Claudia — Tauri Desktop GUI for Claude Code

Introduction

Claudia is a desktop GUI for Claude Code built on Tauri 2 + Rust + React. It wraps the Claude Code CLI in a native window so you can browse sessions, design custom agents, run code in OS-level sandboxes, and roll back to checkpoints — all without leaving the editor or staring at terminal output.

What Claudia Does

• Visualizes every Claude Code session as a navigable timeline.
• Lets you author CC Agents — custom system prompts, tools, and run policies — that you can save, share, and replay.
• Sandboxes each agent execution at the OS level (filesystem and network rules) so a misbehaving agent cannot escape its scope.
• Provides checkpoints: every meaningful diff is snapshotted so you can rewind a session to any earlier state.
• Tracks token spend per session, agent, and project, surfacing it alongside historical analytics.

Architecture Overview

The shell is Tauri 2 (a Rust runtime that hosts a webview), the UI is React + TypeScript + Tailwind, and Bun is the build/dev runner. The backend invokes the local claude CLI as a subprocess and streams its stdout and stderr into the UI. Custom agents are stored as Markdown system-prompt files plus a JSON manifest; sandbox rules are enforced via OS primitives that Claudia configures before spawning the CLI.

Self-Hosting & Configuration

• Requires Rust toolchain, Bun, and an existing Claude Code install on the same machine.
• bun run tauri dev for hot reload, bun run tauri build for a signed local binary.
• Agent definitions live under ~/.claudia/agents/ — copy/paste or git-sync to share between machines.
• Sandbox profiles are declarative; per-agent overrides go in the agent manifest JSON.
• Checkpoints stored in a project-local SQLite database — disposable, no cloud dependency.

Key Features

• Native desktop window — no browser tab, no Electron bloat (Tauri uses the system webview).
• First-class custom agents with their own permissions and tool whitelist.
• OS-level sandboxing instead of relying on Claude Code built-in tool gating alone.
• Full session history with diff-aware checkpoints and one-click rollback.
• Per-project usage analytics — input and output tokens, plus cost over time.

Comparison with Similar Tools

• Claude Code CLI alone — fastest to start, but no GUI, no agent library, no rollback UI.
• Cursor / Continue — full IDE forks; larger surface, but tied to a specific editor.
• Open WebUI for Claude — browser-based; cheaper to deploy, less integrated with the local CLI.
• VS Code Claude extension — lives inside the editor; does not manage long-running agents or sandboxes.
• Aider — terminal-first pair programming; complementary, not a GUI replacement.

FAQ

Q: Does Claudia replace the Claude Code CLI? A: No — it shells out to the CLI. You need claude installed and configured separately.

Q: Are sessions stored in the cloud? A: All session and agent data is local (SQLite plus filesystem). Cloud usage is whatever Claude Code itself does.

Q: Can I run my own agents alongside the defaults? A: Yes — that is the headline feature. Custom agents inherit Claude Code tool set and add their own constraints.

Q: Windows support? A: macOS and Linux are the documented targets; Windows builds compile but are less battle-tested.

Sources

• https://github.com/getAsterisk/claudia
• https://claudia.asterisk.so