What’s new · Your AI Twin

Beta 0.4

Your AI Twin is in its beta phase. This page lists what has changed for you — new things you can do, and problems that have been fixed. Technical release notes for testers live in the project's CHANGELOG.md.

6 November 2026

Creating a partner access key works now. Since the feature arrived three days ago, creating a key was refused with "your plan does not state an access key allowance". The allowance was genuinely missing from plans that already existed, and the message was wrong about why — for most organisations there was no plan on file at all, which needs a different fix than adjusting a number. Plans now carry the allowance, and the message says which of the two is the case.

3 November 2026

You can now create your own partner access keys. Under Partner Access Keys you issue the key your servers use to open learner sessions — the one that identifies your organisation to the platform. Until now only we could create it, you could only ever have one, and creating a replacement destroyed the old one the moment it was made. You can now hold several at once, so moving an integration to a new key no longer means an interruption.
Every key carries a name you chose. Give it a label when you create it, and the list tells your keys apart at a glance — alongside who created it, when it was last used, and the first characters you will recognise from your own configuration.
A key can be limited to one tenant. Choose whether it reaches a single tenant of your organisation or all of them, including tenants you create later. A key limited to one tenant is refused for every other tenant of your organisation, on every single request — not only when the session is opened.
A key can be given an end date. 30 days, 90 days, a year, or never. Before you create it, the page shows the exact date it would stop working.
If a key is refused, you are told which of three things happened — we do not know this key, it has been switched off, or it has expired. Each needs a different thing from you, so each now says so.
The key itself is shown once, when you create it. It cannot be looked up again afterwards. If you lose it, delete that key and issue a new one.

Keys you already had keep working, unchanged. You do not need to do anything.

27 October 2026

All API keys of your organisation on one page. Under Twin API Keys you now see every key across all your Twins at once — who created it, which Twin it belongs to and when it was last used. You can search by Twin, by colleague or by the name you gave the key, sort by last use, and revoke a key straight from the list. Before revoking, the page tells you what that costs: every integration using that key stops working immediately.
Managing a colleague's keys no longer sends you to another portal. From the overview you reach the key section of each Twin without leaving the Partner area.
If we ever have to block one of your keys, you will always be told why. A reason is now mandatory for us, so a blocked key never shows up in your list without one — you no longer have to guess at a dead integration.
Every key now says which kind it is and what it is for. "API key" stood for four different things: the key that identifies your organisation, the key for calling a Twin from your own server, and the two keys for the chat widget. Each is named, explained in one line, and shown with the prefix you will see in your own configuration — so you can tell at a glance which one you are holding, and which one you need. Two of them must never be put in a browser and one is public on purpose; the screen now says so.

26 October 2026

API keys of a shared Twin are now visible to everyone who may manage it. If a colleague had created a key, the list showed nothing at all — no keys, no message — and the counter next to it read zero while creating a new key was refused with "maximum reached". The list now shows every key of the Twin, says who created each one, and the counter and the list always agree.
Revoking a key that a colleague created now works. It used to report a technical error while the key stayed valid. A key that is already revoked now says so, and a key belonging to a connected channel points you to the channel instead of being switched off silently.
Keys created automatically for a channel no longer use up your allowance. Four connected channels used to leave you room for a single key of your own.
If your plan does not include API access, you are told so before creating a key, together with the plan you would need — instead of receiving a key that is refused on first use.

12 October 2026

You can now see how a document re-analysis turned out. While a document was being re-analysed, a "Reprocessing" note appeared and later simply vanished — whether the run had finished properly or had been given up on. The document now keeps a line showing when it was last re-analysed and how many pages were processed, or, if it did not work, what went wrong and what you can do about it. If you are on the page when a re-analysis finishes, a short message tells you right away.
An error message about a form now says which entry is missing. If something required was left out, the message said only that an entry was missing, without naming it. It now names the field, so you can see at a glance what to fill in.

9 October 2026

Deleting your account now removes your login as well. If you had ever approved or set up something for other people, parts of your account could stay behind — your email address among them — while the deletion was already reported as finished. Everything is now removed together, and a deletion that cannot be completed is no longer shown as done.

2 October 2026

You are asked for your consent once. The terms and privacy policy you accept when you register are stored with your account, so you are no longer asked to accept them again after your first sign-in. A registration link from the waitlist can be used once only.
Resetting your password, joining the waitlist and requesting a new registration link now have the same short check. It runs on its own; the emails and their links work as before.
Registration now checks briefly that you are a person. Below the consent boxes you see a short check that runs on its own — there is nothing to click or solve. It needs no cookies and sends nothing to a third party. If you submit before it has finished, the form tells you to wait a moment. The confirmation email now comes directly from Your AI Twin.

1 October 2026

A document no longer uses up its pages twice. If processing a document was interrupted and restarted automatically, its pages could be counted against your monthly page quota a second time — on a trial this could block your next upload. Each document is now counted once. Choosing Re-analyze Document still counts its pages again, as the confirmation says.

28 September 2026

The question analytics and the usage heatmap now open for your organisation's Twins, not only for the ones you created yourself. Both refused a colleague's Twin with an access error, even though the Twin list showed it and everything else about it could be opened.

23 September 2026

New Twins in an organisation now belong to a customer account from the start. When you create a Twin for your organisation, you choose the customer account it belongs to — if there is only one, it is already selected. Usage is then counted for the right account from the first question on. If your organisation has no customer account yet, you are asked to create one first. Nothing changes if you use the platform on your own.
A Twin you have not published can now be used through your organisation's integration. If your Twin is assigned to one of your organisation's customer accounts, it can answer questions there even when its visibility is set to private or restricted. Previously it had to be made public to everyone first. Nothing changes for the chat window or the embedded chat on a website: there, a Twin that is not published stays unreachable, as before.

22 September 2026

Document details can now be edited. Title, author, language and the other details a document carries can be corrected after upload, directly in the portal — the document does not have to be uploaded again.
Deleting an account now removes the waitlist entry too. Previously an e-mail address could remain on the waitlist after the account behind it had been deleted.
Membership of an organisation is now what grants a partner administrator access. A separate personal subscription is no longer required.
Several plan descriptions promised an upload limit that the platform did not actually allow. The descriptions now match what happens.

14 September 2026

Large documents no longer stop halfway. Document processing is given considerably more time, and a document that cannot be processed is reported as failed instead of being retried silently forever.
You can choose whether text recognition runs. Scanned documents benefit from it; documents that already contain text are processed faster without it.
Long words and links no longer stretch a dialog beyond the edge of the screen.
A partner session no longer expires in the middle of a conversation — the session is extended as long as you are working.

7 September 2026

The Creator Portal shows your organisation's Twins, not only your own, and states who created each one.
A Twin's usage allowance is now charged to the organisation that owns the Twin.
An administrator whose access has been suspended no longer retains owner rights.
Deleting a Twin while someone else is editing it is now reported clearly instead of failing without explanation.
A screen-reader label was read out in English on the German interface.

31 August 2026

Every answer carries its legal notice, on every surface a Twin can be reached through.
The running version is shown in the top right of every page.
A Twin without documents can no longer be published — it had nothing to answer from.
"Help & Docs" opens the help page instead of the billing page.
A conversation in which a single field was malformed is now answered, with the problem named, instead of being abandoned.

28 August 2026

Documents carry metadata — title, author, language and more — from upload onwards, visible everywhere the document appears, and correctable in bulk without processing it again.
Documents that failed processing are visible instead of silently blocking further analysis.
The learner interface follows the language set on the Twin.
Embedded Twins keep all their settings when the page switches language.
Deleting an account now completes fully.