Skip to content

AI AGENTS

github-actions[bot] edited this page Nov 17, 2026 · 6 revisions

Languages: English | العربية | Čeština | Deutsch | Español | فارسی | Français | हिन्दी | Bahasa Indonesia | Italiano | 日本語 | 한국어 | Nederlands | Polski | Português | Русский | Svenska | ไทย | Türkçe | Українська | Tiếng Việt | 中文


AI agents on your device

A standing SSH endpoint turns your phone into something an autonomous AI agent can operate directly. Any agent that can drive a shell or an SSH connection (Claude Code, aider, an LLM-in-the-loop script, your own automation) gets a real root session on the device, over the network, with no USB tether and none of ADB's self-exposing debug flags.

This page expands on the idea introduced in the README.

Why SSH, not ADB, for agents

ADB announces itself: adb_enabled and development_settings_enabled are world-readable, the connection is logged, and there is no supported way to hide that debugging is (or was) on. An agent working over ADB leaves that trail on the device for any app or anti-fraud SDK to read. VD Dropbear is the private alternative: a normal SSH login, with credentials you control and a banner that reveals nothing.

What an agent can do

Over one connection, an agent sees a normal Linux-like shell and can:

  • read and edit files, flip settings, and manage modules, packages and services;
  • triage logs, pull artifacts, and run diagnostics or forensics on demand;
  • move and sync files both ways over scp / sftp / rsync, for backups, migrations, provisioning;
  • drive a whole fleet of rooted devices from one place, each reachable by its own key.

Because the project is built to be AI-friendly, with a documented configuration file, an explicit runtime env contract, predictable paths and plain-text status, an agent can discover how to use it without guesswork.

Doing it safely

This is root, so treat it like root:

  • give the agent its own credential, not your personal one, and point it at a lower-privilege account (shell) if you want to limit its reach;
  • scope the binds to networks you trust: loopback or a VPN rather than the open Internet (see Configuration);
  • keep password auth off; a key is revocable and never typed;
  • you can revoke a key or stop the daemon at any time from the WebUI;
  • remember a default connection is a full root session: give the agent only the reach it needs.

A minimal setup

  1. Generate a key pair just for the agent.
  2. Add its public key in the WebUI (Authorized keys).
  3. Bind to loopback or your VPN only, on a non-default port.
  4. Point the agent's SSH config at the device and let it work.

Over USB, with no ADB: turn on USB tethering and bind to its USB address for a cable-only link, handy for an agent on a tethered laptop. See the user guide.


This documentation was generated with Claude (Anthropic) and is maintained by VD171. Corrections and improvements are welcome; see CONTRIBUTING.md.

Clone this wiki locally