DeepSeek Harness Remote
DeepSeek Harness Remote 可將手機或另一台電腦與你運行 DeepSeek Harness 的電腦安全連接,隨時隨地繼續你的工作,方便隨時查看進度並繼續工作。
AI 翻譯
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證For when you just need to be heard
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗Nod is a small iOS app. You open it, and you see a quiet orange face that blinks, slowly, once every few seconds. Underneath, two words: I'm listening.
You type what's on your mind — or tap the mic and talk. A small AI model, running entirely on your phone, reads what you wrote and writes back the way a careful friend would. No advice. No cheerleading. No productivity journey. No "how can I help you today?"
Sometimes, you don't even want a reply. There's a little orange face under the message field — tap it, and the eyes blink slowly, once. No words. Just the quietest possible "I heard you."
That's the whole app. That's why it's called Nod.
Most apps that get your words want to do something with them. Analyze them. Classify them. Score your habits. Nudge you toward their KPI.
Sometimes you don't need any of that. Sometimes you just want to talk to something that listens — not a tool that's trying to fix you, convert you, grow you, or mine you.
We wanted to build that. A small chat app that's gentle, unintrusive, and architecturally incapable of leaking a single word of what you say.
No servers. No accounts. No API keys. No login. Every word you type stays on the device, read only by a model running on the same chip. Out of the box, that model is Apple's own on-device LLM — FoundationModels, part of Apple Intelligence. Nothing to download. It just works.
If you'd rather run an open-weights model on your phone, open settings and switch the engine to Qwen 3.5 4B (iPhones with 8 GB of memory) or the lighter Qwen 3.5 2B (6 GB and up), both via MLX Swift. When you switch for the first time, Nod downloads the weights once (about 3 GB or 1.75 GB, Wi-Fi by default). After that, inference is fully offline.
We don't see what you write. No one does. Not because we promise — because there is no pipe to send it through.
The app has no analytics, no telemetry, no tracking pixels, no third-party SDKs for ads, attribution, or A/B testing. Your conversation sits in a local SQLite database on your device, encrypted at rest by iOS. Tap "Start fresh" and it's gone, irreversibly.
When you open Nod, it blinks. And waits. There's no tutorial, no suggested prompts, no sign-in flow. Start typing when you're ready. Or don't. Nod is in no rush.
Every line is MIT-licensed and sits in this repo. The Swift, the system prompts, the database migrations, the website copy — all of it. If "trust us" bothers you, verify us. Read the code. Build it yourself if you want.
The whole behavior of the app is shaped by a single instruction we give the model. Here's a piece of it:
Your job is not to solve problems. Speak like a real person who cares. Never cheerleade. Real presence doesn't announce itself.
Nod will be listed on the App Store as Just Nod — the name Nod alone was taken, and besides, it matches the in-app button.
Requires an iPhone with Apple Intelligence — iOS 26.0 or later. Works instantly on Apple's on-device model — no extra download needed. If you switch to an open-weights engine (Qwen 3.5 via MLX) in settings, Nod downloads the selected model once (1.75–3 GB, Wi-Fi only by default). Otherwise, nothing is downloaded; nothing leaves the device.
Available on the App Store as Just Nod.
.
├── ios/ Native iOS app — see ios/README.md for build instructions
├── website/ Marketing site at usenod.app
├── prompts/ LLM prompts, shared across platforms
└── LICENSE MIT
For the iOS app's internal architecture, build steps, and engineering
notes, see ios/README.md. For a repo-wide overview,
see ARCHITECTURE.md.
We welcome contributions. Quickest way in:
good first issue
or help wanted.CONTRIBUTING.md for setup, branch
conventions, what we're looking for, and what we've decided not
to build.SECURITY.md — do not
file security problems as public issues.By participating, you agree to follow our Code of Conduct.
Speculative Dynamics Private Limited, a small company in India. If you'd like to reach us: [email protected].
MIT. Do what you want with the code. Attribution is kind but not required.