Robert McDermott has released Astrum v0.11.1, an open-source terminal multiplexer that integrates a persistent AI agent directly into the shell workflow. Unlike standalone AI coding assistants, Astrum operates inside your existing terminal environment, supporting macOS on Apple silicon and Linux on x86-64 and ARM64 architectures. The tool is distributed as a single binary via a curl installer, requiring no sudo privileges or additional dependencies.

Persistent Sessions and Mouse Support

Astrum functions as a session manager where tabs, panes, and shells continue running in the background even after the window closes. This persistence extends across system reboots, restoring directories, splits, and scrollback buffers automatically. The interface is explicitly designed for mouse interaction, allowing users to click tabs, drag dividers, and access context menus, while retaining full keyboard support for power users who prefer traditional terminal controls.

Agent Integration and Security Controls

The built-in AI agent works within the user's real shell session, proposing commands that are visible before execution. Users can configure the agent's autonomy levels, ranging from asking for permission on every command to running read-only operations automatically or operating on autopilot. Security is handled by hiding keys and passwords from the model before any data is sent, and the agent supports various backends including Ollama, Anthropic, Amazon Bedrock, and any OpenAI-compatible endpoint.

MCP Support and Multi-Agent Management

Astrum integrates with Model Context Protocol (MCP) servers, allowing the agent to use external tools via OAuth authentication. Each tab hosts its own independent agent, and a centralized notification system aggregates alerts from long-running builds, bells, or coding agents waiting for permission. This multi-agent view allows developers to monitor and respond to needs across different sessions without switching windows manually.

Key Takeaways

  • Astrum v0.11.1 is open source and installs as a single binary without sudo.
  • Sessions persist across reboots and dropped SSH connections.
  • The AI agent respects shell environment variables and hides secrets from the model.
  • Supports MCP servers and skills defined via SKILL.md folders.

The Bottom Line

Astrum offers a pragmatic approach to AI in the terminal by prioritizing transparency and session persistence over flashy, disconnected chat interfaces. For developers already invested in tmux or zellij, the integrated agent and mouse support present a compelling reason to switch if you want AI assistance that actually understands your current shell context.