TaskHandoff Unifies AI Workspaces Across Local and Remote Machines

Summary

TaskHandoff is an Apache-2.0 project that provides a unified control plane for running and collaborating on Codex and other AI development work across local and remote machines. It manages node enrollment, workspace and instance lifecycles, AI sessions, applications, repositories, and message routing from a shared web, API, and chat gateway surface. Docker is the primary isolated runtime, with a supported Local Runtime for one controlled instance per host user on eligible non-Windows nodes, while the architecture leaves room for Kubernetes adapters. The system is divided into a Control Plane, a Node Agent, and Controlled Instances; instances continue running when the Control Plane restarts, and each AI Session remains the source of truth for conversation state. Users can create workspaces from catalog images or node-local environment templates, combine them with Git projects or local folders, and retain or delete managed volumes explicitly. Environment templates capture a container's writable layer but exclude workspaces, volumes, credentials, processes, memory, and network state, and derived instances receive new identities and managed resources. The project also supports repository inspection, scoped Git credentials, application management, Telegram, DingTalk, WeChat, and Feishu/Lark integrations, plus iOS, Android, desktop, and systemd server deployments. A beta status is stated, with breaking changes possible between releases; local development requires Node.js 24.15.0 through 24.x, pnpm 9.15.3, and Docker for Docker-based features. Debian and Ubuntu servers can install the Control Plane and Node Agent as separate services, and remote machines need only the Node Agent and a one-time join token. The repository describes signed image-market catalogs, release workflows, runtime recovery behavior, and Apache License 2.0 licensing, but does not claim a particular AI model release or benchmark result.