org card: links to the website and the Mac app, Chinese section, ModelScope mirrors

README.md

@@ -11,16 +11,41 @@ pinned: false
   <img src="https://raw.githubusercontent.com/deskmind-ai/deskmind/main/brand/social/README-bilingual-light.svg" alt="DeskMind 得心 — 得心，应手。" width="640">
 </p>
 
-**DeskMind · 得心** — open-source models and tools that let an agent see your screen, decide the next step and act on
-your own Mac, locally. 开源、本地运行的电脑操作 agent：看得清，想明白，稳稳做。
+<p align="center"><b>Small enough to run on your Mac. Smart enough to ask.</b><br>
+小到能在你的 Mac 上跑，聪明到知道该问你。</p>
+
+<p align="center">
+<a href="https://deskmind.dev">Website · 官网</a> ·
+<a href="https://github.com/deskmind-ai/app/releases/latest">Mac app</a> ·
+<a href="https://github.com/deskmind-ai">GitHub</a> ·
+<a href="https://github.com/deskmind-ai/bench">Bench</a>
+</p>
+
+DeskMind is open-source computer use for your Mac. It reads the screen, works out the next step and acts, with small
+models running locally with MLX. When a task could mean two things, it asks instead of guessing.
 
 | Model | What it does |
 |---|---|
-| [brain-0.8b](https://huggingface.co/deskmind/brain-0.8b) | Fast tier: typed, calibrated step decisions (MLX) |
-| [brain-4b](https://huggingface.co/deskmind/brain-4b) | Strong tier: the steps the router escalates (MLX) |
-| [eyes-4b](https://huggingface.co/deskmind/eyes-4b) | Screen grounding: find the target on a screenshot |
-
-Current Brain release: **G18b** (revision `g18b-q8`). On the real macOS desktop (bench v25, 13 tasks × 3 runs through
-the DeskMind app on an M4 Pro), the 0.8B → 4B router at threshold 0.96 passes **39/39** runs with no false "done".
-Decisions take about 0.5 s when the 0.8B answers and about 3.6 s when the 4B checks (about 70% of steps). The server is
-wire-compatible with `/v1/systemone`. Code: [github.com/deskmind-ai](https://github.com/deskmind-ai).
+| [brain-0.8b](https://huggingface.co/deskmind/brain-0.8b) | Fast tier: decides each step, with a probability for every option (MLX, 8-bit) |
+| [brain-4b](https://huggingface.co/deskmind/brain-4b) | Strong tier: checks the steps the 0.8B is unsure of (MLX, 8-bit) |
+| [eyes-4b](https://huggingface.co/deskmind/eyes-4b) | Screen grounding: finds the target on a screenshot, for apps without an accessibility tree |
+
+**Current release: G18b** (revision `g18b-q8`, also `main`). On the real macOS desktop (bench v25, 13 tasks × 3 runs
+through the DeskMind app on an M4 Pro), the 0.8B → 4B router at threshold 0.96 passed **39 of 39** runs and never said
+"done" early. A decision takes about 0.5 s when the 0.8B answers and about 3.6 s when the 4B checks. Small sample;
+what it still gets wrong is in [brain results](https://github.com/deskmind-ai/brain/blob/main/docs/results.md). Any
+agent can use the planner through `/v1/systemone`.
+
+---
+
+**DeskMind 得心**是在你 Mac 上运行的开源电脑操作 AI：看屏幕、想下一步、动手操作，靠的是本机的小模型（MLX）；任务有两种理解时，
+它先问你，不瞎猜。
+
+- **brain-0.8b**：判断每一步，每个选项都给出概率；**brain-4b**：复核 0.8B 没把握的步骤；**eyes-4b**：在截图上找到要点的位置，
+  用于没有辅助功能结构的应用。
+- **当前版本 G18b**：真实 macOS 桌面上（bench v25，13 个任务各跑 3 次，经 App 运行，M4 Pro），0.8B → 4B 路由（门槛 0.96）
+  39 次全部通过，没有一次没做完就说完成。样本不大，还做不好的地方见
+  [brain results](https://github.com/deskmind-ai/brain/blob/main/docs/results.zh-CN.md)。
+- 国内下载：ModelScope 上有同样的文件（[gxcsoccer/brain-0.8b](https://www.modelscope.cn/models/gxcsoccer/brain-0.8b)、
+  [gxcsoccer/brain-4b](https://www.modelscope.cn/models/gxcsoccer/brain-4b)、
+  [gxcsoccer/eyes-4b](https://www.modelscope.cn/models/gxcsoccer/eyes-4b)）。