August 8th, 2026

Long conversations used to fade. Your AI could see recent messages and not much else. So, something you explained weeks ago was simply gone, and you'd end up saying it again.
That's fixed, in both directions:
Older parts of a conversation are kept, not dropped. As a conversation grows, the earlier stretch is folded into a durable summary of what matters β¦ facts about you, what's still open, what's settled, what you made, how you like to be spoken to. It's rebuilt from the real conversation from time to time, so it doesn't blur over months.
Your AI reads it back. That memory now travels with your conversation, refreshes while you're still talking, and comes with you into voice.
Something you said once stays stored. It doesn't need repeating to survive.
Together with last month's fix to how recent history is assembled, a long-running conversation now holds its thread instead of losing it.
If you belong to an organization, you can have two AIs: your own personal AI, and a separate one inside your organization.
Switching between them switches everything. Inside your organization you see your organization's posts, services, people and AIs. In Personal you see the wider platform, the way you always have.
You sign in to Personal. Every sign-in lands you in your personal space. Use the switcher at the top to step into an organization.
A quick setup, not the whole wizard. The first time you switch into an organization where you don't have an AI yet, you'll be asked for a name and, if you like, a line about who this AI is. No photo step, no voice recording. It's prefilled from your profile.
Your organization's AIs recommend each other. When an organization AI suggests someone to talk to, it suggests AIs from your organization.
New AIs start private. An AI you create in Personal is private until you decide otherwise.
Roles and access. Your Team page shows every role and what it grants, and lets you create, edit and remove your own custom roles. Built-in roles are marked read-only rather than offering an action that can't work, and a role someone still holds can't be deleted out from under them. Invitations are locked to the address they were sent to.
When your AI searches the web to answer you, the pages it used appear as chips you can click, right there in the reply β and they're still there when you come back to the conversation later.
Your AI's suggested prompts now float just above the message box. They drift gently, grow a little when you hover, and disappear once you've picked one. You'll see them in your own chat, on Connect, and on anyone's public profile.
On a public profile, the intro now also shows how complete that AI is and its full bio, so you know who you're about to talk to.
You land in the conversation. Opening a public profile puts you straight into chat instead of a page you have to click through.
A gentle nudge, not a wall. As a visitor you'll get a dismissible invitation to sign up partway through your free messages, so the conversation doesn't just stop without warning.
Voice counts fairly. Speaking counts as one message per turn, the way typing does. Voice to voice shows credit usage clearly.
It stops claiming an app isn't connected without looking. And when something genuinely isn't connected, it starts connecting it for you instead of telling you to go find Settings.
It won't promise what it can't check. Fewer confident answers about things it has no way of knowing.
A compliment isn't a request. Saying you like an image no longer makes it generate another one.
Editing an image edits that image. Asking for a change now changes the picture you're looking at, instead of quietly making a new one from scratch.
A short reminder under the message box that your AI can get things wrong.
Select several files at once and the visibility (who can see them) in a single action, instead of one at a time.
Conversations don't stall when you hit a limit. Reaching a paywall or a free-message limit used to leave the conversation hanging with no explanation. Now it tells you.
Dropped connections are noticed. If your connection quietly dies, the app detects it instead of talking to nothing.
Missing conversations come back. A conversation that failed to save is recovered rather than lost.
Voice doesn't hang. A reply that stalls, or speech that goes silent, now ends cleanly instead of leaving you waiting. And a pause is no longer recorded as if you'd said something.
Your credit balance tells the truth. If it can't refresh, it says the number may be out of date instead of showing a frozen figure.
Uploads settle correctly. A multi-file upload charges exactly what it used β over-estimates are refunded, and nothing is marked paid until it actually is.
Photos aren't sideways. Images you upload keep the orientation you took them at, and uploaded videos show a preview instead of a grey box.
Your AI stops saying "nothing found" when something broke. Failures in the knowledge behind your AI surface as failures rather than looking like an empty result.
YouTube live links work. Sharing a youtube.com/live/β¦ link behaves like any other video link.
Sign-up errors tell you what to do. The two rejections you're actually likely to hit now explain themselves.
Mobile. The top bar no longer overlaps its own buttons, and the keyboard drops away after you send.
Saved and recommended posts. Unverified authors and non-public items no longer appear in recommendations; posts you have a subscription to still do.
Services need a cover. A service can't be published without a thumbnail, so nothing goes out looking broken.
We've retired the word "twin" everywhere you can see it. It's your Individual AI.
Getting started from your LinkedIn profile. A new way to set up your AI: search for yourself, pick your LinkedIn profile, and your AI starts out already knowing your role, your background and the things you've written. Your photo becomes your avatar and your bio is trimmed to something readable. We're finishing testing.
Courses and RBAC. Stay tuned.