Agent Device

@thymikee My machines make me wait sometimes while they run/debug/fix my react-native-macos apps with agent-device ❤️

How to build Android emulator cloud for AI agents 1. Create new droplet 2. Install Android SDK 3. Create Android Emulator 5. Run: agent-device proxy 6. Share auth token with agents using agent-device 7. There’s no 4.

@callstackio pretty cool @expensify ngl

Agents reason about code well, but they lack proper tooling to see results of code changes running live on a device. Integrating agent-device into the development loop gives the agent hands and eyes on devices and simulators, so they can reproduce bugs, record the proof, measure performance, and profile renders on its own. We wrote about how Expensify uses agent-device to do that. Read the article here 👇

After a mobile agent writes the code, how does it verify the app? @thymikee and @lukebfarrell talk through the workflow and where @agent_device fits: simulator access, UI checks, logs, and a second QA pass when teams want it ⬇️

Remote iOS Simulator control from a Linux agent container. One command on the Mac: » agent-device proxy Here’s a remote Claude container connecting to my Mac through a cloudflared tunnel and controlling Maps on an iOS Simulator like it’s local.

In the next version of agent-device: audio support Your agent already has "eyes" and "hands". Now it will have "ears" too. PR: github.com/callstack/agen…

PSA: React Native Best Practices skill just got a big upgrade: → agent-first profiling via @agent_device and @callstack/inspector → version-aware RN, @expo, FlashList, Legend List → fewer misleading guidance and brittle benchmarks → smaller footprint » npx skills update

The best mobile agent loops need the right tools → agent-device.dev Verify changes in real mobile devices → github.com/callstackincub… Detect relevant skills from package.json → claude -p / codex exec Get second opinion from a different agent

Manual QA eats hours. Every login, every form, every screen tapped by hand. So I handed it off to an AI agent. CallStack's open-source agent-device drives a real React Native app and runs the boring tests for me. No scripts to maintain. Watch it work: youtube.com/watch?v=dfVG_a…

The tests we’re running here are skillgym by @ChmalSzymon: reading instructions for LLM (skills or help cmd) and asserting on the output, tool calls, skill reads. It’s pretty cool incubator.callstack.com/skillgym/

ok gpt-5.4-nano is pretty cheap and surprisingly decent at following instructions: - 96 test suites run - 81 passed - 5 mins total (parallelized run) - $0,0075 per test

hello, world!