PaduPadu
View as markdown

Getting started

Padu is a high-performance, native desktop and web workspace for orchestrating local AI coding agents. Built in Rust with GPUI (the GPU-accelerated engine behind Zed), Padu runs directly on your machine, keeping all code, transcripts, checkpoints, and credentials 100% local.

Download the native release for macOS, Linux, or Windows from padu.dev/download or the GitHub releases page.

The desktop app bundles its own lightweight daemon and starts it automatically on a loopback port. No separate server install or cloud account is required.

2. Headless Daemon (Remote Devboxes / Servers)

For remote devboxes, headless Linux servers, or continuous environments, you can run the standalone daemon binary:

# Start the standalone daemon on loopback port 34123
padu-daemon --bind 127.0.0.1:34123

# Or allow remote connections over private VPN / Tailscale
padu-daemon --bind 100.101.102.103:34123 --allow-non-loopback

The daemon can also serve the bundled web client directly. See Self-hosting the web UI.

Configuration and local state live under ~/.padu/ (or your OS user data directory).

3. Prerequisites

Padu manages external agent CLIs; it does not bundle AI models itself. Before launching an agent in Padu, make sure you have installed and authenticated at least one supported agent CLI:

See Supported providers for the full list of supported agents.

Next Steps

  • Workspaces — Understand Padu's workspace, session, and worktree model.
  • Git worktrees — Run concurrent agents in isolated branches.
  • Connectivity — Connect remote web or mobile clients via LAN, Tailscale, Cloudflare Tunnel, or an SSH relay.
  • Performance Architecture — Learn how GPUI delivers 120 FPS rendering with zero UI-thread blocking.
  • Security & Privacy — How Padu protects your source code and credentials.
  • Configuration & Settings — Configure ~/.padu/settings.json and provider binary overrides.