v2.7.20: Read What Your Employee Remembers, Calls That Know the Conversation, and a Browser Share That Stops Expiring

Your employee has been building a memory for months. Everything you told it, every decision you made together, every correction. Until now the only way to see any of it was to ask, one question at a time, and hope it surfaced the right thing.

That changes in this release.

Memories: read and manage what your employee remembers

There is a new Memories tab in the chat sidebar, sitting between Files and Kanban. It opens the whole memory as a board: every entry on the left, newest first, grouped by the day it was written, with the one you select opening on the right at a comfortable reading width.

There are two ways to search, because they answer different questions. The plain search matches words. The semantic search matches meaning, so you can describe what you are looking for in your own words and it ranks entries by how close they actually are. That is the same search your employee runs on itself, now pointed at the screen instead.

You can add a memory, edit one, or delete one. Editing re-indexes it, so a corrected memory is found on its new wording rather than the old. Deleting is permanent and asks first.

Underneath whatever you are reading sit its details and then Related memories: the five closest entries, one click away. It answers the question that reading a memory naturally raises, which is what else does it know about this.

On who can see it. The owner of the employee and instance admins. Nobody else, and that deliberately includes people the employee has been shared with. A memory holds things said by everyone who has ever talked to that employee, so being able to talk to it does not entitle you to read what it remembers about other people. This is enforced properly, not by hiding a tab.

Memories is an Enterprise feature and is off on the free tier.

Calls now know what you were just talking about

An in-app call used to begin cold, with no idea what you had been discussing thirty seconds earlier in the chat.

Calls now open carrying the last five exchanges of the conversation you are actually in. You can stop typing, start talking, and continue the same thought.

Your call rules, in front of you and editable

The call screen used to show Language, Voice, and a large empty space. That space now shows the rules that employee is about to be given, verbatim, so what governs the call is visible before you dial rather than buried in settings. Employees without any rules say so plainly.

Better, it is now a field you can type in, with a Save button. Previously those rules could only be edited from the phone settings, and only for an employee that had a phone number, which on a typical instance meant one employee out of sixty-four. Every employee can have call rules now, phone number or not.

Ask your employee to write its own call rules

Next to Save there is a new Enhance prompt button. It hands your current rules to the employee and asks it to improve them. The employee reads them, thinks about what it actually does and who calls it, and comes back with a proposal, better phrased and filling gaps you had not considered. Approve it and it saves them itself.

It proposes rather than rewrites, on purpose. Nothing changes your call script without you seeing it first.

With an empty box, it will write you a first version from scratch.

The empty message box now offers a call

With nothing typed, the Send button used to sit there greyed out doing nothing at all. It is now a call button: one click opens the live call screen for that employee, with language and voice ready to choose. Type anything and it turns back into Send.

Sharing your browser stops expiring every week

If you share your browser with an employee, that connection used to quietly stop working after seven days. Everything would go silent and the only fix was to share it again by hand.

It now renews itself. The connection refreshes automatically in the background, and its lifetime goes from 30 days to 90. You should not have to think about it again.

One practical note: this arrives through the Chrome Web Store, so it reaches you on the next extension publish, and employees already set up need one final re-share to get started on the new system. After that, never again.

Connectors: Edit means edit

To change an employee's permissions on a Telegram or Discord bot you had to press Assign, which is not what that word means, and it always reopened the first assignment, so with several employees on one bot the others could not be edited at all.

Every assigned employee now has its own Edit button that opens on that employee. Assign only appears when a bot has nobody yet. Assignment rows also tell the truth now, showing "3 allowed chats" when a list is restricted instead of looking identical to an unconfigured one.

Telegram permissions can also be changed on the free tier again, which had been blocked by a shared dependency on a paid connector.

Kanban: better attachments, readable cards

The board advertised images and quietly accepted anything at all. It now accepts one clear list, matching exactly what your employee can already attach: png, jpg, jpeg, gif, webp, svg, bmp, pdf, txt, md, json, csv, mp4, webm, mp3 and wav. Anything else is refused before it lands, with a message telling you what is allowed instead of a bare error.

Cards your employee writes are also readable now. They used to arrive with raw formatting marks scattered through them, or as one unbroken paragraph. Headings, lists, bold and links now render properly, and the skill guidance was rewritten to describe what a good card looks like rather than how to mark one up.

Existing employees pick this up with Update skill at the top of the board.

Free tier: an idle mailbox no longer wakes your employee

On the free tier, connecting a Gmail or Outlook account meant your employee woke up on every incoming message, which is the fastest way to burn a free allowance on nothing.

Automatic reception is now off there. Your employee still reads and sends mail whenever you ask it to; it simply stops starting work on its own because something landed in the inbox.

Also in this release

• Outlook replies keep their line breaks and actually send to the people you put in copy
• Work started by an inbound message, a scheduled task or a webhook now shows in the chat while it happens, instead of the conversation looking idle
• The settings panels fit on a tablet screen properly
• Coming back from a file returns you to where you were scrolled, not the top
• Nova now lives in the dashboard only, rather than in two places
• The Chrome extension no longer shows your own message twice

Update and go and read what your employee remembers about you. It is a stranger feeling than you expect.