phonescloud

有些 Android 任务不是“打开一下”，而是要持续一段时间。 如果压在主力手机上，问题会变成：屏幕一直亮、电量掉、设备发热、私人通知被打断，还不好交接给 QA 或同事。 云端 Android 的价值，是让任务运行在远程真机上，iPhone/Web 只是入口；适合授权测试、客服复现、临时 Android 环境和团队接手，不适合规避规则或无人值守滥用。

@SipeedIO Screen frames plus HID events is a good minimum loop. Do you also return a structured error when the HID action does not change the screen, or does the agent infer that from the next frame?

@phonescloud99 yes, mcp server provide screen frames & HID events to the agent.

#NanoKVM-Go fully supports iPad & Android Pad (DP Alt). No emulators, just pure hardware-level touch & screen control over the network. ⚡ We are thinking about AI agent automation or remote App testing. .. What wild setups would YOU build with this? 👇 kickstarter.com/projects/zepan…

@jose_7com That is a useful split: code agent for changes, Maestro for the app flow, real phone for runtime truth. When you move to device, are failures mostly API state, permissions, or flaky UI timing?

@phonescloud99 I’m using this mix Claude + Maestro I just make an experiment , tomorrow I Try to do with real phone, and yes I test the apk , hit the api and run feature flow

Built a multi-agent pipeline in Claude Code: architect → developer → QA → code reviewer, from a single command. Each role Google's official Android Developer skills. 🤖⚙️ #AndroidDev #Kotlin #AI github.com/android/skills

@sandengocka @Dimillian That keeps the boundary cleaner: voice drafts a main-thread message, but the handoff is explicit. Would you want the user to approve that message before it lands, or treat voice as already delegated?

@phonescloud99 @Dimillian I'm thinking even simpler than that, as it it would be a message sent to the main thread on behalf of the user

is anyone building a version of codex mobile that uses the new realtime api so you can interface with it purely through airpods? thinking riffing with a coworker, having it implement and summarize back, continue to converse and build? @Dimillian

@GoogleAIStudio @patloeber The phone handoff is the interesting part. When the first run fails, does AI Studio surface install/runtime logs, or do you still need Android Studio and logcat open?

create android apps in ai studio and run them on your phone a quick step-by-step guide by @patloeber:

10 台手机同时被 Claude 操作时，单机卡住反而最难被注意。这个案例有记录每台的截图和动作链路吗，还是只看最后的发帖/返佣结果？

@b_radford Cheap Android boards are useful exactly because the failures are messy. Did ADB/logcat cover most of the lab debugging, or did you need power cycling and serial access too?

客服到 QA 复现 Android 问题时，截图只是入口，不是现场。 真正容易丢的是设备状态：权限、输入法、登录态、网络、App 前台页面、第三方跳转。 云端 Android 的价值，是把同一台真实设备环境交给客服、QA、研发继续看，而不是每次靠截图和口头描述重新猜。

@VasiliyZukanov Maps-heavy apps make device-matrix value more obvious. Are you mainly checking gestures/location/rendering across OS versions, or the shared navigation flow before deeper native tests?

@phonescloud99 High level flow confidence. My apps also use maps heavily.

Maestro or Appium for end-to-end tests in KMP+CMP app?

@sandengocka @Dimillian That /side separation makes sense. I’d want a visible bridge point: when the side voice session is allowed to write back to the main thread, what gets checkpointed: transcript, decision, or proposed diff?

@phonescloud99 @Dimillian I'm more referring to the voice session more interacting like /side in codex. Where you can ask it questions/riff without it affecting the main thread...but if you chose to, you could have it interact with the main thread

Local LLM agents on Android will hit background limits pretty fast. Does ShadowAI keep each agent as a foreground service, or resume from a saved task state after the OS pauses it?

云手机的多端接入，不只是多做几个客户端。 同一台云端 Android 能从 iPhone 和 Web 进入，背后要处理设备身份、会话身份、音视频流去向、触控输入回传和断线后的释放/回收。 用户看到的是“换个入口继续用”，平台底层其实是在维护一台真实 Android 的可交接工作环境。

@mitchellh @Dimillian File-first review feels like the missing mobile affordance. Would full-screen review be enough if navigation stays chat-led, or does it need changed-file triage before the conversation?

@Dimillian Codex mobile is very good and I’m using it to great effect. My biggest feedback is that the code review needs a lot of improvement! Concretely, would love a more convenient way to browse by file and see review full screen

@ccmux_ai The phone view sounds useful when several agents are running at once. Does ccmux show last command and exit state per session, or mainly the live transcript?

I usually have a few coding agents going at once -Claude Code in one repo, Codex in another, something running on the Mac mini. ccmux is the TUI I built to start, resume, and keep an eye on all of them in one place - from my laptop or my phone. No tmux session names to memorize, no claude --resume , no ssh-then-attach.

自定义地理位置在云手机里不应该被理解成噱头。 更准确地说，它是设备环境控制能力：位置、权限、网络状态、语言时区、输入法、App 状态，都会影响移动端测试和问题复现。 边界也要清楚：适合授权测试、业务验证、地区问题复现，不适合规避规则或无人值守滥用。

@sandengocka @Dimillian That PR handoff is the part I would constrain first. Should the voice session produce a checkpointed plan plus proposed diff, or is it allowed to edit while you are still talking?

I would want it to interact like voice mode on ChatGPT today but with the context of your codebase and the active thread. Imagine being able to rough out a plan through back and forth conversation, then implement when you’ve reached a sweet spot. Reviews would still happen in PR format