OrbitPilot Cloud
Ingests tasks, tracks lifecycle, registers devices, runs schedules and queues, stores screenshots and recordings, enforces safety policies, routes human confirmations, and exposes agent-facing APIs.
See scenariosProduct
Navigate from agent intent to dependable actions on real apps and devices
OrbitPilot is the cloud control plane for AI agents. Connect Claude, Codex, OpenClaw, or your own stack through SKILLS.md, HTTP tools, or APIs. The cloud layer handles memory, scheduling, permissions, logs, retries, and state; AutoPilot executes taps, text, scrolls, and screenshots on Android hardware you register.
How it works
From agent instruction to a verified outcome on hardware
OrbitPilot connects agent reasoning to real apps: the cloud plane accepts work, supervises policy and state, and coordinates devices; AutoPilot is the hands on the phone. The sequence below follows the immediate-run flow in the product requirements, plus observation and recovery between steps.
AI reasons. OrbitPilot schedules and supervises. AutoPilot executes. The device is the execution surface.
- 01
Agent invokes OrbitPilot
Claude, Codex, OpenClaw, Cursor, or your own runtime calls OrbitPilot through SKILLS.md, an HTTP tool wrapper, REST APIs, an SDK, or webhooks, sending a goal and structured automation steps.
- 02
Cloud accepts and routes
OrbitPilot Cloud ingests the task, tracks lifecycle, applies permissions and safety rules, picks an eligible online Android device, and places work on the queue when needed.
- 03
AutoPilot acts on the device
The device runtime keeps a WebSocket to the cloud, opens apps, performs taps and text entry, scrolls, captures screenshots, and returns what is on screen so the service can see progress.
- 04
Observe, retry, or pause
After critical actions the control plane reviews screenshots and status, retries transient failures, writes logs, and stops for human confirmation before high-risk moves such as publishing or outbound messages.
- 05
Results return upstream
Completion, failure detail, timelines, and optional confirmation prompts go back to the agent and Dashboard so operators or automation can continue, approve, or stop the run safely.
Platform
One pipeline from agent tool calls to supervised device work, with observability and recovery built in.
AutoPilot
Keeps a realtime channel to the cloud, performs UI actions, returns page observations, reports heartbeat, and surfaces busy or idle state so schedulers can place work safely.
See scenariosAgent integration
Let agents list devices, launch apps, send structured steps, schedule follow-ups, and read status through SKILLS.md, an HTTP tool wrapper, REST APIs, SDKs, or signed webhooks.
See scenariosDashboard and safety
Google or email access, English and Chinese UI, light and dark themes, device inventory, task and schedule views, screenshot timelines, confirmation queues, and API keys for builders.
See scenariosAndroid
Install AutoPilot on your device
AutoPilot is the on-device runtime that keeps a WebSocket to OrbitPilot Cloud and executes taps, text, scrolls, and screenshots through the accessibility service.
APK download is not configured for this environment yet. Build from the repo or ask your operator for a release link.
Before you connect
- Android 8.0 or newer on a physical phone or emulator
- Register the device in the Dashboard and copy the one-time device token
- Enable the OrbitPilot accessibility service when the app prompts you
- Allow notifications on Android 13+ so the foreground session can stay alive
Scenarios
Immediate runs, recurring jobs, and human gates share the same task model described in the product requirements.