# Agent Manager from Huggingface

> Private cloud workspace for running and managing Claude Code, Codex, Gemini CLI and other coding agents from the browser.

Agent Manager is an open-source project from Hugging Face that runs AI coding CLI agents such as Claude Code, Codex, Gemini CLI, opencode, Hermes, OpenClaw and fx in one private browser workspace. It is deployed as a Docker Space on Hugging Face and is licensed under Apache 2.0. Agents keep running in the Space backend after the browser tab is closed.

## What It Is

Agent Manager is a browser-based manager for coding agent sessions. Each session has its own conversation, and sessions can be grouped by project and arranged side by side. An Overview shows which agents are working and which replies need attention. Users sign in with their own provider login or API keys.

## Working With Sessions and Files

Each session can be read in a live terminal or in a conversation reader with formatted replies, expandable tool activity, search and earlier history; reader support depends on the agent's transcript format. Users can browse workspace folders, preview and edit files, upload documents or screenshots, and keep a shell beside the agents. Reusable skills can be managed in one place, and agents can discover, message and wait for other sessions through a local API. Token usage, estimated costs and available quota information are shown, with details varying by provider. Prompts can be scheduled, device notifications enabled, and an agent on another machine can be connected to the same workspace. Each CLI can be updated from Settings without rebuilding the Space.

## Setup Path

Users duplicate the public template Space as a private Space, then mount a private storage bucket at /data with read-write access. A first-run window lists the agents in the image and guides sign-in or API key setup. The same setup can be scripted with the huggingface_hub Python client.

## Privacy and Persistence

Agent Manager has no login of its own, so anyone with access to the app can use its shell and logged-in agents; the documentation says to keep both the Space and its bucket private. A privacy lock helps detect an exposed instance, and the public template has its terminals disabled. A Space sleep, restart or rebuild ends running processes. With the bucket mounted, saved files and checkpointed CLI state survive, and eligible sessions can be restarted and conversations resumed where the CLI supports it. Shared Codex sessions are described as a work in progress and are not enabled by default.

## Features
- Run multiple coding agents side by side in the browser
- Group sessions by project with an Overview of agent status
- Live terminal and conversation reader with search and history
- Workspace file browsing, preview, editing and uploads
- Managed shared skills across agents
- Local API for agents to discover and message other sessions
- In-app CLI updates from Settings
- Token usage and estimated cost tracking
- Scheduled prompts and device notifications
- Remote agent connection from another machine
- Session restart and resume after reboot with a mounted bucket
- Privacy lock to detect exposed instances

## Integrations
Claude Code, Codex, Gemini CLI, opencode, Hermes, OpenClaw, fx, Hugging Face Spaces, Hugging Face Storage Buckets

## Platforms
WEB, API, CLI

## Pricing
Open Source

## Links
- Website: https://huggingface.co/spaces/lvwerra/agent-manager-template
- Documentation: https://github.com/huggingface/agent-manager/blob/main/docs/development.md
- Repository: https://github.com/huggingface/agent-manager
- EveryDev.ai: https://www.everydev.ai/tools/agent-manager-from-huggingface