The last update was at the end of September. Since then we have shipped sixteen changes, some you will notice in the first conversation and some you will notice the first time something goes wrong. Here is the short version; each item has its own entry in the changelog with the details.
A progress line instead of "Thinking…"
An agent that looks things up takes a few seconds per lookup, and until now the visitor watched a fixed "Thinking…" for all of it. The chat page and the widget now show a short line that follows what the agent is doing: "Looking that up…", "Checking the documents…", "Reviewing what I found…", and after 15 seconds on one step, "Still on it — this one takes a little longer…". The lines come from a fixed list in every chat language, so a visitor never sees the name of one of your tools or what was sent to it. Why the wait exists at all, and what you can do to shorten it, is in why AI answers take a few seconds.
Polish and Croatian
Agents now reply in Polish and Croatian, which brings the count to 14 languages. Both languages run across the product: the chat page, the widget, WhatsApp and Telegram, the dashboard, the emails, read-aloud and dictation. The website has its own /pl and /hr versions, blog included. The launch post has the details.
Version history for agents, MCP servers and tools
Every change to an agent or an MCP server now leaves a version on a new History tab: who changed it, from where (the dashboard, the API, MCP or an import) and what changed. Compare any version with the live one, restore it in one click, and name the versions you want to keep for good. A restore is itself a version, so it can be undone too. Tools in the Tools library have the same history. More in change your agent without fear.
👍/👎 feedback that reaches you
The thumbs under each reply on the chat page and in the widget now save to your workspace, and you decide what each one asks. By default 👎 opens a short form with a category and a comment, and 👍 just records the vote. A new Feedback tab in the agent and MCP editors sets the form and shows what visitors said, with links into the conversations. Also in the API and MCP. How to put it to work: chatbot answer feedback.
Teams for hand-overs, and agents with no AI at all
Hand-overs to people now go to teams. An agent lists its teams and picks the right one from what the visitor asked, each team has its own opening hours and an optional backup team, and when everyone has reached their chat limit the customer waits in line and is told their place. On WhatsApp and Telegram a welcome menu lets people choose before they write. Teams live under Settings → Teams; the old Team tab is now called Members. The full walkthrough is in routing hand-overs to the right team.
For conversations that should never be answered by AI, there are now person agents: a chat page and channels your team answers from the Inbox, with no AI wording anywhere on the page. You can limit who may chat to listed email addresses and domains, so a staff-only line is a setting, not a project. Both are in the API and MCP. When this is the right call: chat without AI.
The Inbox, upgraded
- Messages arrive the moment they are sent, in both directions, instead of on the next refresh.
- A reply editor that works like Gmail: bold, italic, links and lists as you see them, with Ctrl+B, Ctrl+I and Ctrl+K.
- Resizable panes, sounds for new conversations and messages, and a settings cog that also chooses what each card shows.
- A full-screen Inbox with a collapsible list, a compact Filters button, and counts as small badges on each view.
- A Spam folder where junk conversations are filed, out of every other view and out of the counts.
- The transcript now shows the whole answer: the cards and videos a reply showed, the sources it used, and when it sent a payment link, started a browser task or sent a verification code.
Photos, documents and voice notes
Visitors can send photos, documents and voice notes on the chat page and in the widget, with a record button and a waveform player. On person agents your team sees them in the Inbox, and the team can reply with files and voice notes to the web chat, the widget, WhatsApp and Telegram. AI agents get voice notes transcribed before they answer. More in voice notes and files in AI chat.
Verify who you are talking to
An agent can now confirm a visitor's mobile number or email address with a six-digit code, sent by WhatsApp, SMS or email, before a tool acts on that person's account. A phone the conversation has already proven skips the second code, the code box survives a page reload, and the verification card in the dashboard was redesigned around the checks you turn on. The how-to: verifying chat visitors.
More control on WhatsApp
- The cost control's answer window is now set in minutes, not hours.
- A new cap on answers per conversation, so a long back-and-forth moves to the web chat at the point you choose.
- While a person from your team is answering, the cost control stands aside and counts nothing.
- The link to continue on the web goes out as a button instead of a long address.
- The Channels tab is a grid of WhatsApp and Telegram, each with its own page and its own welcome menu.
Knowledge and MCP servers
Long documents such as laws, regulations and manuals can be split into sections at their headings (beta), so the agent reads the exact section it needs instead of a loose passage. Every knowledge source can now be edited in place, including its stored text, instead of being deleted and added again. And hosted MCP servers introduce themselves properly to MCP clients, with their own name, a description and instructions, instead of a technical identifier.
The chat page, polished
- A sent message glides to the top and the reply writes itself below it.
- On phones the composer docks at the bottom edge, like the chat apps people already use.
- The sidebar is resizable, long titles scroll on hover, and a title shimmers while its reply is being written.
- Dots in the sidebar show which chats are replying and which have a reply you have not read, and "Thinking…" is still there when you switch away and come back.
- Read aloud starts in about a second, and the disclaimer under the composer is shorter.
Smaller things
- The dashboard is available in English, Portuguese and Spanish, and since 26 October in Polish and Croatian too. Pick yours on the Profile page.
- Each agent can have its own chat page background, separate from the widget's cover image.
- Custom domains show an honest status: a live check of each DNS record, the reason in plain words, and correct instructions for root domains.
- Agents answering in Portuguese use European Portuguese unless the visitor clearly writes Brazilian Portuguese.
- New agents start with search-engine indexing off; turn it on when the agent is ready to be found.
- The Tools tab cards are accordions whose switches save at once, and HTTP tools handle large JSON responses without cutting them in the middle.
Everything here is live today. The changelog has one entry per item, and that is where we keep writing these up.
Frequently asked questions
Do I need to turn any of this on?
Most of it works straight away: the progress line, the Inbox changes, the chat page polish and version history need nothing from you. Teams, person agents, visitor verification and the WhatsApp caps are settings you choose per agent, and the 👍/👎 form starts with sensible defaults you can change on the Feedback tab.
Which languages does SuperCognit support now?
Agents reply in 14 languages, Polish and Croatian being the newest. The dashboard is available in English, Portuguese, Spanish, Polish and Croatian, and the website in the same five. See Polish and Croatian.
Can I undo a change to my agent?
Yes. The History tab keeps every change, lets you compare any version with the live one and restore it. A restore is recorded as a new version, so it can be undone as well. See agent version history.
Where do I see what changed each month?
The changelog has one entry per change, with the details, and these roundups summarise each period. The previous one is what's new in late September.
