anywhere-labs/Agents-Anywhere

Control any coding agent on any device — from your phone.

What it solves

Agents Anywhere solves the problem of needing to be physically at your computer to manage and interact with AI coding agents. It allows users to remotely monitor, control, and intervene in agent sessions running on a separate machine (Mac, Windows, Linux, or cloud sandbox) using a mobile device or web browser.

How it works

The system acts as a remote control layer rather than a new execution environment. It consists of three main components:

  1. Client: A Web console or Android app where the user interacts with the agent.
  2. Server: A FastAPI-based backend that handles authentication, session state, and RPC (Remote Procedure Call) forwarding.
  3. Connector: A local daemon (Desktop App or CLI) installed on the machine where the agent is actually running. The Connector interfaces with the local Agent Runtime (like Codex or Claude Code) and the local workspace files/terminal, forwarding instructions from the server to the local machine.

Who it’s for

It is designed for developers who use AI coding agents and want the flexibility to approve tasks, check logs, preview code, or run terminal commands on their development machines while away from their desks.

Highlights

  • Unified Control Plane: Manage multiple sessions across different devices from one interface.
  • Runtime Integration: Deep support for Codex and basic support for Claude Code, with an extensible adapter system for future agents.
  • Remote Workspace Access: Ability to browse files, read/write code, and access an interactive remote shell.
  • Non-Intrusive: Does not proxy model credentials or move code to a central server; everything runs locally on the connected device.