Tianshu v0.5: I said I'd build it. Now it runs.

A few months ago I posted a video saying I'd build an open-source, self-hostable AI agent platform from scratch. Back then there wasn't a single line of code — just three things I wanted to fix: make progress visible, real per-user isolation, continuity across devices. This is me keeping that promise. Each one, shown running: A Kanban board + worker pool — tasks actually run in parallel, and failures are marked, not silent. A per-tenant Linux sandbox with a network egress policy — browser, files, credentials never cross. A channel system (WeChat connected first, more coming). Plus a few extras: Workforce Studio — pull your whole agent config (main agent, workers, plugins, prompts) into a single "Solution" you can edit, diff, export, and activate in one click. Manage your agent the way you manage code. Key-free web search + web_fetch (no API keys to wire up). Point-and-click config for models and more. Open source and self-hostable. One npm command runs an instance on your own machine — real browser, real sandbox, real workers. ⭐ github.com/tianshu-ai 📦 npm install -g @tianshu-ai/tianshu