imbue-ai/mngr

What it solves

mngr is a Unix-style CLI tool designed to solve the complexity of deploying and managing multiple AI coding agents across diverse environments. It eliminates the need for opaque, managed cloud services by allowing users to run agents on their own compute—whether locally, in Docker containers, or on remote hosts like Modal—while providing a unified interface to control, monitor, and sync data with them.

How it works

The tool operates as a management layer built on standard open-source primitives: SSH, git, tmux, and Docker.

• Agents as Processes: Agents are treated as processes running within tmux sessions on a host. Each agent has its own working directory and configuration.
• Host Management: mngr can launch agents on various "providers" (e.g., Modal, AWS, Docker). It uses SSH to access these hosts.
• Lifecycle Control: Users can create, destroy, list, and connect to agents via the CLI. It includes features like idle detection to automatically shut down hosts to save costs.
• State Handling: It supports snapshotting and forking agent states, as well as rsync and git integration for moving code between the local machine and the agent's environment.

Who it’s for

It is built for developers and AI engineers who want full control over their agent infrastructure, prefer a CLI-driven workflow, and need to scale from a single local agent to hundreds of remote agents without being locked into a specific provider.

Highlights

• Provider Agnostic: Run agents locally, in Docker, or on remote cloud providers like Modal.
• Cost Efficiency: Agents auto-shutdown when idle, ensuring you only pay for active compute and inference.
• Infrastructure Control: Full control over network allowlists, SSH key isolation, and container security.
• Composable Workflows: Ability to snapshot, clone, and migrate agent states across different hosts.
• Seamless Data Sync: Built-in rsync, git push/pull, and a pair mode for continuous synchronization between local and remote environments.