Skip to main content

Workspaces and accounts

A workspace is the container for one company’s knowledge. Every document, folder, tag, and property definition belongs to exactly one workspace. Every workspace belongs to an account, which holds the people and the paperwork: members and owners, the plan and billing, invitations, and access tokens. An account has exactly one workspace, created when you name your company during setup, and the workspace goes by the account’s name. An account owner can change that name under Settings → General. It is the name shown in the account switcher, on invitations, and when you connect an agent. A workspace also has Instructions: the rules every connected agent reads before writing, such as your top-level structure and where things go. Anyone in the account can edit them under Settings → Instructions. See Workspace instructions. Your profile is you: the email you sign in with. One profile can belong to several accounts, and the user menu at the bottom of the sidebar switches between them. See Belonging to several accounts. How the app looks is a per-browser setting; see Theme. You can also share a folder with someone outside the account. They sign in with the email you shared it with and can read the folder and everything in it, including documents added later, but they cannot change anything, and they do not join the account. See Sharing a folder. The same boundary applies to your agents. Each agent connection reaches exactly one workspace, and no tool lets it reach another. To give an agent a second workspace, add a second Tatara entry in your client. See One entry per workspace.

Documents

A document is a title and a body, plus a few properties that make it findable later: The description deserves more attention than it looks like it needs. It is what an agent sees when it scans your workspace for relevant context, and it is usually what decides whether the agent opens the document at all. A vague description is the most common reason good knowledge goes unused. Descriptions and tags are optional when you write in the app. They are required when an agent writes: anything an agent creates must arrive with a description and one to five tags. Agents are held to a higher standard than people here on purpose, because they are the ones producing volume. Documents also record where they came from. Every document is stamped with what wrote it, taken from the credential used, so an agent’s work is always attributed to that agent. It cannot be supplied, overridden, or faked by the writer. When you are reviewing your workspace, you can always tell what a person wrote and what an agent did.
There is no draft, published, or archived state. Nothing is hidden from search because of where it sits in a lifecycle. A document either exists in your workspace or it does not. The one flag that comes close is the lock, and it governs who may change a document, never who may see it.

Locked documents

A document can be locked. A locked document is approved, canonical content, such as an ideal customer profile, a brand guide, or a privacy policy. It refuses every change from agents and people alike until someone unlocks it: no edits, no retagging, no renaming, no moving, no deleting. Reading is untouched; it still appears in search, listings, and links like anything else. Locking is deliberately a human act. The document’s owner, or an account owner, selects the padlock in the document’s top bar, next to the favourites star. Anyone else sees the padlock lit while the document is locked, and pointing at it says who can unlock it. There is no agent tool for locking, and there is no way around it: an agent that tries to change a locked document is refused, told who can unlock it, and told explicitly not to work around the lock by writing a copy somewhere else. The padlock is on documents that open in the app’s main view. Workbooks, data pages, slide decks, and whiteboards open on pages of their own, which have no padlock. This is the mechanism that lets you run agents at volume without your source of truth drifting. Lock what has been agreed; leave the rest open. Because a locked document is pinned in place, the folders above it are held too: a folder containing a locked document anywhere inside it cannot be renamed, moved, or deleted until that document is unlocked.

Extra properties

Beyond the standard properties, documents can carry others your own way of working needs: a review date, a customer name, a renewal quarter, a confidence level. They appear in the Custom card of the document’s details rail, and your agents can read and filter on them. “Every customer document not reviewed since June” is a question a base can answer only because the review date is a property rather than a sentence buried in the text. Each extra property is defined once, on the Fields page in the sidebar, with a name and a type: text, number, checkbox, date, date and time, select, or list. That is what keeps them worth having: an agent can only write a property that has been defined, so a workspace never accumulates four spellings of the same idea. An agent can define a new one when it genuinely needs it, but it cannot invent an ad-hoc label on a single document and move on. Documents that arrive from elsewhere carrying properties nobody has defined still show them. Those properties are listed on the Fields page under In use, unregistered, where Define field adds one to the registry. It is the same curation pattern as the tag registry.

Kinds of document

The plainest document is prose: a title and a body you write in the editor. A document can also be drawn as a board, a tracker, a table, a mind map, a saved view over other documents, or a dashboard. Underneath it is still Markdown, and the app draws it in that format. You can start a board, tracker, table, or mind map yourself from New document in the sidebar. Some kinds of document are their own thing rather than a format, and each opens on a full-screen page of its own. Your agents build these. Ask for one by name.

