Library

Library in the left sidebar collects the files that have passed through an agent: the ones you attached, and the ones agents produced themselves.

It is an index, not a store — Croli records the path, and never moves, copies or alters a file. The page itself is strictly read-only.

There are two ways in: Library in the left sidebar, and the quick-actions menu in the status bar. The sidebar row can be switched off under View options → Navigation items; doing so only drops the shortcut, the status bar still gets you there. Its position is not in the address bar, so a reload lands you back in the conversation workspace.

What it collects

One row per file, keyed on conversation plus path: the same file in two conversations is two rows. The default order is last seen, newest first, and there is no sort control.

Only seven kinds are indexed:

Code, Markdown, txt, csv, json and archives are not collected — the library is for documents and media, not for the source files in your repository.

Source is one of two things: Uploaded (what you attached) or Generated. Generated covers more than what a write-file tool produced: paths are also swept out of command lines and command output, but those only count when the file exists right now and its modification time falls within half an hour of the turn that mentioned it — so cat-ing a PDF that was already in the repository does not index it. Images are never discovered from output (source trees are full of them, and one ls would drag the lot in), so a chart.png written by a script will not show up unless a write-file tool produced it or you attached it yourself.

Finding something

A row of filters along the top, all of which stack:

• Source — All / Uploaded / Generated, single choice
• Type — the seven kinds above, multi-select, each labelled "{count} files"; that number counts everything and does not follow the other filters
• Agent — which CLI handled it; multi-select and counted too
• Search — the box says "Search file names", but it matches file names and session titles, and never file contents
• Date range — three buttons (7 days / 30 days / 90 days) plus a More menu (This month / This year / All time / Custom); All time by default, filtering on last seen
• Grid and List views, and your choice is remembered

Sixty at a time; Load more and "{shown} of {total}" appear below only while there are more. An empty index reads "Nothing here yet…", and filters that match nothing give "No files match these filters."

What you can do with a file

Click opens the detail sheet, double-click (or Enter) opens the file, and Space also opens the detail sheet. The right-click menu and the detail sheet carry the same four actions:

Open does not mean "open in your system's app". PDFs, audio and video have no built-in preview, so reading them as text just produces garbage — to hand one to your own applications, use Reveal in Finder.

There is no rename, delete, download or move. The library does not touch your files; do that in the file system.

The detail sheet shows Type, Source, Size, References, First seen, Last seen, Path, Session, and Folder (that last row is absent when the conversation belongs to no folder). Only images get a thumbnail — under 2 MB, and not read from disk until it scrolls into view; everything else shows an icon.

Once you move or delete a file, its row is dimmed and marked "Missing", and only the detail sheet spells out "This file is no longer on disk". Open and Reveal stop working, and double-click does nothing.

When the index updates

Opening the page runs an incremental sync by itself — only when there are stale conversations to read, and silently. The Index menu in the corner holds the two manual actions:

• Refresh index — reads only the conversations that changed. This item also carries a "Last refreshed {time}" line, or "Never refreshed" if it has not run.
• Rebuild from scratch — re-reads every conversation. Use this when a file you are sure exists is missing from the library.

While it runs you get a progress bar: on the left the title of the conversation being read, on the right a done / total count. A manual run ends with "Indexed {count} files"; the automatic one on open says nothing. If some conversations could not be read it says "Some conversations could not be read", and the rest are indexed regardless. Only one sync runs at a time, and progress is broadcast — a sync another window started drives your progress bar too.

Worth knowing

• It costs nothing until you open it — the sync runs when you open the library, or when you expand the Assets block in a conversation's details, never in the background. After a Croli update the first visit may rebuild the whole index, and that one visit is slower.
• The index is derived: it is re-parsed out of your conversation transcripts. Delete a file on disk and rows with solid provenance — uploads and write-file tools — are marked Missing, while rows swept out of command output simply disappear on the next parse. Deleting a conversation removes its files from the library at once, and restoring the conversation brings them back.
• Agents do not use the library. It is what comes out of their work, not an input to it: there is no retrieval tool and nothing is injected into a prompt. To give an agent a file, use + → Attach files in the composer.
• Session details carries a small Assets block too, listing just that conversation's files (up to 24, opened with a single click), with uploads and generated output in two columns.
• Attachments are not always copied. On a local desktop workspace the file is referenced where it is; only in a browser or against a remote workspace is an attachment uploaded, into a folder per conversation under the server's ~/.croli/uploads/ (CROLI_HOME / CROLI_DATA_DIR move that location). A backup packs it up along with the database.
• Connected to a remote workspace, the paths here are paths on that server — and neither a browser nor a remote workspace offers Reveal in Finder.