Droidrun

Droidrun's pitch is narrow and honest: agents need a body, not just an API key. Most agent tooling assumes the work happens behind a web endpoint, but a large share of real tasks — messaging, ride-hailing, delivery, banking, social posting — live in Android apps that assume a phone. Droidrun's answer is to hand the agent an actual cloud Android device with a real identity, stream it, and let the agent act on structured text extracted from the app interface rather than on brittle coordinate-clicking scripts. Two things separate it from the device-cloud crowd. First, mobilerun.framework is open source and publicly benchmarked at 91.4% on AndroidWorld, which gives you a measurable control loop to evaluate instead of a demo video. Second, the cloud product is designed around fleets, not single sessions — the point is running many agents on many phones with streaming and identity attached, which is what an operations team actually needs. Where it fits: teams building agents or RPA flows that must touch Android-only surfaces, and infrastructure companies that want to embed phone control rather than build it. Where it doesn't: if your workflow is web or desktop, a browser agent or a standard automation stack gets there faster and cheaper, and if you don't want to write code, this isn't an out-of-box tool — it's an API and a framework. The honest weak spots are the ones the site leaves blank. There's no published per-device rate, no named underlying model, and no public changelog or release notes with dated entries, so you can't judge reliability from history. Treat the free tier and the open framework as your evaluation path, then scope fleets and support directly.