giuliastro/harness-remote

What it solves

It addresses the fragmentation of AI coding agents, where each tool (like Claude Code or Codex CLI) operates as an isolated "island" with its own sessions, transcripts, and controls. This makes it difficult to switch between agents for different tasks (e.g., switching from planning to debugging) without manually copy-pasting context, and prevents remote access to these sessions from other devices.

How it works

Instead of replacing coding agents, Harness Remote acts as a local-first control plane. It connects to the machines where your development environments live, discovering and orchestrating the native sessions of installed agents. It provides a continuity layer that allows you to observe, resume, and hand off work between different agents while keeping the source code, credentials, and session data on your own hardware. It uses a machine daemon to expose projects and sessions to a client interface available on desktop, web, or Android.

Who it’s for

Developers who use multiple AI coding agents and want a unified way to manage their sessions, switch between tools without losing context, and remotely monitor or interact with their coding agents from different devices.

Highlights

• Native Fidelity: Maintains the original behavior, tool execution, and session semantics of the underlying agent rather than using a generic abstraction.
• Cross-Agent Handoff: Allows explicit continuation of work from one agent to another, preserving objectives and decisions without merging them into a single "mega-agent."
• Local-First Architecture: Keeps source code, API keys, and subscriptions on the local machine, requiring no cloud uploads.
• Multi-Device Access: Provides a unified interface for managing sessions via desktop, web, or Android apps.
• Broad Integration: Supports agents including OpenCode, Claude Code, Codex CLI, Oh My Pi, and PI.