What it does
TaskHandoff is a self-hosted workbench for running AI coding agents, with Codex and OpenCode named as the supported agents. A control plane manages nodes (your local machine or remote ones) and instances, and sessions for either agent can be created and supervised from one place. Tasks run in instances, which you create from a Git repository or a folder on disk with either a Local or a Docker runtime. Local Runtime uses tools and directories on the host, and each host user can run only one Local Runtime instance, so the site points to Docker for multiple isolated instances. Instances and worktrees let several tasks run in parallel without trampling each other's files or branches, and work carries on if you close the browser because tasks live in the instance, not the tab. You review tool activity and Git changes, then deliver through a branch or worktree. A Stories feature keeps long-running work and its context together and is used to orchestrate repeatable workflows. The same control plane can be opened from the desktop app, the web, mobile or a chat bridge; the site advises setting up authentication and HTTPS before using remote access. The site also offers an agent skill that teaches a connected agent how to drive TaskHandoff.
Capabilities
- Codex and OpenCode sessions in one workbench
- Local or Docker runtime for each instance
- Isolated workspaces using instances and worktrees
- Git change review before delivery
- Local and remote nodes under one control plane
- Stories for repeatable, long-running workflows
When to use it
- You want to run Codex or OpenCode on your own machines or server.
- You need several agent tasks in parallel without branch or file clashes.
- You want to review diffs and tool activity before agent changes ship.
- You want to supervise remote agent tasks from desktop, web or mobile.
