The workbench in the cloud
The same workbench as the one on your computer, in your browser, at an address of its own. The same pages, the same projects and folders, the same record of every question put to a model, for when you work on more than one computer, or away from your own.
It keeps your research on the server rather than in a folder on your computer. The two are each complete on their own; syncing is how a project travels between them, when you choose.
Built The cloud workbench is built and being tested, with a short list of people. You can ask to be on it.
Signing in
With your account on this site, once you have been given the workbench.
The cloud workbench has no accounts of its own. You sign in with the account you have on this site - its password, a magic link or a passkey - and if you are already signed in here, you go straight in. Signing out of the workbench signs you out of this site too.
While it is tested, it is open to a short list of people. Ask for it by ticking the box when you register, or from your account's Cloud workbench page. An administrator reads each request, and you are told by email; the same page says where your request stands.
An administrator can also invite an address. If there is no account with it yet, the invitation waits until one is registered and confirmed. Signing in before you have been given the workbench says so, rather than letting you in.
Your work, and nobody else's
Kept apart twice over.
Everything you bring in or make is kept in a workspace of your own. Every page the workbench shows, and every piece of work it does for you in the background, is filtered to it; and beneath that the database itself refuses to show or change anybody's rows but the person's the work is for, so a mistake in the workbench still could not reach another person's research.
Models are asked with your own AI keys, which are kept encrypted and are never shown back to you or to anybody else. What you spend is yours, and nobody else can spend it. Your settings - the models you use, the look, the widths of the tables - go with your account, whichever computer you sign in from.
What it does differently
Less capable where it has to be, and it says so where it is.
- Models. Only hosted ones - Anthropic, OpenAI or Gemini - with your own key. There is no model on your computer to ask from the cloud.
- Searching. By the words you type, in documents and in the passages they are cut into, and answering from your papers with the passages found that way. Searching by meaning is off until you turn it on in Settings, Searching: then each passage is sent once to OpenAI's embedding model with your own key, and so is each search by meaning. A document labelled personal or special category data is kept back, unless its folder has an exception for hosted models; its passages are found by their words, and the page says how many were kept back.
- Code. Code a model writes is not run in the cloud: the hosting cannot run it in isolation. The workbench on your computer checks, builds and tests it in a sandbox.
- Zotero. The workbench on your computer reads your Zotero library there. The cloud's will use Zotero's own web service, which is still to come.
- Saving from the browser. The browser extension saves into the workbench on your computer; a page saved there reaches the cloud when its project syncs.
Still to come
Projects shared with people you invite, notes written together, and a log of who did what that cannot be altered unnoticed.