FlowMux：为AI编程Agent打造的Linux终端多路复用器，集成浏览器与智能通知

FlowMux is a GTK4-based Linux terminal multiplexer designed specifically for AI programming agents. It integrates WebKitGTK browser tabs, smart desktop notifications, session persistence, and agent lifecycle hooks, supporting seamless integration with Claude Code, Codex, and OpenCode. This tool addresses the limitations of traditional terminal multiplexers (like Tmux) that are not optimized for AI agents' needs.

Terminal multiplexers have evolved from GNU Screen to Tmux, Zellij, etc., but traditional tools are designed for human interaction, not AI agents. AI agents generate large outputs, need long-running sessions, require user approvals, and execute parallel tasks. Traditional tools fail to support these scenarios: developers either check terminals frequently or miss important agent status updates.

FlowMux's design follows the principle of adapting to agents' workflows. Key decisions include:

1. Agent native notifications: Built-in lifecycle hooks for Claude Code, Codex, OpenCode convert agent events (task completion, approval needed, errors) into native desktop notifications, routed to the correct workspace and silenced if the pane is focused.
2. Browser integration: WebKitGTK 6.0 browser tabs share the pane tree with terminals; agents control browsers via IPC sockets (DOM snapshots, clicks, input) without separate Chromium processes.
3. Session persistence: Agent sessions survive app restarts, restoring workspaces to their previous state.

FlowMux integrates three key functions:

• Terminal multiplexing: Sidebar workspaces, shared pane tree for terminal/browser tabs, VTE4/Ghostty backends, SSH support via russh.
• Embedded browser: WebKitGTK 6.0 engine, agent control via IPC, cookie import from major browsers (Firefox, Chrome, etc.), side-by-side with terminals.
• Smart notifications: Lifecycle hooks for agents, workspace routing, focus-aware silencing, cross-window isolation, OSC sequence support.

FlowMux is written in Rust using GTK4 and libadwaita, offering memory safety and performance. It uses a modular crate structure:

• flowmux-core: Domain types (Workspace, Pane, etc.)
• flowmux-config: Configuration handling
• flowmux-terminal: Terminal backends
• flowmux-browser: WebKitGTK integration
• flowmux-notify: Notification system
• And others (ipc, ssh, vcs, etc.) This modularity ensures maintainability and extensibility.

FlowMux supports three AI agents:

• Claude Code: Captures events like task completion, approval requests, errors, and progress updates.
• Codex: Supports key event notifications.
• OpenCode: Provides integration support. Integration is managed via flowmux doctor (audit status) and flowmux fix (repair components), both idempotent commands.

Use cases:

• Multi-task agent development (managing concurrent sessions).
• Browser automation testing (agent-controlled clicks/input).
• Long-running tasks (reliable notifications on completion).
• Approval-intensive workflows (timely alerts to avoid blocking).
• Linux desktop users (GTK/GNOME ecosystem preference).

Relation to cmux: FlowMux is an unofficial GPL-3.0-or-later reimplementation of cmux (a macOS AppKit app), extending its concept to Linux while ensuring legal compliance and open-source commitment.

FlowMux represents a shift toward agent-native tools. Future trends may include:

• Deep agent integration instead of surface adaptation.
• Redesigned notification and state management.
• Seamless browser/external system connections.
• Session persistence and context recovery. Its open-source code provides a foundation for expanding agent support, integrating more tools, or porting to other platforms.