What a document actually is

Prose documents and every format are stored as plain Markdown with structured properties attached. That is the same open foundation as Google Cloud’s Open Knowledge Format, the emerging standard for the context AI agents read. That has two consequences:
  • Agents read it natively. Markdown is what language models are trained on and what they write best. Your knowledge does not have to be translated, scraped, or embedded into something else before an agent can use it. It is already in the shape agents want.
  • It is yours, in the open. Nothing is locked in a proprietary format. A board or a table document is not a special file type or a database row. It is all text you could read in any editor, search with any tool, or take somewhere else entirely.
The other kinds are different. An HTML page or a slide deck is HTML. A workbook keeps its cells in database tables of its own, and you can download it as .xlsx, or one of its tables as .csv. Data pages and whiteboards store a structured body that only their own agent tools write. You never have to write, read, or think about any of that. The app shows you documents, boards, tables, maps, workbooks, decks, and whiteboards; your agents handle the structure underneath. It matters because it explains how Tatara works, and because it means your knowledge outlives any decision we make about the product.

Titles and addresses

The title is the document’s identity, and its address follows from it. Rename a document and it moves; every link pointing at it is rewritten in the same operation, so nothing breaks and nothing has to be tidied up afterwards. A document’s address is its folder, then the document, using simplified, URL-safe versions of each name: product/pricing/pricing-page-spec. You see the full address in your browser’s address bar; everywhere else you see the real title. If a name is already taken in the same folder, Tatara adds a number rather than refusing or overwriting.
A title cannot contain /, #, |, [, or ], or a line break. Each of those means something in the way documents link to each other, so a title using one would produce links that could never resolve. Tatara refuses it up front rather than quietly mangling the name.

Folders

Folders group documents, and they nest up to twelve levels deep. Two folders in different places can share a name. A notes folder under product and another under marketing are different folders. What tells them apart is their full address. If you ask an agent to file something in “notes” and the name is ambiguous, it will ask you which one rather than guess. Documents link to each other, and Tatara tracks those links in both directions: every document knows what it points at and what points back at it. That backlink half is what a plain folder of notes is missing. Open your pricing page and its details rail shows a Links card listing what it points at, and a Linked mentions card listing every document that links to it. To link to a document while you are writing, type its name inside double square brackets:
The link becomes clickable as soon as you finish typing it. To show different words, add a |: [[Pricing Page Spec|the spec]]. Your agents use the same links, plus a couple of longer forms for pointing at an exact address or embedding one document inside another. A link to a document you have not written yet is not an error. It is kept, shown dimmed, and connects itself the moment a document with that name appears. You can write toward a document that does not exist yet and let the link connect when you get to it. Deleting works the same way in reverse: links to a deleted document dim rather than disappear, and restoring it brings them back.

Tags and the taxonomy

Documents carry up to five tags. Separately, your workspace keeps a tag registry that sorts every tag into one of three tiers: The three are independent, not a hierarchy. A topic does not belong to an area, and any goal can apply to a document in any area. A pricing page spec might be area product, topic pricing, goal acquisition: three separate answers, not one path through a tree. It means you can ask “everything about pricing” and “everything aimed at acquisition” and get useful answers to both.
The registry is advisory. Tags on documents are not checked against it, and coining a new tag never fails a write. A tag that is not registered turns up under Needs review on the Tags page, where you can add it to the registry in a tier or merge it into an existing tag. Agents can coin tags freely and you curate afterwards, rather than agents stalling on a vocabulary they do not have.
The registry suggests one area, up to two topics, and up to two goals per document. That is guidance for you and your agents, not a rule the product enforces. The only hard limit is five tags. See Organizing with tags.

Deletion and recovery

Deleting is soft. Tatara keeps the content rather than destroying it, which is what makes recovery possible. Every delete in the app shows Deleted "…" with Undo for about five seconds, and the message stays while you point at it.
After that message has gone, the app has no way back. There is no trash, and nothing in the app lists deleted documents or restores one.
From then on, recovery runs through a connected agent. A folder comes back when you tell the agent its name and where it was. A document comes back only if the agent already knows its id, because deleted documents can’t be found by searching. See Getting something back later.

Where to go next

Document formats

Boards, trackers, tables, maps, and live views over your workspace.

Connect an agent

Give your agent access, and put your workspace to work.