I’ve been building TaskHandoff, a self-hosted workbench for AI coding agents, and I’d like to share it here.

What it is: one web UI where you run Codex or OpenCode against your own local or remote machines. You create workspaces, dispatch tasks, watch the AI session stream, review diffs, and merge or reject the result. The agents run on your hardware, and the orchestrator is something you host yourself, so nothing has to leave your network.

Why self-hosted: the whole point is that repo access and agent credentials stay on your side. Session data, task history and workspaces live in your own instance.

Stack / install: Apache-2.0. If you already have Docker, it’s basically task-handoff install. Docs (EN + ZH): https://docs.thandoff.com/

Source: https://github.com/edgestorage/task-handoff

Disclosure: I’m the author, so this is a self-promo post. Happy to answer questions about the architecture, what runs where, or how remote workspaces are handled. If you try it and something is confusing, tell me and I’ll fix the docs.