
Every AI assistant that runs entirely on someone else’s server takes your location, camera, and voice data with it.
The problem with phone-based AI has always been who owns the pipeline
Running a capable AI agent on mobile has meant either accepting a closed cloud system or stitching together clunky workarounds that break the moment you need real device sensors. There was no clean middle path between full privacy control and a genuinely useful mobile assistant.
Your phone becomes a sensor node, not the brain
OpenClaw Releases iOS and Android Companion Node Apps That Connect a Phone to a Self pairs a self-hosted Gateway, which you run on macOS, Linux, or Windows via WSL2, with lightweight companion apps on iOS and Android that contribute camera, location, voice, and a Canvas interface. The phone never processes sessions or routes commands on its own. Nodes connect to the Gateway over WebSocket on port 18789 and require explicit approval before any privacy-sensitive tool goes live.
Privacy-first operators are the first to feel the difference
- Security engineers who need an AI agent that never touches a third-party cloud and want full audit control over every tool call
- Developers building multi-device agent workflows who need a real mobile sensor layer without rebuilding the entire stack
- Researchers handling sensitive data who need persistent memory and local model support without routing through a subscription API
The self-hosted AI agent category has accelerated sharply in 2025, with projects like Home Assistant’s AI integrations and local-first LLM runners pushing the market toward infrastructure users own. OpenClaw now adds a mobile sensor layer that most self-hosted stacks still lack entirely, which pulls capable agentic workflows away from closed platforms faster than most vendors anticipated.
What you can actually do once the node is paired
- Send camera images directly to your agent for real-time visual context
- Trigger location-aware tasks from your phone without exposing data to third parties
- Use voice input on mobile routed through your own Gateway and model
- Manage agent skills and plugins across WhatsApp, Telegram, Signal, and Discord from one Gateway
Both iOS and Android companion apps are free to download. The Gateway runtime requires Node 24 or Node 22.19 and runs on hardware you already own. Pricing not listed for hosted tiers — check our directory.
OpenClaw requires you to run and maintain the Gateway yourself, which is a real technical barrier for non-developers.
Teams that want a managed alternative can look at Beam AI for hosted agentic workflows. For local-first setups without mobile nodes, Ollama combined with Open WebUI covers most of the same ground minus the phone sensor layer.
Self-hosted AI is growing a mobile layer the closed platforms cannot easily copy
The architecture gap between cloud-locked assistants and truly private agent networks is narrowing in a direction that favors operators who move early.
We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.