What’s new? v4.47. Undo for task changes, and more..

Discover the improvements we’ve built into each release to make Iurefficient a more powerful, intelligent, and secure tool for your professional practice. v4.47.6 — The assistant now sees tickets, inventory, and billing “Which project has the most tickets?” was a question with no possible answer. The assistant answered by looking at tasks and…

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

v4.47.6 — The assistant now sees tickets, inventory, and billing

“Which project has the most tickets?” was a question with no possible answer. The assistant answered by looking at tasks and documents, because data from the add-ons never reached it: the mechanism to query them existed, but it was only used on a path the application had stopped using. Once it was reconnected, the assistant suddenly started seeing what six add-ons were already providing —CRM, collaboration, tax, ERP, personal, and knowledge graph— without anyone having noticed it wasn’t getting through.

And three add-ons that contributed nothing now do. Inside a project, the assistant knows its tickets with status and priority, the assigned equipment with make, model, and serial number, and its invoices with balances. In the general chat it can compare across projects: which ones have the most tickets, how inventory is distributed by type and status, which warranties and licenses expire in the next 60 days, how much is receivable, and who owes the most. The projects it mentions appear as links, so you can open them right from the answer.

No migrations; update backend and frontend together.

v4.47.5 — Typing in the chat is instant again

In a long conversation, typing became torture. Each letter took a while to appear, as if the system had to think about it before letting you continue. In a freshly started thread you didn’t notice; with forty messages on screen it was unusable. The cause: every keystroke redrew the entire conversation and reprocessed the formatting of every message —tables, lists, formulas, code—, and all that work happened inside the key handler itself. Measured with forty messages: from 113 milliseconds per keystroke down to 15. It wasn’t a new problem; it had been around for a while and only became obvious in long conversations. The improvement also reaches other screens that display formatted text.

New toggle: conversation or deep analysis. It’s in both chats, in the same place. By default the assistant responds with the fast model, which is more than enough for asking questions and chatting, and starts writing within three seconds. When you turn on deep analysis, the most capable model responds: it takes longer to start because it reasons before writing, and it’s what you want for a forty-page contract or a demanding question. The choice is remembered in each browser, so anyone who works with long documents can leave it on and forget about it.

Two bugs from the streaming launch, fixed: typing while the assistant was responding was extremely slow —the incoming text forced the chat to repaint dozens of times per second— and the answer ended up being displayed several times in a row, even though only one remained in the history. In addition, the statistics card for the general knowledge index shows data again: it had never shown any on any instance.

v4.47.3 — The chat responds before your eyes, and you can choose a faster model

Asking the assistant something meant staring at a blank screen for fifteen to thirty seconds, with no way of knowing whether it was working or how much longer it would take. This release tackles that wait from both sides.

The answer is written before your eyes. The text appears as the assistant drafts it —in the project chat and in the general chat— instead of appearing all at once at the end. If you close the tab and come back, the answer is still there: nothing is lost. What you see while it’s being written is a preview; when it finishes, the system replaces it with the final version, which is the one that includes the action cards and the verified citations.

Configurable chat model. You can now use a fast model for conversation and reserve the powerful one for drafting documents and analysis. Measured with the same question and the same case file: the first word goes from 13.7 to 3.2 seconds, and the complete answer from 17.3 to 5.4. It’s an optional setting that is turned on and off from the Central system without touching anything else.

The “what the AI is doing” notice is finally visible, and it says something useful: “Reviewing 12 project documents”, “8 sources found · drafting with…”, with a timer. It had existed since June but never actually showed up because of a configuration error — the same one that had disabled the lock that prevents two simultaneous queries from rebuilding the same index at the same time.

v4.47.2 — Documents are indexed again, and the chat can now read them

A silent failure that had been affecting every instance since July 9: uploaded documents never finished processing. The file was saved and showed up in the list, but internally it was left half-done — it wasn’t indexed for AI search, text recognition didn’t start on scanned files, and it didn’t make it into the general knowledge base. None of this triggered any warning: the document seemed perfectly fine. It’s now fixed, and the affected documents repair themselves the first time their project is queried from the chat.

The general chat now reads what you attach. Before, it replied “the document is still being indexed” for files uploaded hours earlier, and there was no way to get it out of that state. You can also analyze a document without having assigned it to a project: before, the system required you to create one first, even though the general chat already knew how to do it.

Details you notice every day: the general chat now shows the sources for each answer —it always cited where each piece of data came from, but there was no button to open the details—; a web address pasted into a project chat is clickable again; long messages are no longer cut off in the middle of a table when collapsed; and accents were corrected in 109 texts spread across the application, the welcome tutorial, and the add-ons (“Busqueda”, “Informacion”, “Ano Fiscal”).

v4.47.1 — Spreadsheets, the way they’re meant to be read

The chat accepts Excel again. When you tried to attach a spreadsheet with the paperclip, the file appeared grayed out and couldn’t be selected; no error message appeared because it was never actually sent. Dragging it into the conversation did work, which made the problem even more baffling. Now the paperclip offers exactly the file types the administrator has enabled in Settings › Documents, and changing them there takes effect immediately.

Excel files open as spreadsheets, not as printouts. The system used to show an image of how the file would look printed: cut off across the width, split into pages, with columns getting lost. Now they open in the spreadsheet viewer, with their grid and their tabs, and the “Edit” button is still where it was. Word files and presentations keep the quick preview, which does reflect the document for them. Opening a spreadsheet takes a few seconds longer, because the editor starts up; in exchange, you see it in full.

.xlsm and .xlsb files now open. They could be uploaded, but when opened the system treated them as Word documents and they couldn’t be viewed — in the viewer, in links shared with third parties, in the client portal, and in templates. .xlsm files can also be edited.

v4.47.0 — Undo in Tasks, and an import that no longer deletes what it doesn’t mention

In Tasks, almost everything was permanent. You’d import a file that overwrote existing tasks, change the status of forty at once, delete a phase with its twenty subtasks, or drag something onto the wrong parent, and there was no going back: you had to rebuild it by hand. Now, after any of those operations, a notice appears with an Undo button. If you close it, the operation is still available: the history is kept for 30 days, and you can also redo what you’ve undone.

Undo respects other people’s work. It isn’t a blind rollback: if, between the operation and your change of heart, someone else modified those same fields, those tasks are left as they are and you’re told which ones they were.

Deleted tasks are no longer lost. They go to a trash bin —their logged hours, expenses, and comments still point to them— and restoring them brings back the complete tree with its subtasks and hierarchy. Along the way, the delete warning now tells the truth: before, it promised to “delete 2 tasks” while taking down the thirty that hung from them.

Importing a CSV no longer deletes what the file doesn’t mention. A file with four columns would wipe out the description, leave tasks without an owner, delete estimated hours, and flatten the hierarchy. Now only what the file contains is written. And reimporting an MS Project plan no longer duplicates it: the corrected plan is reconciled with the tasks that already exist, even if it came out of Iurefficient, was worked on in Project, and came back.

chat Improvements fast Tasks undo

← Back to the index