Manage your document library

The documents tab at /admin/documents is the page you'll spend the most time on after the first week. New policies show up, old ones get replaced, the IT runbook needs another section. This article covers the day-to-day work of keeping the library current.

The documents tab at a glance

Three stat cards at the top: Total documents, Total chunks (every doc is broken into smaller searchable pieces; chunks are how the system retrieves the right passage), and Errors (files that failed to ingest or got stuck mid-process). Below the cards, the documents table lists every file the system knows about. The Errors counter is your "look at me" signal. When it's anything but zero, the table tells you which rows need attention.

What the columns mean

Type (the file format), Filename, Status (Pending / Processing / Indexed / Failed, the same four-state lifecycle covered in upload your first documents), Chunks (how many searchable pieces the doc was broken into; a 30-page PDF is usually a few dozen), Last updated (relative time), and a three-dot row menu.

The status badges tell the story. Anything stuck in Processing for more than a few minutes is treated the same as Failed by the Errors counter.

Adding documents

Click the Upload button (top right) or drag files directly onto the page. You can select multiple files at once. The upload dialog shows per-file progress and the status badge as each one moves through ingestion. File type, size limit, and what to do when something rejects are all covered in upload your first documents.

Deleting a document

Each row's three-dot menu has a Delete action. A confirmation dialog opens before anything happens because deletion is final. For larger cleanups, the checkboxes on the left of each row let you select several rows at once, and a bulk Delete button appears at the top of the table showing how many you've selected.

Deleting a document removes its chunks immediately. Your team can no longer get answers from it the moment the row disappears. One thing that does survive deletion: the historical Q&A log entries that cited the document stay where they are. The pills in those old log rows just become non-clickable references. The reasoning: questions your team asked last quarter were real questions, and the answers they got were real answers. Wiping that history when the underlying file is removed would lose useful context.

The Sync Now button

Sync Now re-enqueues failed and stuck documents back into the ingestion queue so they get another try. Two real use cases: you fixed a previously failed file (for example, you OCR'd a scan-only PDF and re-uploaded it, but a sibling file is still showing Failed from before, and Sync Now retries the sibling), or a file got stuck mid-process and you want to nudge it. For the standard upload flow you don't need it.

When ingestion fails

The failure modes (scan-only PDFs, corrupt files, oversized uploads, unsupported types) and the recovery for each are covered in detail in the when a file fails section of the upload article. Sync Now is the in-place retry for files that are still on disk; re-uploading is the fallback if the underlying file needs fixing first (OCR pass, format conversion).

One canonical version per document

Don't upload three different revisions of the same policy hoping the system will pick the latest. It cites whatever retrieval ranks highest for a given question, which can be any of them. That's how employees end up reading the 2019 PTO policy on a Monday and the 2024 version on a Tuesday.

Keep one canonical copy of each document in the library. When something changes, delete the old version and upload the new one. The version history belongs in your file system or document store, not in the chunks the chat reads from.

What to do next

With the library under control, the next thing worth setting up is your team. Invite and manage your team walks through bringing employees in, picking the right role, and handling the lifecycle from invite through password reset.