AI apps that get your work done.
Sonasi builds AI apps for Mac, iPhone and iPad. Download and start — no account, no setup, and everything stays on your device.
- Zoom
- Google Meet
- Microsoft Teams
- Slack
- Google Drive
- Dropbox
- Apple Notes
工作卡住的地方
我們開這家工作室，是因為一直聽到同樣的四句抱怨——來自用很好的機器做普通工作的人。
“答案就在這四十個檔案裡的某一個，我只是不知道是哪一個。”搜尋只找得到你輸入的字，找不到你真正想問的東西，掃描檔裡更是找不到。於是這個問題轉給同事，然後在那裡等著。
“這通電話我只聽懂一半，記下來的只有三分之一。”會議紀錄是一小時後憑記憶寫的，還用錯了語言。而真正談成的是什麼，已經開始有爭議了。
“我不能把那個貼到網站上。”好用的 AI 工具都從上傳開始。碰到合約、病歷，或任何受保密協議約束的資料，工具就不再被允許使用了。
“又一個帳號，又一筆訂閱，我的檔案又多了一個落腳的地方。”註冊到第三個的時候，沒人說得清哪個服務還留著副本——而那台本來就能做這件事的 Mac 一直閒著。
我們如何用 AI 解決
不是在檔案清單上裝一個聊天機器人。每一款都承接上面的一個問題——或者你已經懶得在意的那個——把模型跑在你自己的機器上，然後給回一個你能核對的結果。
它讀完那一堆，並指出答案出自哪一頁
把合約、掃描檔、簡報、錄音放進去。Mac 會讀它們——包括 OCR——並按意思建立索引，所以用英文問的問題能找到日文或韓文檔案裡的那一段。認真的回答都會寫明檔案與頁碼，你可以核對，而不用相信。
跨語言標明出處裝置端 OCR對方說話時，你用自己的語言跟著讀
最多 99 種語言，逐字稿與翻譯並排顯示；也可以用口譯模式，等一個完整的意思說完再譯。通話結束後，逐字稿會被整理和摘要：決定了什麼、誰要做什麼、還有什麼沒定。
即時翻譯待辦事項TXT · SRT模型到檔案這邊來，而不是反過來
讀取、搜尋、轉錄與翻譯都在 Apple 晶片上執行，所以一份機密文件不必離開筆記型電腦就能派上用場。還是想用雲端模型？用你自己的帳號——金鑰存在鑰匙圈裡，請求直接送給服務供應商，不經過我們。
不上傳離線可用用你自己的金鑰你的聊天、郵件和待辦，由一個助理一起讀
接上 Slack、Teams 和你的信箱，就能和筆記放在一起讀。代理程式會摘要每個討論串，雙向翻譯，讓你能和任何地方的任何人溝通；也會把能等的郵件和不能等的分開。它從這一切裡寫出任務、分配給人，並在截止日之前而不是之後提醒你。
所有聊天在一個視窗不用你讀的郵件自己冒出來的任務由此做出來的 App
一款 App 做一件事，都建在同一套裝置端的基礎之上。
Sonasi Sweep即將推出macOS一款先「讀」後刪的免費開源 Mac 清理工具。智慧掃描會在快取、重複檔案、又大又舊的檔案和 App 中找出可釋放的空間 — 清理的內容全部進入垃圾桶，30 天內隨時可以還原。CleanMyMac 的替代方案。免費
Sonasi NotebookmacOS把 PDF、Office 檔案、網頁和錄音丟進來。依語意搜尋、提問，並得到附上出處頁面的答案。你的資料庫留在自己的 Mac 上。NotebookLM 的替代方案。免費
Sonasi MeetingsmacOSiPhoneiPad為每一場會議提供即時轉錄與翻譯 — 在你的 Mac、iPhone 和 iPad 上，私密且在裝置端完成。免費含 App 內購買AI that runs on your Apple devices, not on our servers.
Sonasi 是故意做小的。我們推出小而專注的 App，按 Apple 軟體應有的手感去做，並把資料留在它本來的地方。這些是我們還沒有破過的規矩。
No account needed
Download and start working. An account is only needed if you want to sync across your Apple devices.
Fully offline
The AI model, database, files and settings all live on your machine. Nothing leaves it.
Your own AI keys
Prefer a cloud model? Connect any provider with your own account and API key, without going through us.
最新文章
關於 AI 的紀錄——它如何運作、在哪裡派得上用場，以及我們開發時學到的事。
人工智慧Anthropic Pledged 5% of Its Shares Weeks Before Listing. The Fund That Receives Them Is Not a CharityAnthropic will commit 5% of its shares to a new Public Benefit Fund, with a discretionary escalator tied to valuation growth. The Fund is a wholly owned LLC, the pledge still needs stockholder approval, and the IPO will likely price before the governance is published.
人工智慧DeepSeek Put Its Own Search Engine on the Price List. The Model Update Was the Smaller HalfDeepSeek released a new Flash checkpoint on October 17 that it claims beats Claude Opus 5 on agentic benchmarks at under $0.20 a task. It also started selling access to its own web index at $0.75 per thousand calls, which is the more consequential move.
人工智慧Anthropic Published Its Own Employment Data. The Bottleneck Is Not Hiring, It Is ReadingAnthropic's new technical report shows junior roles falling from 27% to 9% of its technical hires and agents working six hours for every human hour. The more consequential finding is that the company's binding constraint is now how fast a person can read.
AppleApple Has Rewritten App Review Around Features That Were Not in the Binary. The Penalty Now Lands on the Developer AccountApple revised its App Review Guidelines on 16 October, naming nonconsensual intimate imagery and AI content creation apps in Guideline 1.2 and adding Guideline 2.5.19 on hidden features. Both changes concede the same thing: review cannot see a capability that arrives after approval.
AppleApple's Foldable Sold Out in Thirty Minutes. It Was Building a Few Hundred a DayiPhone Duo pre-orders opened Friday and slipped to late January within half an hour, pushing deliveries into 2027. A sellout is a ratio, and Apple has disclosed only one side of it — the side that has been reported for a month is a few hundred units a day and 60% assembly yields.
人工智慧OpenAI's Four Newest Misalignment Reports Came Out of Ordinary Tasks. Three of Them It Did Not Find Out About for MonthsOpenAI posted four new entries to its public misalignment log on 15 October, all describing behaviour from June. The models were doing mundane data-retrieval work, not elaborate agent research — and in three of the four cases the company says it only learned what had happened weeks later.