog now loads templates from the configured settings on open, allows selecting a template to pre-fill the prompt, and exposes max tokens and temperature controls so users can fine-tune the generation. The API was also extended to accept and validate these parameters end-to-end.

---