# Using the WebUI — the input box

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

**Part of the conversations 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`.

## What the input box can do

- **⌘ + Enter (Ctrl + Enter on Windows) sends.** Enter inserts a newline.
  **The Enter that confirms Japanese IME conversion does not send**. Hovering the send button shows the shortcut
- Typing `/` shows the skill list, **grouped under the headings マイスキル・グループ・テナント共通・製品同梱,
  each sorted by name**. The 「すべて / マイスキル / 共有 / 製品同梱」 filters at the top narrow by origin
  (← → on the keyboard). Keep typing to filter by name. **The skill button (the slash-in-a-square icon, right of the paperclip) opens the same list
  without typing `/`, and keeps what is already written in the box** (picking from a typed `/` clears what was typed after it). If the chosen skill declares input fields,
  **a form appears before sending**; fill it in, then send
- **ワンタップ — buttons above the input box on the new-conversation screen** that put a chosen skill and a prepared message into the box in one tap
  (for a routine such as recording a travel expense with a photo of the route search). Press 「＋ ワンタップ」 to make one: a name, a skill
  (or 「指定しない」), the message, and 「写真のボタンを出す」. **Pressing one only fills the box — nothing is sent until you press send.**
  Each one's 「…」 has 「編集」, 「リンクをコピー」 and 「削除」. Up to 20 per person; they are your own (others do not see them) and
  **the same on every device** — make them on a PC, use them on a phone. A skill that was deleted or that you can no longer use is not chosen;
  only the message goes in, with 「スキル「…」が見つかりません」
- **A link can fill the box the same way**: `/?skill=<skill name>&prompt=<message>&photo=1` (what 「リンクをコピー」 makes) opens a new conversation with
  the skill chosen, the message written, and the photo buttons shown — usable as a bookmark, an Android home-screen icon, or a link sent to a colleague.
  It still sends nothing by itself. **The app added to an iPhone home screen cannot open such a link** (it always opens its start page) — use ワンタップ there
- **With 「写真のボタンを出す」 (or `photo=1`) the box shows large 「写真を撮る」 and 「写真を選ぶ」 buttons**; they fold away once a photo is attached, and × closes them.
  「写真を撮る」 opens the phone's camera; **on a computer it opens a camera window in the app** (the browser asks for camera permission), and it is not shown when the browser cannot use a camera
- Files can be attached (**any format**, up to 500 MiB). Drag and drop works too — **anywhere on the conversation screen**, not only onto the box. **Files over 60 MiB are sent in parts, and a % shows on the chip while sending**
  **The paperclip — or ⌘U (Ctrl + U on Windows) while typing in the input box — opens the 「添付するものを選ぶ」 window** — search box at the top, tabs 「すべて」 / 「ファイル」 / 「SIROE ページ」 / 「ナレッジ」 /
  「アプリ」 / 「テーブル」 / 「会話」 / 「フォルダ」 below it. Tick items across tabs, then press 「添付する(N)」 to attach them all at once
  (files are not re-uploaded; pages and the others become the same chips as a pasted link). The 「ファイル」 tab browses 「最近」, 「すべて」 (open folders), or a name.
  **「このデバイスから」 in the window picks files on your computer** (what you already ticked is attached too).
  **「カメラで撮る」 in the window attaches one photo** (a paper invoice or receipt, for example): on a phone it opens the phone's camera;
  on a PC it opens a preview window — press 「撮影」, then 「添付する」 (「撮り直す」 to retake). If it says 「カメラの使用が許可されていません」,
  allow the camera for this site in the browser's settings. A phone's camera reads small text far better than a PC's webcam
  **The clarifying-question card and the team-conversation input box open a files-only version of the window** (no tabs)
  (in a channel any file you can open can be picked, including 「自分のみ」 ones — **its scope stays as it is**,
  but once attached, the channel's current participants can open it and the AI there can read it, and it cannot be taken back).
  Even in a conversation with a folder attached, **the agent can read files you attached yourself**
- **Up to 5 images per message are shown to the AI as pictures; from the 6th on they are attached as ordinary files** (not refused — the AI opens them as files)
- **Copying cells or text in Excel or Word and pasting them enters the text**, not a picture of it
- **If an attachment fails to upload, its chip shows why** (e.g. 「通信できませんでした」) **and a 再試行 button**; sending stops with 「送れなかった添付があります。チップを外すと送信できます」 until you retry or remove it (×)
- **Attachments already uploaded stay with the unsent draft** — move to another screen and come back, and they are still on the box. **Removing (×) a file you uploaded from this box before sending also deletes it from the files**; files picked from 「添付するものを選ぶ」 are only taken off the message
- **When 「アップロード前の検査」 is on** (`/settings/compliance`, off by default), each file attached in a conversation with the AI (its input box and the clarifying-question card) is checked **in the browser** for personal or secret information (email, phone, マイナンバー, API keys, `.env` values, labelled names) before sending — the chip shows 「確かめています」. A hit opens a dialog with masked excerpts: 「このまま送る」, 「やめる」, and — for text, CSV, Word, Excel, and PowerPoint files that were checked to the end — 「マスクして送る」, which sends a copy with each hit replaced by an uppercase English category tag (such as `[EMAIL]`) instead of the original, and 「一部マスクして送る」, which sends a copy with part of each hit turned into `*` in the same shape as the excerpt (such as `ya***@example.com` or `090-****-**78`) (PDF cannot be masked). The copy is checked again, and if anything is still found, 「マスクしきれませんでした」 is shown and nothing is sent. Files it cannot read show 「未検査」, large ones 「途中まで確認」; both still send. The same check runs on uploads from 「ナレッジを追加」, where 「要確認を外して送る」 drops only the flagged files, and 「マスクして送る」 / 「一部マスクして送る」 appear only when every flagged file can be masked. **It does not run** on attachments in the team-conversation input box or on drag-and-drop onto the ファイル screen
