Summary

mngr addresses the challenge of managing AI coding agents, which are often locked into opaque, expensive cloud services. It provides a Unix-style, Git-like CLI for orchestrating agents across diverse compute environments, from local machines to remote hosts and containers. This empowers users with cost transparency, enhanced security, and composability, enabling powerful agent workflows without vendor lock-in.

How It Works

mngr adopts a "Git for agents" philosophy, leveraging familiar primitives like SSH, git, and tmux. Agents are managed processes running within tmux sessions on designated "hosts" (local, Docker, Modal, etc.). Hosts can be shared among multiple agents and are designed to automatically pause when idle, significantly reducing compute costs. This approach prioritizes user-controlled infrastructure, offering flexibility and avoiding the limitations of managed agent platforms.

Quick Start & Requirements

• Primary Install: curl -fsSL https://raw.githubusercontent.com/imbue-ai/mngr/main/scripts/install.sh | bash (installs uv and mngr).
• Core Prerequisites: ssh, git, tmux, jq.
• Optional Prerequisites: rsync, unison, claude.
• Documentation: Interactive help via mngr --help and mngr ask. Shell completion is automatically configured.

Highlighted Details

• Cost-Effective: Agents auto-shutdown when idle, with costs primarily limited to inference and active compute time. Startup times are typically under 2 seconds.
• Secure: Employs SSH key isolation, network allowlisting, and full container control for agent environments.
• Composable: Supports sharing hosts, snapshotting/forking agent states, direct command execution (mngr exec), and bidirectional data synchronization (mngr push/pull/pair).
• Observable: Provides direct SSH access (mngr connect) for debugging and maintains message transcripts (mngr transcript).

Maintenance & Community

The provided text does not detail specific contributors, sponsorships, or community channels (e.g., Discord, Slack). It is structured as a monorepo containing core mngr code and several related plugins.

Licensing & Compatibility

The license type and any compatibility notes for commercial use or closed-source linking are not specified in the provided README content.

Limitations & Caveats

The snapshotting feature is explicitly marked as "experimental." The README does not detail other potential limitations, unsupported platforms, or known bugs. The absence of explicit licensing information represents a significant adoption blocker requiring clarification.