What’s new? v4.48. Transcription improvements and assistant persistence.

Discover the improvements we’ve added in each release to make Iurefficient a more powerful, intelligent and secure tool for your professional practice. v4.48.1 — What the assistant writes is no longer lost You asked it for minutes, it wrote them in full and nothing arrived. Neither the card to save them nor the text: on…

Discover the improvements we’ve added in each release to make Iurefficient a more powerful, intelligent and secure tool for your professional practice.

v4.48.1 — What the assistant writes is no longer lost

You asked it for minutes, it wrote them in full and nothing arrived. Neither the card to save them nor the text: the screen showed “I’ve prepared the executive minutes…” and that was the end of it. The document travels to your browser inside a format that forces the model to escape by hand every quotation mark in a ten-thousand-character text, and a single stray quotation mark was enough to invalidate it. Worse: since it couldn’t read it, the screen deleted it from the message. Ten thousand characters of work vanished without a trace or a warning, and the longer the document, the more likely it was. Now the server repairs it before delivering it, and whatever can’t be repaired is shown in full with a warning instead of disappearing.

In the CRM you can now link a document that already exists. On a lead or an opportunity you could only upload a new file or create a blank one; the most common thing was missing: linking the proposal, the annex or the spec sheet that’s already in the system. The search box offers you recent documents as soon as you open it and finds them by name or by content. Only the link is created —the file remains the same, with its versions and permissions, and it can be linked to several records at once— and removing it from the record doesn’t delete anything.

Includes everything in 4.48.0. No migrations; update backend and frontend together.

v4.48.0 — Transcriptions now exist for the assistant

You transcribed a meeting, asked for the minutes, and the assistant replied that the file “has no extracted text yet.” It was right: the text was saved and visible on screen, but none of the paths that feed the assistant looked at it. The text of an audio or video file lives in its own field —separate from the one for written documents and the one for scanned ones— and the assistant’s search never read it, so a fifty-thousand-character transcription contributed exactly zero. Now it goes in through all four paths, and the transcriptions you already had are added automatically: the general index is rebuilt twice a day.

A document you attach to the chat is read in full. Previously the assistant was handed loose fragments and only read the whole thing when the search found nothing, with the paradox that indexing a document made the answer worse: when you asked for minutes, a single fragment arrived. Attaching means “use this document,” and now that’s what happens.

Transcribing long audio no longer fails at two hours. A 54-minute meeting needs about four hours of processing, and the previous limit aborted jobs that were running perfectly; worse still, they were requeued and competed with each other for the same service. Progress bars frozen at 85% are also a thing of the past.

Web research searches for the topic, not the sentence. Questions like “what do you think of the agreement I uploaded?” were sent to the search engine as is, references to documents included. Now the terms are extracted.

And deleting a document really removes it from the index. Because of an old bug, removing it from the general index always failed silently: its text remained searchable and every reindex stacked up one more copy.

No migrations; backend only, although backend and frontend are updated together.

← Back to the index