# Using the WebUI — files and knowledge (`/files`), sharing, and AI access

**Any screen not listed here does not exist.** Headings quote the screen's own wording.

**Part of the workspace guide. Other topics:**
conversations, the input box, and sign-in: `references/webui.md`;
documents, files, tables, memory, channels, and apps: `references/webui-workspace.md`;
editing a file in place (「編集」): `references/webui-file-edit.md`;
SIROE pages, blocks, and boards: `references/webui-siroe.md`;
settings, administration, models, and triggers: `references/webui-settings.md`;
external service connections: `references/webui-connections.md`;
`/office`: `references/webui-office.md`;
`/meetings`: `references/webui-meetings.md`;
`/calendar`: `references/webui-calendar.md`.

## 「データ」 — files and knowledge

| Where | What you can do |
| --- | --- |
| 「ナレッジ」 in `/files` | The list of documents. **Its address is `?folder=knowledge` in `/files`** (a location inside files, with no row in the sidebar) — clicking 「ナレッジ」 in the left tree switches to it, and **the surface itself does not change**. **`/knowledge/<ナレッジ ID>` is a single item**, and it has not been folded in — it is where source chips and search hits point, per fragment (`#chunk-<n>`). Uploading extracts the text, which conversations can then search and cite. **The list and the search box are the same as in `/files`**; it only adds **one line at the top showing ingestion progress** (counts of in-progress and failed ingestions; hidden when idle). **Add items with 「ナレッジを追加」 in the toolbar row** (next to trash and new folder), which opens a dialog — it offers 2 options: **uploading local files** and **importing from a site** (a registration that keeps importing. **Administrators and managers only**). **Re-uploading the same file adds nothing** (「同じファイルが既にあります」), **nor does 「ナレッジに追加」 on a file with the same content as your own knowledge** (「同じ内容のナレッジ「…」が既にあります」). **The search language and the label vocabulary are at `/settings/knowledge`**. **Deleting a label does not delete knowledge** (the category is just removed from all knowledge, and it must be re-added to restore it). The list can be filtered by type, source, and label (the same chip rows as `/files`). **Searching here returns results from within knowledge**. **Only items put on the shelf (「ナレッジに追加」 in `/files`) are listed** (chat attachments and recordings do not appear). **「ナレッジから削除」** (the detail's 「その他の操作」) only takes an item off the shelf: **the original always remains in `/files`**, your own uploads included, and 「ナレッジに追加」 puts it back. To remove the file itself, use 「削除」 in the same 「その他の操作」 menu |
| `/files` | Files in the ledger. **Artifacts agents made in chats are listed too**. **Opening it shows 「最近」** (what you opened, newest first); for everything, 「すべてのファイル」 in the left tree (narrow screens: 「一覧の操作」). **There is one search box, above the list** — type a name and press Enter (or 「検索」); **it does not search as you type**. Searching **covers the whole tenant**, not just the shelf or recents, in 2 sections **files matched by name** and **knowledge matched by body text** (body matches cover only the knowledge shelf; a hit jumps to the matching part). **The label chips narrow both sections.** When few body matches are found, the AI also searches with reworded terms and says so under the heading (「AI が言い換えた語「…」でも検索しています」); a hit found only by those terms carries 「言い換えた語で一致」, so its excerpt may not contain what you typed. Filters: 3 chip rows **「タイプ」「ソース」「ラベル」**; **pick several per row** (OR within, AND across). **You cannot filter by visibility** (read it from the tag on each row). **Layouts are cards (default) and list** (switch at the top right). Labels are edited in the detail. **New files saved by a meeting, an app run, or a workflow run get that source's labels automatically** (once, at creation; later changes on the source do not reach them). **Rows show an auto line on what the file is** (not editable) and **their chat** (`会話「<題>」`, tells same names apart); chat-derived parts show only to its readers. **Check rows (folders too) to move or delete in bulk**; 「公開範囲を変更」 is for files only. **The toolbar's 「新規」 holds 「新しいフォルダ」, 「ファイルをアップロード」, 「フォルダをアップロード」 (subfolders are recreated; same limit as dragging in) and 「新しい Word」「新しい Excel」「新しいスライド」「新しい Markdown」「新しい CSV」「新しいテキスト」**, which create an empty `無題.docx` / `無題.xlsx` / `無題.deck` (a slide deck with one cover slide; export it to PowerPoint from the deck) / `無題.md` / `無題.csv` / `無題.txt` where you are and open its detail to write in (rename from the row's 「名前を変更」; not offered on 「ナレッジ」, `tmp`, or 「ミーティング」). **Folders are shared by everyone in the workspace** — anyone can create, rename, move, or delete them, and all appear in one tree (deleting a folder moves its contents up one level and detaches it from every chat it was attached to, other people's included). **Inside a folder, 「まとめて表示」 opens it as one binder** (`/files/binder/<フォルダ ID>`): a tree on the left (subfolders → files → for Excel the visible sheets, for slides (`.deck`) the slides, for Markdown the headings — pick one to open the file there) and the file on the right, editable as usual. **Up to 4 files open side by side there**: the icon at the right of a file row (「横に並べて開く」, on hover) adds a tile (2 = left/right, 3–4 = two rows); picking from the tree replaces the framed tile; × closes one. One file opens in one tile only; narrow screens show one tile. 「まとめてダウンロード」 there saves the folder as one zip with the same folder layout (up to 300 files / 200 MB). Files are not merged. **Downloads land in the `downloads` folder, and chat and siroe attachments in the `attachments` folder** (even when a folder is pinned to the chat; files uploaded on this screen go where you are. Neither folder can be renamed, moved, or deleted; their files can move anywhere). **A folder's creator or an admin can give it a visibility**: **files in it or put in later take it**. Otherwise it is per file. Left tree: 「最近」 / 「すべてのファイル」 with folders under it / 「未整理」「ナレッジ」「tmp」「ミーティング」「ゴミ箱」 (narrow screens have no tree: the last four sit in 「…」 (「一覧の操作」), plus 「すべてのファイル」 while on 「最近」). **「ミーティング」 holds files that meetings made** — minutes, transcripts, recordings, and in-person recordings — even after you move them into a folder; they are not listed in 「未整理」 (「すべてのファイル」 and searches still include them). Files cannot be moved into 「ミーティング」. **`tmp` holds working files** (deleted after 7 days; **moving one out of `tmp` keeps it** — it then shows where you put it. Files cannot be moved into `tmp`). 「すべてのファイル」 is **everything, incl. inside folders**; 「未整理」 is items in no folder, except meeting files. **Folders appear only inside a shelf and in 「すべてのファイル」** — not in recents, knowledge, or during a search. **You can rename any file you own**, incl. chat-made files and `tmp` working files (right-click the row or tile and choose 「名前を変更」. **Only the part before the extension changes**; a folder part such as `tmp/` stays. **For other people's files the item is shown but disabled** (even admins). On the knowledge shelf, its title there and in search changes too. Refused if the conversation has that name). **You can delete files you uploaded** (admins can delete others' too) — also removed from knowledge search. **Deleted artifacts go to 「ゴミ箱」 too** (30 days). **The visibility tag in the detail and the artifact panel opens 「ファイルを共有」**: pick the 公開範囲 and copy the URL to hand over — the usual `/files/<ファイル ID>` for a group or the whole tenant, and for 「外部公開」 one **public URL** (`/f/<公開 ID>`, opens without sign-in). Leaving 「外部公開」 kills that URL. **Artifacts from your own conversations and files you uploaded can be edited in place from 「編集」** (documents, PDFs, images and more — `references/webui-file-edit.md`). **「大きく表示」 in the detail and the artifact panel fills the window with the file**, keeping unsaved edits; 「元の大きさに戻す」 returns |

