@@ -95,6 +95,32 @@ step 4:
 }
 ```
 
+Worth deciding now: **jack's copy of a download outlives the import.** Radarr
+and Sonarr read the finished file out of this folder and write your library
+copy, but nothing cleans up jack's — so the folder grows with every grab. Add
+`unlinkImportedFiles` to have jack drop its copy once the import is confirmed:
+
+```jsonc
+{
+  "downloads": {
+    "completedPath": "/data/torrents/completed",
+    "unlinkImportedFiles": true
+  }
+}
+```
+
+It's off by default because it deletes files, and it's a plain `unlink` of that
+one file — if your *arr hardlinked into the library, the library keeps the data;
+if it copied, you get the space back. You can also flip it later from the
+management UI without a restart. See
+[After the import](/guide/how-it-works#after-the-import) for the full picture,
+and [`unlinkImportedFiles`](/reference/configuration#downloads-unlinkimportedfiles)
+for the exact guarantees.
+
+Every other key in this block is an optional tuning knob with a sensible
+default — see [`downloads`](/reference/configuration#downloads). All of them are
+editable later from the management UI, so you don't have to get them right here.
+
 ### `servers`
 
 One entry per Radarr/Sonarr. Set each server's URL, and reference the API keys
 
@@ -1,5 +1,5 @@
 ---
-description: Learn how jack connects Radarr and Sonarr through Torznab and qBittorrent-compatible APIs to search and transfer media between trusted peers.
+description: Learn how jack connects Radarr and Sonarr through Torznab and qBittorrent-compatible APIs to search, transfer, and import media between trusted peers, and what happens to downloaded files afterwards.
 ---
 
 # How it works
@@ -64,6 +64,8 @@ sequenceDiagram
     ARR->>JACK: GET /api/v2/torrents/info (poll progress)
     JACK-->>ARR: completed → content_path = finished file
     Note over ARR: scans completed folder, imports into library
+    JACK->>ARR: poll history — did the import land?
+    Note over JACK: confirmed → optionally unlink jack's copy
 ```
 
 1. You grab a release. Your *arr's download client is the **qBittorrent**
@@ -78,6 +80,45 @@ sequenceDiagram
 5. *arr polls jack's `/api/v2/torrents/info` for progress; once jack reports
    the torrent complete, *arr scans the completed folder and imports the file
    into your library, renamed and tracked.
+6. jack watches that *arr's history until the import is confirmed, then — if
+   you've turned it on — removes its own copy from the completed folder. See
+   [After the import](#after-the-import) below.
+
+### After the import
+
+Importing doesn't consume the file in `completedPath`. Radarr and Sonarr read
+it and write your library copy; jack's copy stays where it was, and jack has no
+further use for it — a finished download is never re-served to peers or
+re-imported.
+
+That leaves you with two on-disk outcomes, and which one you get is decided by
+your *arr, not by jack:
+
+- **Your *arr hardlinked** (its default when the completed folder and the
+  library live on the same filesystem) — the library entry and jack's copy are
+  two names for the same bytes. Nothing is duplicated, but the completed folder
+  keeps filling with entries you'll never look at.
+- **Your *arr copied or moved** (different filesystems, or hardlinks disabled) —
+  the library now holds its own bytes, and jack's copy is a genuine second copy
+  of every file you've ever grabbed. Left alone, `completedPath` grows without
+  bound.
+
+[`downloads.unlinkImportedFiles`](/reference/configuration#downloads-unlinkimportedfiles)
+is the switch for this. Turn it on and jack unlinks its copy as soon as the
+import is confirmed: in the hardlink case that just drops the redundant
+directory entry and your library is untouched, and in the copy case it frees the
+space. It's **off by default**, so an instance you set up and forget will
+accumulate.
+
+The unlink is deliberately narrow. It runs only on an import jack has confirmed
+— the destination *arr reports the download in its history, or the manual import
+jack pushed reports `completed` — so a queued, in-progress, or failed import
+keeps its file, and so does a file another download still needs. jack removes
+that one file and nothing else; it never touches your library.
+
+Flip it from **Settings → Downloads** in the [management
+UI](/guide/management-ui) or set it in `config.jsonc`. It's the one key in the
+`downloads` block that applies without a restart.
 
 ## 3. Serving — being a peer to others
 
 
@@ -1,5 +1,5 @@
 ---
-description: Manage jack servers, peers, API keys, downloads, catalog requests, and logs from the self-hosted web management console.
+description: Manage jack servers, peers, API keys, downloads, catalog requests, and logs from the self-hosted web management console, with live updates over Server-Sent Events.
 ---
 
 # Management UI
@@ -11,11 +11,11 @@ instance without hand-editing `config.jsonc`. With the
 operate jack day to day:
 
 - **Overview** — your configured servers and peers, and whether each one
-  initialized cleanly.
+  initialized cleanly, with transfer activity updating live.
 - **Catalog** — browse everything your peers share, and request titles into
   your library.
 - **Downloads** — inspect, cancel, retry, and delete in-flight or finished
-  grabs.
+  grabs, updating live as they progress.
 - **Peers** and **Servers** — add, edit, and remove your friends and your
   Radarr/Sonarr connectors.
 - **Settings → Downloads** — the whole
@@ -28,6 +28,28 @@ operate jack day to day:
 Don't want it? Delete the `jack-ui` service and the backend's
 `JACK_MANAGEMENT_KEY` line to run jack headless.
 
