Skip to main content

​
Finding and recovering content

The command palette matches document titles, paths, and slugs. It does not look inside document bodies. It also does not match tags, descriptions, or a document’s type.So if you remember a phrase but not the document it sits in, ⌘K will not find it.Full-text search across document content does exist: it is available to a connected agent, through the search tool over MCP. The workaround today is to ask your agent to find it: describe the phrase you remember and let the agent search the bodies and hand you the document.More detail in Finding documents.
Not from the web app. There is no trash, no undo, and no restore button.Deleted content is kept, not destroyed, so a connected agent can bring it back:
  • The agent restores a document by its id.
  • It restores a folder by the slug or full path the folder had before you deleted it.
The two behave differently when something is in the way. A restored document takes the next free name if something else has since taken its place. A folder restore refuses instead of renaming. If a live folder now sits at that path, move or rename that one first. And a folder whose parent folder is also deleted cannot be restored on its own: restore the parent, which brings the whole subtree back with it.
There is no way to search for a deleted document to recover its id. If you are about to delete something you might want later, note its id first.
More detail in Deleting documents and folders.

​
Organizing

The web app has no way to move a folder. There is no drag for folder rows and no menu item for it. Moving a folder is only possible through a connected agent. Ask it to move the folder, and the move updates the paths of everything inside.Documents are different: you can drag a document onto any folder row in the sidebar, or use Move… in its menu.Renaming a folder in the app does work, and it updates the paths of every document and sub-folder inside it.More detail in Folders and organization.
A folder that still contains documents or sub-folders cannot be deleted from the web app. The dialog opens, but the Delete button stays disabled and the text names what is still inside.Two ways forward:
  • Empty the folder first (move its contents elsewhere or delete them), then delete the folder.
  • Ask a connected agent to delete it. An agent can remove a folder together with everything in it, which is the practical route for a large branch. It confirms first: the agent reports how many documents and sub-folders would go, and only removes them once you confirm.
More detail in Deleting documents and folders.

​
Saving and rendering

The most likely cause is tags. Five tags per document is the enforced limit, and the tag field does not stop you adding a sixth. The save fails instead. Remove tags until you are within the limit and the save goes through.The other cause is your plan’s document limit. If the workspace is over it, documents become read-only until you upgrade or delete some. Reads and deletes keep working at any level, so you can always free up space.See Organizing with tags and Plans and limits.
The document declares a format that could not be rendered, so the app fell back to the plain editor. The banner names the format the document was meant to render as, with a short note on what went wrong.Three causes:
  • The declared format is not one Tatara recognizes.
  • The format’s configuration is invalid.
  • The body no longer parses as that format, usually because an edit broke the structure the format requires.
The document itself is intact, and you can edit it as markdown in the meantime. Fix the body so it parses against its format and the rich view comes back.More detail in Document formats.
HTML documents render in a sandbox with no JavaScript and no network access. Scripts never run, and external stylesheets, fonts, and images never load, so anything the page was going to build or fetch at load time is missing.Inline everything the document needs: CSS in a <style> block, images and fonts embedded as data URIs, and any interactive result worked out in advance and written into the markup as static HTML.
Only create_html_document creates an HTML document, but an agent can fix one afterwards with write_document, passing the document id and the corrected markup. The replacement is re-checked against the same HTML rules.
More detail in HTML documents.

​
Agents

A rejected write is a normal step in a working agent loop, not a fault. Every rejection states what is wrong and how to fix it, so a capable agent corrects itself and retries without your help. If your agent stops and reports the rejection to you, reading it usually tells you exactly what to ask for.If an agent keeps failing the same write, the usual cause is that it is retrying without reading the response. The agent skills teach it the document formats and the recovery loop up front.
Check the server address first. It is exactly https://www.tatara.app/api/mcp. No other host or path works.If the client cannot complete the browser sign-in, create an access token under Settings → Access tokens and configure the client to send it as a bearer credential instead. Only a workspace owner can create one.If an agent that used to work suddenly loses access, check two things. An access token is checked on every call, so revoking one under Settings → Access tokens stops the agent on its very next request. Disconnecting a browser-signed-in client under Settings → Agent access works differently: it stops that client renewing its access, so it must sign in again, but the credential it already holds keeps working until it expires, up to an hour later.If the person the agent connected as had their membership deactivated, they lose access to the workspace entirely, and so does anything connected under their account. A workspace owner can reactivate them from Settings → Members.More detail in Connecting an agent.