Agent Manager

Agent Manager

Runs coding agents such as Claude Code, Codex and Gemini CLI from one private browser workspace, so you can follow several tasks side by side and continue from another device.

Made by lvwerra on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Agent Manager is a private workspace in your browser for AI coding agents that normally run in a terminal. Supported agents include Claude Code and Codex, plus Gemini CLI, opencode, Hermes, OpenClaw and fx. You keep several tasks in view, read each conversation or open its terminal, and carry on from another device while your Space keeps running.

Who it suits: Developers who use more than one command-line coding agent and want them in one place.

Industry | Who uses it | How they use it | Example
Software development | Developers using several coding agents | Run agents side by side in one workspace | Giving Claude Code and Codex the same folder
Solo work | People who switch between computers | Pick up a running session from another device | Checking a long task from a second machine
AI tooling | People who compare agents | Group sessions by project and watch them in Overview | Seeing which agent has finished and which needs a reply

If you do not use command-line coding agents, or you need a service with its own login for several people, you can skip it.

Where to use it:

• A web browser, on a Docker Space you duplicate on Hugging Face
• Windows, macOS or Linux with Docker, for a copy on your own computer

Use case | Who | What they get back
Run several tasks at once | Developers | Sessions grouped by project and shown side by side
Read an agent's work | Reviewers | A conversation reader with formatted replies and search
Work on shared files | Small projects | A file browser, preview, editing and uploads next to the agents

How to set it up: Before you start, have a Hugging Face account and at least one supported coding agent you can use. The app runs as a Docker Space. A private bucket mounted at /data keeps your files, sign-ins and chat history across restarts.

• A Hugging Face account
• At least one coding agent that Agent Manager supports
• A private storage bucket

1. Open the Agent Manager template on Hugging Face and choose Duplicate this Space from its menu. Set visibility to Private.
2. Open the Storage Buckets settings of your new Space. Create or pick a private bucket, then mount it at /data with read and write access. Finish this before any agent sign-in.
3. Once the Space is up, open the app. A first-run window shows which agents are ready and helps you connect one. You can sign in from inside a session, or save the provider's API key as a secret on the Space.
4. For Codex, use the device code sign-in option in its session, because the ChatGPT sign-in cannot be reached from a Space.
5. Send your first task.

For a copy on your own computer, these are Hugging Face's own steps.

pip install -U huggingface_hub
hf download lvwerra/agent-manager-template --repo-type space --local-dir agent-manager-template
cd agent-manager-template
docker build -t agent-manager-template .
docker run -it -p 7860:7860 agent-manager-template

How to use it: Add a session for an agent, send it a task, and add another session to review the result. Group sessions to keep both in view, and use Overview to return to all of them. The Docker run command above serves the app on port 7860.

docker run -it -p 7860:7860 agent-manager-template

Tool | What it does
Overview | Shows which agents are busy and which replies are waiting on you
Conversation reader | Shows replies with formatting, tool activity you can expand, search and older history
Live terminal | Lets you work in the agent's own terminal
File browser | Browses workspace folders, previews and edits files, and takes uploads
Skills | Keeps reusable skills in one place for every agent
Settings, General, Agents | Updates each agent CLI without rebuilding the Space

Example: You start a Claude Code session and a Codex session on the same folder and ask the first, "Add tests for the login page." You ask the second to review that work. Overview then shows which session is busy and which one is waiting for you.

For newcomers and experts: Think of it as a control room for coding agents that lives in a browser tab. To start, duplicate the Space, keep it private, connect one agent and send a small task. Later you can add a second session next to it.

Under the hood it is a Docker app with a server and a web front end. Agents can find, message and wait for other sessions through a local API. Groups only organise the interface, so they do not separate files or credentials. Reader support depends on each agent's transcript format.

Duplicate the Space and keep it private
Mount a private bucket at /data
Connect an agent with sign-in or a key
Run sessions and watch them in Overview

Good to know:

• License: Apache 2.0
• Cost to run: each agent subscription or API plan, the Space hardware and the storage are billed by their own providers
• Data it reaches: workspace folders, uploaded files, agent logins and conversation history on the mounted bucket, plus any API keys you save as Space secrets
• Not for: public access, because the app has no login of its own, so anyone who can reach it gets its shell and its signed-in agents

On Hugging Face, 14 November 2026

• Likes: 46
• Built with: docker
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/lvwerra/agent-manager-template
• Files: https://huggingface.co/spaces/lvwerra/agent-manager-template/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.