+## Settings → Downloads
+
+The whole [`downloads`](/reference/configuration#downloads) block is editable
+here, so there's no reason to hand-edit `config.jsonc` for it:
+
+- **Completed folder**, **simultaneous transfers**, and the
+  **[drop imported files](/reference/configuration#downloads-unlinkimportedfiles)**
+  switch — the settings you actually choose — sit at the top.
+- **Transfer retries** and **import watcher** tuning are collapsed below, each
+  labelled with how many of its values you've moved off jack's defaults.
+
+Every field's placeholder is jack's default, and **clearing a field means "use
+the default"** — jack drops the key from the config file rather than storing a
+blank. `completedPath` is the exception: it has no default, so it can't be
+cleared. If your config has no `downloads` block at all, the page offers to set
+one up.
+
+One **Save** covers the form and only lights up once something changed;
+**Revert** discards your edits. `unlinkImportedFiles` takes effect the moment
+you save — **every other key here is read at startup, so restart jack** to apply
+it.
+
 ## Access control
 
 The UI supports two auth modes, depending on where the management key comes
@@ -60,6 +82,39 @@ separate listener the UI never touches.
 See [`apps/ui/README.md`](https://github.com/roziscoding/jack/blob/main/apps/ui/README.md)
 for the full UI configuration reference.
 
+## Live updates
+
+The UI doesn't poll. Overview, Downloads, and Settings each hold a
+[Server-Sent Events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events)
+stream open through the BFF, and jack pushes a fresh snapshot the moment
+something changes — a transfer's progress, a connector's state, a peer or server
+you just edited. There is no refresh interval to tune and no manual refresh
+button.
+
+The streams are plain `GET` endpoints on the management API:
+
+| Stream | What it pushes |
+| --- | --- |
+| `/overview/stream` | Connector overview + download state |
+| `/downloads/stream` | Every persisted download change |
+| `/config/stream` | Peer and server config changes |
+
+Each sends its current snapshot immediately on connect, then a full snapshot per
+change — no deltas to reassemble — plus a `ping` event every 15 seconds so idle
+proxies don't drop the connection. The Logs page tails `/logs/stream`, which
+works the same way except each event is a single log line rather than a
+snapshot.
+
+The browser reconnects on its own if a stream drops; Overview and Downloads show
+a **Live** badge that flips to **Reconnecting…** while it's down.
+
+**Behind a reverse proxy:** SSE needs response buffering **off** and an idle
+timeout **longer than 15 seconds** on whatever sits in front of the UI. jack
+sends `X-Accel-Buffering: no` (nginx and compatible proxies honour it) and the
+BFF forwards that header, but proxies that buffer regardless — or cut idle
+connections early — will leave the UI stuck on **Reconnecting…**. See
+[Troubleshooting](/guide/troubleshooting#management-ui-stuck-on-reconnecting).
+
 ## Management API
 
 The UI's backing API — the **management API** — is a separate listener on its
 
@@ -1,5 +1,5 @@
 ---
-description: Diagnose jack connectivity, Radarr or Sonarr registration, qBittorrent client checks, downloads, imports, permissions, and peer errors.
+description: Diagnose jack connectivity, Radarr or Sonarr registration, qBittorrent client checks, downloads, imports, permissions, management UI live updates, and peer errors.
 ---
 
 # Troubleshooting
@@ -126,6 +126,50 @@ depends_on:
 (This needs `healthcheck` blocks on those services — the linuxserver.io images
 ship with them.)
 
+## The completed folder keeps growing
+
+Everything imports fine, but `downloads.completedPath` never shrinks — every
+title you've ever grabbed is still sitting there.
+
+That's the default behaviour, not a bug. Importing doesn't consume the file:
+Radarr/Sonarr read it and write your library copy, and jack's copy stays put.
+Whether that costs you real space depends on your *arr — a hardlinked import
+shares bytes with the library, a copied one doesn't — but either way the folder
+fills up.
+
+**Fix:** turn on
+[`downloads.unlinkImportedFiles`](/reference/configuration#downloads-unlinkimportedfiles),
+in `config.jsonc` or from **Settings → Downloads** in the management UI. It
+applies immediately, with no restart. jack then unlinks its copy as each import
+is confirmed — safely: only on a confirmed import, only that one file, never
+your library. Details in [After the import](/guide/how-it-works#after-the-import).
+
+This only affects downloads imported **after** you enable it. Files already in
+the folder are yours to clear out — deleting the download rows from the
+management UI's Downloads page removes their files too.
+
+## Management UI stuck on "Reconnecting…"
+
+The Overview and Downloads pages show a **Live** badge while their
+[event stream](/guide/management-ui#live-updates) is connected. If it sits on
+**Reconnecting…** — or flips to it every few seconds — the stream is being
+closed or buffered before it reaches the browser. The pages still load; they
+just stop updating on their own.
+
+Three usual causes:
+
+- **A reverse proxy is buffering the response.** SSE only works if the proxy
+  streams bytes through. jack sends `X-Accel-Buffering: no` and the UI's BFF
+  forwards it, which nginx honours; for others turn buffering off explicitly
+  (`proxy_buffering off;` on nginx, `flush_interval = -1` on Traefik, Caddy's
+  `reverse_proxy` streams by default).
+- **The proxy's idle timeout is too short.** Streams are idle between changes,
+  with only a `ping` every 15 seconds. Give the UI route a read/idle timeout
+  well above that — 60 seconds or more.
+- **The management API is down or unreachable.** The BFF answers `503` when it
+  can't reach jack. Check `docker compose ps` and the backend logs for
+  `Management API listening`.
+
 ## Seeing what jack is doing
 
 Set `LOG_LEVEL=trace` to log every HTTP request (method, path, status,
 
@@ -34,5 +34,5 @@ features:
     details: No trackers, no swarm — files transfer over plain, authenticated HTTP.
   - icon: 🖥️
     title: Web management console
-    details: Manage servers, peers, API keys, and downloads without hand-editing config files.
+    details: Manage servers, peers, API keys, and downloads without hand-editing config files — transfers and connector state update live.
 ---