LeraLink Development Update: 20 November 2026LeraLink Team

LeraLink Development Update: 20 November 2026

A new Tenant Admin ready for pilots, faster pages, and publishing that waits for nothing

date: . This update covers the released between 18 and 20 November. Tenant Admin, where an installation is set up and run, has been rebuilt so that an organisation joining a pilot can set up and run its own installation without help. Pages open faster, on phones and for returning readers, and the first reader anywhere in the world now receives a prepared page. and people are linked to a text only where the text really names them. Publishing no longer waits for any machine work, and machine-made titles and metadata in other languages now come to an editor for approval.

A new Tenant Admin, ready for pilots

Tenant Admin is where an installation's owner and administrators set up its brand, languages, people, email and . We redesigned every page of it with one aim: someone who has never seen should be able to set up an installation on their own, on a phone or a laptop.

Your brand, on every screen. Tenant Admin now uses the installation's own accent colour, heading typeface and colour mode. It is in for every operator, whatever language the site is read in, and on a phone as fully as on a desktop, with a menu that opens from the top bar.

A place to start. The first page is now an Overview with a setup checklist: brand, languages, email sender, first member, first Project, Wall, and custom address. Each item says whether it is done and opens the page where it is done.

Everything in one place, once. The pages are grouped into six sections: Overview, Site, , People, Publishing and Installation. Each page says in one line what it is for. Every page and tab has its own address, so a refresh, the back button and a shared link all return to the same place. An installation sees only what its plan includes, and that holds on the server too, not only on the screen.

Daily work moves to the Studio. Work done every day now lives in the Studio, where are written. Projects are created there with a "New Project" button and a guided set-up, and their policies, status, copies and archiving are managed there too. Creating Projects is now a permission of its own, which the owner can give to anyone through a role. Permanent deletion stays with the owner and administrators, and only for a Project that is still an empty draft. Owners and administrators also manage Pieces in bulk in the Studio: unpublishing, archiving, deletion with a typed confirmation, and scheduled deletions. The Studio's reading surface now has version history, so an earlier version can be restored.

Reader notes, moderated with care. The notes readers leave in a text's margin are now managed under Reader notes. Moderators can select many notes at once and hide, delete or restore them. A deleted note is kept privately, so a mistake can be undone. Permanent deletion asks for a typed confirmation and is recorded in the installation's audit log. Notes can be downloaded in full as CSV or JSON. When an author withdraws their own note, its text is erased.

Clearer settings. The AI settings have one tab for each service: , images and search metadata, each saved on its own, with an overview of how AI is used in the installation. The Library and public profiles are searchable lists that stay quick with thousands of entries. Where a setting is locked by the plan, the page says who can unlock it.

With this release, is ready for its first pilot installations. Before release, every one of Tenant Admin's 52 pages and tabs was checked at phone and desktop width, in light and dark mode, on an -only installation and on an English– one, shaped like the organisations we expect in a pilot. Organisations interested in a pilot can write to hello@lera.link.

Pages that open faster

The largest share of the public work in this release is speed, measured the way search engines measure it: a mid-range phone on a slow connection.

A prepared page in every data centre. LeraLink is served from 's data centres around the world. Until now, the first reader to reach a page through a data centre that had not yet shown it waited while the page was drawn, which could take between one and nine seconds. Every public page is now also kept in one shared store inside the , and the first reader anywhere receives it from there. Measured from Amsterdam just after a publication, those first requests were answered in 66 to 126 milliseconds. Only pages open to everyone are kept there. Restricted and personal pages never are.

Pages are prepared when you publish. A few seconds after you publish or withdraw a , its pages, the home page and the Wall are drawn in advance, so the first reader does not wait for them. Publishing one Piece also no longer makes every other page of the site be drawn again: only the pages that show that Piece are refreshed.

Less work when a page opens. When a page appears, the reader's browser takes it over so that buttons and menus respond. On a phone, that takeover kept the browser busy for long stretches in which a tap could go unanswered. We removed the repeated work behind it. In our release test of the home page on a phone, the time in which the page could not respond fell from 842 milliseconds to 13. Search engines count anything under 200 milliseconds as good.

A smaller program behind every request. The program that answers each request is now less than half its former size, and parts used only occasionally, such as print charts and passkey sign-in, are loaded only when needed. The median first response on sajadsepehri.com fell from 320 to 189 milliseconds.

Returning readers get the prepared page. A reader who comes back to a site now receives the prepared copy as well, instead of a page drawn anew for them.

Lighter pages. A reader of a site whose interface is in English no longer downloads the full Persian interface vocabulary, about 300 KB less on a first visit. The first picture on the Wall now appears with the page itself.

Pages that stay still. On the Hub, a few parts of some pages, such as the list of people, an event's time zone and the promotion button, were completed only after the page had appeared, so the page moved slightly. They are now part of the page from its first moment.

links you can rely on

LeraLink recognises the and people a published text talks about and links them to it. On the Hub we noticed , the scholar of the tenth and eleventh centuries, listed as a related figure on texts that never mention him: privacy and data-processing documents, among others.

The reason was a . , the open knowledge base LeraLink draws on, records «بیرونی» () as one of his names, and in Persian the same word also means "external", as in "external AI services". Every text that used the ordinary word was linked to him.

now work as they do in careful writing. A one-word short name counts only in a text that also names the person or concept in full, and only in the language it belongs to. Where a surname is a name on its own, as "Bourdieu" is, an editor can say so in that identity's settings. Machines never make that decision. Short names suggested automatically are also checked before they are kept.

After the whole Hub had been read again, Al-Biruni appeared on one text, where before he had appeared on 18.

Publishing that waits for nothing

Publishing is immediate. Publishing no longer waits for , search metadata or indexing. The Piece is live as soon as you publish it, and that work continues in the background.

Machine work only in your public languages. After publication, LeraLink prepares a Piece's title, subtitle, keywords and search metadata in the installation's other public languages. Languages that are selected but not public no longer receive any automatic work. The body of a Piece is never translated by a machine unless an editor asks for it.

An editor decides. These machine-made titles and descriptions now appear to readers marked as not yet reviewed. Until then, an editor approves, corrects or rejects each one in the Piece's settings, under Languages. What an editor approves or corrects becomes the Piece's own text.

Importing that can be resumed. The import page now warns you before you leave it while an import is under way. If you come back to it, it shows what was already done and offers to open the Piece, publish it again, or cancel the import. Cancelling removes everything an import wrote, if its Piece was never published.

Restricted content

A restricted Piece or Project is readable only by the people and roles it names, and no model ever reads it. In this release we extended that protection across more of the platform, including events, publication files, statistics and the AI services. The publication files of a restricted Project are now available to the installation's owner and administrators.

Search engines

Each site's sitemap, the list of pages a search engine reads, is now divided by kind of content, so it stays quick to produce however large a site grows. The earlier addresses continue to work.

What comes next

The speed work in this release is also the foundation for our next direction: LeraLink as an app. We are working towards making every installation installable as a progressive web app, so that readers and authors can add it to a phone or computer from the browser and open it from the home screen like any other app, with no app store in between. We will describe it in a later update.

Two pieces of work continue. When a site has not been visited for a while, the first response can still take more than half a second; the next step is to divide the program in two, so that the part answering most requests stays small. On phones, the home page is still the slowest page to show its main content, and it remains at the top of our list.

In this Piece