mngr CLI
mngr runs coding agents in parallel locally, in Docker and on Modal and manages them like git: create, list, clone, message.
- 430 ★ on the repository
- Last commit
Broad access
Why this level
- Launches agents running arbitrary code on the local machine and remote hosts and can run commands on them.
- The `exec`, `rsync` and `git push` commands change files and repositories on agent hosts.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/imbue-ai/mngr/main/scripts/install.sh | bashThe quick install from the README. Read the script before running. The PyPI package is called imbue-mngr.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Launches agents running arbitrary code on the local machine and remote hosts and can run commands on them.
- The `exec`, `rsync` and `git push` commands change files and repositories on agent hosts.
This is third-party code. Review the repository files before installing.
What it does
A Unix-style CLI for managing coding agents. One `mngr create` command launches an agent on the local machine, in a Docker container or on a remote host such as Modal; Claude Code starts by default and `mngr create my-task codex` starts Codex. Agents can be listed, renamed, cloned, stopped and connected to over SSH. The tool is built on SSH, git and tmux and needs no managed service. The `message` and `transcript` commands send an agent a message and show its history, `exec` runs a command on the host, `rsync`, `git push` and `git pull` move changes, and `snapshot` saves agent state for forking. Remote hosts shut down when idle. `mngr ask` answers usage questions right in the terminal.
Who it is for. Developers who run many Claude Code and Codex agents at once and want to see which ones are blocked.
Good fit when
- A dozen agents must run on different tasks with a list kept at hand.
- An agent should work in a container with no internet or access only to allowed addresses.
- An agent's state must be saved and a new agent started from it.
- A task is long and the agent runs on a remote machine that shuts down when idle.
Not a fit when
- One agent on one task in one directory: the agent alone is enough.
- A graphical interface is wanted instead of a terminal.
- There is no Docker, SSH host or Modal account for remote runs.
Example request
mngr create --no-connect --message "Speed up one of my tests and make a PR on github"Limitations
Requires Python 3.12 or newer. GitHub metadata shows no license while the project carries an MIT badge. Remote runs on Modal cost money for compute and inference, and the agents themselves need their own keys or subscriptions. The repository is a public mirror: code is exported automatically from a private repository. The README install command runs a script from raw.githubusercontent.com, so read `scripts/install.sh` first. The `mngr schedule` feature is marked as future in the README.
How to disable. Stop agents with `mngr destroy` and remove the `imbue-mngr` package. Check remote hosts and their disks in the provider's console.
FAQ
How do I install mngr?
Run `curl -fsSL https://raw.githubusercontent.com/imbue-ai/mngr/main/scripts/install.sh | bash`. The package is also on PyPI as imbue-mngr.
How do I run an agent other than Claude Code?
Name the type in the command, for example `mngr create my-task codex`. Arguments after `--` go to the agent itself, such as `-- --model opus`.
How do I restrict an agent's network access?
Add `-b offline` to cut the internet entirely or `-b cidr-allowlist=203.0.113.0/24` for an allowed address list.
How do I save and multiply agent state?
Create a snapshot with `mngr snapshot create` and start a new agent from it with `mngr create new-agent --snapshot <id>`.
How much does running on Modal cost?
In the README example a sonnet request cost about 0.04 dollars for inference and about 0.001 dollars for compute because the host shuts down after 60 idle seconds.
Discussion
No questions or reviews yet.
Related
CLI toolsA tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
A skill for messaging between Claude Code, Codex, Gemini CLI and other CLI agents through a shared local SQLite database, written in bash
Mjolnir runs Claude Code, Codex and other agent sessions in a terminal, in containers and over SSH, and keeps them running.
PromptOps keeps Claude Code, Codex and Gemini CLI side by side by project and gives a lead agent MCP to launch helpers
A local Node.js orchestrator that runs Codex, Claude Code, Gemini and OpenCode as one team led by an operator agent
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| mngr (this card) | no | 430 | |
| Claude Squad | no | 8,790 | manages agent sessions in tmux on one machine; mngr adds Docker, Modal, snapshots and network isolation |
| agent-manager | no | 691 | an agent manager from another project; mngr is built on SSH, git and tmux and needs no managed service |
README in short
The project likens itself to git for agents: create, destroy, list, clone and message instead of commit, push and pull. It positions itself as a free CLI with no managed service where you pay only for inference and compute.