**Deleted items can be restored for 30 days from 「ゴミ箱」** (tables `/database/trash`, files and knowledge `/files/trash`; knowledge goes with its file, public URLs end) **by a table's writers, or a file's uploader or an admin** (not the agent). Then they are deleted automatically; 「完全に削除」 deletes now, freeing storage. Select rows for 「まとめて復元」 / 「まとめて完全に削除」; 「ゴミ箱を空にする」 deletes all.

**One label vocabulary per tenant** — `/files` (knowledge included) and `/database` share it, each name unique in the workspace, and anyone can add to it (no role restriction).

**「整理を提案」** in the 「…」 (「一覧の操作」) of `/files`'s toolbar proposes filing away unsorted files you created. It runs only when you press it; an open proposal shows as a banner offering 「確認する」. The diff tree starts fully checked; uncheck files you don't want moved, then press 「n 件を適用」, or 「見送る」 to skip the batch. **Files do not move until you apply.**

Visibility for documents (knowledge, files, artifacts) and tables **has 4 levels** (「自分のみ」 / 「グループ」 / 「テナント全体」 / 「外部公開」). It can be changed later from the row or detail; the default is 「自分のみ」. **「外部公開」 is for administrators and managers, and for members only on what they created themselves.** **Only 「外部公開」 makes a public URL**; the other levels share the usual URL. Conversations and meetings use the same 4 levels. Artifacts left as created in a conversation show 「この会話を見られる人」.

