Verital
Verital
Verital
讀論文、寫論文、做簡報、錄課堂——都在同一台電腦上，同一條時間軸上。 AI 說的每一句話都指得回它讀過的那一頁；指不回去的，它會說它指不回去。 Read papers, write them, build the deck, record the lecture — one desk, one timeline. Every sentence the AI writes points back to the page it read; when it can't, it says so.
為什麼是這個名字The name
標誌是一束光譜收進稜點，再射出一道白光——那是反過來的稜鏡。 分光是「一件事拆成很多」；收斂是「很多件事變成一個說得清楚的」。 你要的是後者。 The mark is a prism run backwards: eight rays converge to a point and leave as a single beam. Splitting light turns one thing into many; converging turns many into one you can defend.
真相來源是你機器上的那個資料庫，不是我們的伺服器。 拔掉網路，全部功能照樣在——搜尋、筆記、排版、離線的模型。 雲端是管道，不是來源。 The system of record is the database on your machine, not our server. Pull the network cable and everything still works — search, notes, typesetting, on-device models. The cloud is a pipe, never the source.
摘要、問答、考題、簡報，產出的每一段都帶著它讀過的段落， 點一下跳回那一頁那一行。查不到出處的引用會被擋下來， 不會變成一條看起來很像真的參考文獻。 Summaries, answers, quizzes, slides — each carries the passage it came from, and one click jumps to that line on that page. Citations that can't be verified are held back rather than dressed up as real ones.
課堂錄音、你當下寫的那一筆、老師換到那一頁——掛在同一條時間軸上。 點你三週前寫的那個字，錄音就跳到他講那句話的那一刻。 The recording, the stroke you drew at that moment, and the slide that was up — all pinned to one timeline. Tap something you wrote three weeks ago and the audio jumps to the sentence being said.
做得到的事What it does
保留原版面的對照翻譯（公式照排，不變成亂碼）、段落透鏡、 框一塊就問 AI、真的手寫筆記本。 Side-by-side translation that keeps the original layout (formulas stay typeset), paragraph lenses, ask-about-this-region, real ink notes.
跨全部專案的檢索問答、深度研究報告、Podcast、解說影片、 字卡與測驗——選擇、填空、計算，還有 AI 按給分點批改的簡答。 Retrieval across every project, deep-research reports, podcasts, explainer videos, flashcards and quizzes — multiple choice, fill-in, calculation, and short answers graded point by point.
用投稿目標的引擎排版：LaTeX 走 Tectonic 出 PDF、Word 走真正的 docx 編輯器。改稿、追蹤修訂、交叉引用、文獻庫。 Rendered by the engine your target actually uses: Tectonic for LaTeX PDFs, a real docx editor for Word. Revisions, tracked changes, cross-references, a bibliography that checks itself.
每一頁由模型按設計系統重新設計，不是套版。 圖片切成可以拖動的物件，資料圖從你的表算出來。 Each slide composed against a design system rather than poured into a template. Images decomposed into draggable objects; data charts computed from your own tables.
錄音、即時逐字稿、板書轉成可以擦掉的手寫筆記、 作業與考試範圍自動挑出來變成待辦。 Record, transcribe live, turn the blackboard into ink you can erase, and pull homework and exam scope out into your planner.
一條連結，教授不用註冊。內容在你這邊加密，
金鑰只在連結的井號後面——伺服器上沒有一個字看得懂。
One link, no account for them. Encrypted on your machine; the key lives
after the # and never reaches the server, so there is
nothing readable on our side.
方案Plans
付費買的是我們代你去打模型與雲端那一半。 功能不鎖——你用自己的 API 金鑰或自己的帳號登入，一樣跑得動全部東西。 What you pay for is us running the models for you and the cloud half. Nothing is feature-gated: bring your own API key or your own account and everything still runs.
| Free自己的金鑰Bring your own key | ProUS$10／月（sandbox 佔位） | Pro+US$20／月（sandbox 佔位） | |||||
|---|---|---|---|---|---|---|---|
| 本機全部功能 | All local features | ✓ | ✓ | ✓ | |||
| 自己的 API 金鑰／帳號登入 | Your own key or account | ✓ | ✓ | ✓ | |||
| 官方 AI（我們的金鑰） | Official AI (our key) | 試用額度 | Trial credit | 月配額 | Monthly quota | 更高配額 | Higher quota |
| 多裝置同步 | Multi-device sync | — | 兩台都開著的時候（點對點，我們零留存） | While both are running (peer-to-peer, nothing kept by us) | ✓ | ||
| 雲端備份 | Cloud backup | — | — | ✓ | |||
| 加密信箱（批改連結、離線收件） | Encrypted inbox | — | 短期 | Short retention | 30 天 | 30 days | |
| 問卷雲端保留 | Survey retention | 7 天 | 7 days | 30 天 | 30 days | 90 天 | 90 days |
學生與教職員的優惠正在辦。定價確認後會更新這一頁。 Student and faculty pricing is in progress. This page updates when it lands.
FAQ
不會。檔案與索引都在你的電腦上。走官方 AI 的時候， 只有那一次請求需要的片段會送出去，而且送出去之前會先過去識別化。 我們不留提示詞——記帳的表裡連一個放得下提示詞的欄位都沒有。 No. Files and indexes live on your machine. When you use the official AI, only the passages that request needs leave, and they pass through de-identification first. We keep no prompts — the usage table has no column a prompt could fit in.
能。閱讀、筆記、排版、搜尋、本機的語音與 OCR 模型都不需要網路。 需要網路的只有：雲端的模型、多裝置同步、以及每隔一陣子的授權續期 （續期失敗不會馬上把你鎖在外面）。 Yes. Reading, notes, typesetting, search, and the on-device speech and OCR models need nothing. Only cloud models, multi-device sync and the occasional licence renewal do — and a failed renewal doesn't lock you out on the spot.
不用。填自己的 API 金鑰，或用你已經有的帳號登入，全部功能都跑得動。 買方案是因為共用一把上游金鑰時，重複的請求會命中同一份快取—— 同樣的錢比較耐用，而且不用自己管金鑰。 No. Paste your own API key, or sign in with an account you already have, and everything runs. The plan exists because a shared upstream key means repeated requests hit one cache — the same money goes further, and you don't manage keys.
不需要。你發一條連結，他在瀏覽器裡就能讀、能改、能留評語。 他的角色寫在那條連結裡——「只能留言」的人改不動你的正文， 這件事是在你的機器上擋掉的，不是靠他那邊守規矩。 No. You send a link; they read, edit and comment in the browser. Their role is carried by the link — a commenter cannot touch your body text, and that is enforced on your machine, not by trusting theirs.
一堂 90 分鐘的課，原始 WAV 大約 173 MB。備份前會轉成 Opus， 大約 21 MB——差八倍，所以雲端備份的成本不會失控。 A 90-minute lecture is about 173 MB as raw WAV. It's converted to Opus before backup — roughly 21 MB, an eight-fold difference, which is why cloud backup stays affordable.
macOS 先上。Windows、Linux、iPad 的框架已經在專案裡， 但還沒到可以請你依賴它的程度——好了會在這一頁公布，不會先寫在這裡。 macOS first. Windows, Linux and iPad scaffolding exists in the project but isn't at a point where we'd ask you to rely on it. It'll be announced here when it is — not before.