Knowledge: one place for every document

Documents arrive in GPTree from several directions. You upload a file. You add reference documents to a project. You drop a file into a chat to ask about it. An agent writes a report every morning, or a crew of agents writes a brief. The desktop app keeps a folder on your computer in sync. Your own systems send documents in through the Knowledge API. Each of those used to live in its own corner of the product, found a different way, and only some of them kept any history at all.

The Knowledge page is the one place they all appear. Same list, same history, same way of reading them, no matter which of those five things produced the document.

Finding something

The search box searches names and the text inside your documents. Results come in two groups: documents whose names match, and, under Inside your documents, passages whose text matches, each shown with the sentence that matched. Clicking a passage opens that document at those words. There is more about this, and about seeing which answers used a document, in Find a document by what it says.

You can also group documents into collections, so the ones you need for a job sit together: see Collections.

What you see

Open Knowledge from the sidebar or the left rail. Each entry shows the document's title, where it came from, which project it is in, how many answers have used it, and which version is current. The Came from label is the honest answer to "why does GPTree have this": Document for a file you uploaded or added to a project, Web page for a page you added by its address, Chat attachment for something you attached to a message, Saved conversation for a conversation you chose to keep, Written by an agent for a report one of your agents wrote, Written by a crew for a crew's final output, with a new version each time a run says something new, and Synced for a document your own systems sent in, followed by the name of the source it came from. A document the desktop app syncs from a folder on your computer reads Synced · This computer and shows the file's path. A synced document also shows the identifier your system uses for it, so you can match it to the original.

You can add to the library from the page: upload a file, save a conversation, or paste the address of a web page and GPTree keeps a copy of what the page said. That copy is taken when you add the page: GPTree does not go back to check the page for changes later. You can rename a document, and remove one. What removing does depends on where the document came from, and the confirmation says so: removing a file you added to a chat takes it out of your knowledge and leaves the chat's copy alone, while removing a document you uploaded deletes it.

Documents that reach you through a project are here too: anything in a project someone shared with you, and anything a collaborator adds to one of your projects. Each one shows the project's name and who shared it, and you can open, read, and compare their versions like your own.

Your workspace's company library is here too: the documents everyone in your workspace works from, added by the workspace's owners and admins. When you belong to a workspace that has one, a Company choice appears alongside the others.

Choose All, Only mine, Shared with me, or Company to decide whose documents you see, pick where documents came from when you want one kind, or search when you know roughly what you are after.

Versions, and what changed

Open a document and you get its version rail, newest first, with the current version marked. Click any version to read what that version actually said, not what the document says today.

Next to any older version is Compare with current. That shows a line by line difference: what was removed, what was added, what stayed. This is the part that is hard to do by hand and easy to get wrong from memory, especially for a document that has been revised a dozen times.

A version is written when the content changes. Re-uploading an identical file, or an agent routine producing the same report two days running, does not create a new version. So the version rail answers "what changed and when" rather than "how many times did something touch this".

Why it works this way

Keeping history is only half of it. The other half is that the knowledge engine reads from this same collection, so a document is not just something you can find, it is something GPTree can use. That is why an agent's report and a file you uploaded sit in one list: once a document is here, it can be cited in an answer, and it can be reached through the API, regardless of what created it.

Good to know

• Deleting the source removes the document from this page and from search. Citations in older answers report the passage as unavailable rather than showing content you removed.
• Restoring a document you deleted brings it back here, with its history intact.
• Other people's documents appear only when a project or your workspace's company library gives you access to them. If you are removed from a project or a workspace, its documents drop out of your list.