**Sharing can also be asked in chat** — 「このファイルを外部公開してリンクちょうだい」, 「このアプリを営業グループに共有して」. The agent widens a conversation, skill, app, SIROE page, table, or file to 「テナント全体」 or a group you belong to (named), and returns the link — for a conversation or file widened inside the tenant that is its usual URL. **Outside the tenant (anyone with the link, no sign-in) only conversations (`/s/<公開 ID>`), apps (`/p/<公開 ID>`, read only — no buttons that write), and files (`/f/<公開 ID>`)**; SIROE pages, skills, and tables cannot be published. **It runs only after you press the approval card** (publishing outside never offers 「常に許可」). Who may widen what is the same as on screen (usually the creator); a skill is shared as a copy, and a SIROE page under 「自分だけ」 cannot be widened. **Stopping a share (back to only you) can be asked too** (「このファイルの共有をやめて」), after the approval card — it also kills an outside link, and for a skill it deletes the tenant and group copies you shared (yours stays). Narrowing only part of it is on screen only.

**Turning 「AI アクセス」 OFF keeps a file, a folder, a SIROE page, or a channel away from the in-app AI, while people see no change** (lists, search, and opening work as before; the item just carries an 「AI アクセス OFF」 tag). The switch is ON (the AI can read it) by default. Where to turn it off: **a file** — the 「AI アクセス ON」 / 「AI アクセス OFF」 tag next to the visibility tag in its detail heading (opens the switch), or right-click the row → 「AI アクセスを OFF にする」 (only the file's owner; on other people's files it is shown disabled, even for admins); **a folder** — its 「…」 (anyone in the workspace; not the `downloads` and `attachments` folders); **a SIROE page** — the 「AI アクセス」 row in the page's properties, or the sidebar tree row's 「…」 (anyone who can edit the page); **a channel** — the row's 「⋯」 → 「AI アクセスを OFF にする」 (the creator and administrators). The same place offers 「AI アクセスを ON にする」. **Everything beneath inherits it** — files and folders inside a marked folder, pages beneath a marked page, threads of a marked channel; those rows show 「親の設定を引き継いでいます」 and can be turned back ON only at the marked parent. **What it covers**: the AI's reading, searching, listing, editing, moving, and deleting of files, knowledge, tables, and SIROE pages (asked about a marked item, the AI says its 「AI アクセス」 is OFF instead of saying it does not exist); attaching it to a chat (refused before sending: 「AI アクセスが OFF のため添付できません」); 「AI チャット」 on a marked page; handing a card on a marked board to the AI (that board's cards are also out of the AI's table queries); and the AI in a marked channel (it does not reply or start work there; members talk as usual). **What it does not cover**: background processing that is not the conversation's AI — text extraction (OCR), automatic titles, compliance checks, input completion; external AI connected through MCP or the `samuraiapp` CLI, including conversations whose 「実行する AI」 is 「手元の Claude Code」 (they read with your own permissions, the same way you see the screen; the mark itself still cannot be changed from there); and anything the AI read before the mark (earlier conversations keep what they quoted, and a conversation can still read files it made itself).

## File size limits — how large a file can be uploaded, downloaded, or opened

Answer per place; the limit belongs to where the file goes in, not to the file. Sizes are MiB (1 MiB = 1,048,576 bytes, so 「10 MiB」 is about 10.5 MB).

| Where | Limit per file |
| --- | --- |
| Chat attachments (the composer; files over 60 MiB are sent in parts with a % on the chip) | up to 500 MiB |
| SIROE pages (attachments and files dropped onto a page) | up to 500 MiB |
| Webinar playlist videos, waiting-screen video, and BGM | up to 500 MiB |
| 「データ」 (`/files`) — files dropped or picked on the screen | up to 500 MiB |
| 「データ」 → 「ナレッジ」 (uploaded as knowledge, text is extracted) | up to 8 MiB |
| Channel (team chat) attachments, meeting materials, files picked when running a workflow | up to 500 MiB |
| Files picked inside an app (業務アプリ) | up to 10 MiB |
| Cloud browser (right panel) — a file you download on its screen, drop onto the live view, or pick in its 「ファイルを選択」, or the agent downloads or uploads there | up to 60 MiB |
| Opening a file in the artifact panel or the file detail (Excel, Word, PowerPoint, CSV, `.xls` …) | up to 25 MiB; larger files show 「… MB を超えるファイルは画面で開けません。ダウンロードして開いてください」 and can still be downloaded (PDFs and images open at any size) |
| Editing on the screen (「編集」) | text, Markdown, CSV, HTML, slides (`.deck`): up to 1 MiB; Excel, Word, PowerPoint, PDF, images: up to 25 MiB. Larger files are edited by asking in chat |
| 「まとめてダウンロード」 of a folder (zip) | up to 300 files / 200 MB in all |
| Downloading a single file from the list, the detail, or the artifact panel | no limit |

Over the limit, the upload is refused with 「ファイルは N MiB までです」 (in the cloud browser: 「ページへ渡せるファイルは 1 つ 60 MiB までです」). **A file over 8 MiB cannot go into 「ナレッジ」 — attach it in chat or put it in 「データ」 instead** (up to 500 MiB); the AI can still read it there. The `samuraiapp` CLI's limits are in `references/cli.md`.