Workspaces and your library

A workspace is an ordinary folder. Everything for one project lives in it: your papers, the drafts, the agent’s notes, and the conversations. Open one with abstract <folder>, or from the Workspaces screen.

Creating and switching

On the Workspaces screen you can:

• Create a new workspace. It’s made in ~/Abstract/<name>.
• Open a recent workspace (the last 8 are listed), or type any existing folder path (~ works) to open it.
• Delete a workspace you’re not currently in. This permanently removes the folder and everything inside it, your files included. As a safeguard, abstract only deletes folders that contain an .openpaper/ folder (created the first time a folder is opened as a workspace), and never your home folder or the workspace that’s open.

Running abstract with no folder reopens the last workspace you used, or creates ~/Abstract/default the first time.

abstract never uses your home folder (or the disk root) as a workspace, because it would scan everything on your computer. If you ask for it, abstract opens ~/Abstract/default instead.

What’s in a workspace

Folder	What goes there
sources/	papers you add or the agent downloads
drafts/	documents the agent writes, plus exports
figures/	page images the agent rendered to look at
analysis/	scripts and outputs from data analysis
.openpaper/	the library database, conversations, memory, and reading notes (hidden)

You can add your own folders and files anywhere.

The Files panel

The panel on the right shows the workspace. Folders come first; drafts/ and sources/ open by default, and it refreshes every few seconds.

• Click a file to preview it: PDFs and HTML, images, rendered Markdown, or plain text.
• Hover a file to download it, delete it, or (for Markdown) open a printable view with PDF so you can save it as a PDF.
• Drag files between folders to move them. A moved paper keeps its library entry and notes.
• folder+ creates a folder; + Add saves files into sources/.
• The virtual notes folder shows the agent’s private reading notes, one per source. They help it remember what it read, but they’re never cited as evidence.

When you’ve been searching the literature, a screening bar at the top shows how many papers were identified, screened, and included.

Adding papers

Click + in the chat box:

• Save files to workspace: PDFs, Markdown, text, LaTeX, BibTeX, or images go into sources/. A file with the same name is replaced.
• Attach to this message (not saved): up to 6 files (PDF, Markdown, text, images) sent with one message only, for quick questions.
• Import from URL: paste a paper URL, DOI, or arXiv id and the agent fetches it.

Or simply copy files into the workspace folder.

How papers become searchable

Saving a file does not index it. The agent ingests a paper when it needs it, or you can ask it to (“ingest everything in sources/”). Ingesting:

1. extracts the text, page by page, and splits it into passages;
2. looks the paper up in Crossref and OpenAlex (by DOI if it finds one, otherwise by title), so citations use the official record;
3. assigns a grade: peer-reviewed, preprint (for example arXiv), or note (no registry match, such as your own notes);
4. catalogs figures, tables, charts, and equations so they’re searchable too.

Ingested files show INGESTED and their grade in the Files panel.

Scanned PDFs (image-only pages) are detected automatically and transcribed by your model, ten pages at a time. A scanned PDF must be under 18 MB, so split larger ones first.

Reading before citing

The agent can only cite sources it has actually opened: read page by page, viewed, or summarized in a note. A downloaded but unread paper can’t be cited. If you ask for a draft before anything was read, it will tell you which papers it still needs to read.

Conversations

Each workspace keeps its own conversations, listed in the sidebar (up to 50). + New chat starts a fresh one; the last open one comes back when you reload. The agent can search your other conversations in the same workspace when you refer to earlier work.