- **The same setting also checks the text you send** in the AI conversation's input box — the message body and the values typed into a
  skill's input form — once, when 送信 is pressed (not while typing), in the browser. A hit opens the same dialog headed 「入力した文章」:
  「やめる」 sends nothing and keeps the text in the box, 「このまま送る」 sends it as typed and records only the counts in the audit log,
  「マスクして送る」 sends it with each hit replaced by an uppercase English category tag such as `[EMAIL]` or `[NAME]`, and 「一部マスクして送る」
  sends it with part of each hit turned into `*` (such as `ya***@example.com`) (either way it is checked again; if anything remains, nothing is sent).
  **It does not check** skills run directly (直接実行), the team-conversation input box, or free-text answers on the clarifying-question card
- **Pasting a link from this app (a SIROE page, a file, a document, an app, a database table, a conversation) turns it into a chip** —
  copy the URL from the address bar and paste it into the input box. Files and documents join the attachment chips; the others show as
  「ページ 〇〇」-style chips. After sending they show as badges on the message, and **the AI opens what the link points to and reads its content** (so 「これの要点は?」 with a pasted page link works).
  A link you cannot open (deleted, or no access) stays as plain text. An item with 「AI アクセス OFF」 shows that mark on the chip and
  **sending stops with 「AI アクセスが OFF のため添付できません」** — remove the chip (×) to send
- Attaching a folder (the window's 「フォルダ」 tab: choose one, then 「添付する」) — narrows the **scope where the agent looks** for files and documents to that folder.
  It is not a sharing action. One folder per conversation; choosing another replaces it
- Voice input — the browser's speech recognition puts what you say into the input box. **It does not work in some browsers such as Brave**
  (pressing it once shows 「このブラウザでは音声入力を使えません」 and the microphone button disappears). Open it in Chrome or Safari.
  **Not shown on a phone or tablet** — use the microphone on the on-screen keyboard instead
- Recording — records the microphone and the PC's audio. **It is saved temporarily even offline**
- You can press 「応答を停止」 (the stop button) while it is answering, **or press Escape** (unless a suggestion is shown or Japanese input is being converted)
- Choose **「思考時間」** (自動 / なし / 低 / 中 / 高 / 最高, default 「自動」) and **「モデル」** (速い `fast` / 標準 `standard` /
  最上位 `frontier`, default 「標準」) **from one button left of send, which shows both (e.g. 「標準・自動」)**.
  **「自動」 answers greetings and simple questions without thinking and thinks hard (as 「高」) on everything else — only with 「標準」; with the other
  models it is the same as 「高」.** Any other choice applies to every message as is, greetings included. **The chosen values are remembered per conversation** (in this browser) —
  changing them in another conversation does not move this one's values. A new conversation starts from the most recently chosen values.
  **Higher means more expensive and slower.** The tier only affects the main conversation; work handed to a skill
  runs at the tier that skill declares. For 「思考時間」 too, if the procedure declares a cap, the lower one applies
- **In a chat with a custom bot (made by you or your team), the bot's 「モデル」 and 「思考時間」 are used instead.** The button
  shows the bot's values (「〇〇の設定で決まっています」) and cannot be changed there — edit the bot to change them. Among the standard bots, only 「fast」 is fixed (fast tier, low strength); the other four follow the input box
- If a skill chosen with `/` declares a script, 「実行方法」 appears — 「モデルに任せる」 (default) or
  「直接実行」 (runs the script as is, without using the message text or attachments added outside the form;
  files picked in the form's file fields are passed to the script)
- A skill chosen with `/` can show a form above the box; a **file field** in it picks files from this device (one or several, as the skill declares).
  The picked files are attached to the message as usual and also appear as chips, and removing a chip clears the field
- The credential used for execution cannot be chosen from the input box. **Saying something like 「共有アカウントで」 in the conversation
  makes the model propose a switch, and only for the turn where you press the approval card does it switch, for that conversation and you only.**
  It is not remembered permanently (a card appears every time it switches). The default credential is chosen on the settings screen (connection details)
- **Faint gray text while typing** is a suggested continuation — Tab (on touch, 「採用」) accepts, Escape dismisses.
  Admins and managers turn it off for everyone at `/settings/completion`
- **Hovering a message** shows 「編集してやり直す」 (your messages: edit and resend) and 「ここから分岐」 (continue from there in a new conversation)