Open Source · MIT · v0.4.14

AgentHubA local-first multi-agent desktop tool

One GUI and one CLI to manage AI coding agents: install environments, logins and connections, Skills, usage, and local sessions.

Download See how AgentHub works Open GitHub Releases

Download for macOS Download for Linux

AgentHub home: Overview of which agents are ready, plus local usage — Overview · which agents are ready, plus local usage

01 — Product

What AgentHub is

AgentHub is a local-first multi-agent desktop tool. It manages different AI coding agents through one GUI and one CLI.

When several AI coding agents are used together, install dependencies, logins, Skills directories, and usage stats are usually scattered across tools. AgentHub puts install, login and connection, Skills, usage, projects, and sessions in one place.

It does not replace Claude Code, Codex, Kimi Code, or the other tools. The workbench makes each tool’s differences explicit, then puts common actions together. Install steps and capability limits follow the GitHub repository README and the source.

Claude CodeCCodexXKimi CodeKGrokGPiPWorkBuddyWZCodeZDeepSeek HarnessDKiroKCursor AgentR

AgentHubOne GUI · one CLI

Install
Skills
Projects
Token
Connections

Several AI agents → AgentHub → one place to manage them

02 — Features

Core features

Screenshots are from the current desktop app; email addresses, local paths, API keys, and other private details are covered with white blocks. The sidebar is split into workspace and management; Settings → Features can show or hide sidebar pages. Release packages may lag the working tree — use the matching Release notes. MCP write support, plugin update, mixed vendors, and some protocol conversions are still being built and are marked in progress.

Overview

1 / 9

See which agents are ready, plus usage and cost estimates parsed from local logs. Go to Connections when you need to sign in or switch.

MCP

In progress

Only scans and lists MCP servers already found. It does not install them or change each tool’s settings. Write and manage are not built yet.

Plugins

In progress

Lists plugin packs already installed in Claude, Grok, and Pi. Claude and Grok packs can be installed, uninstalled, enabled, or disabled here. Pi remains list-only. Update is still a proposal.

CLI

Command-line entry with doctor, env, agent, provider, account, skill, usage, backup, run, and config.

03 — Agents

Multi-agent support

Currently manages Claude Code, Codex, Kimi Code, Grok, Pi, WorkBuddy, ZCode, DeepSeek Harness, Kiro, and Cursor Agent. Cursor Agent is hidden from the sidebar and Connections by default, and can be unhidden on the Agents page.

C
Claude Code
npm or native install
X
Codex
npm install
K
Kimi Code
No independent Skills directory
G
Grok
Native install
P
Pi
npm install; requires Node 22.19+
W
WorkBuddy
Native install opens the official site; custom models appended by catalog
Z
ZCode
Native install opens the official site; API keys appended by catalog without replacing other entries
D
DeepSeek Harness
Not the DeepSeek API itself
K
Kiro
Manages kiro-cli: detect, install, login guide, and API Key
R
Cursor AgentHidden by default
Hidden by default from the sidebar and Connections; can be unhidden on the Agents page

DeepSeek Harness is not the DeepSeek API itself. Do not treat Cursor IDE as a separate agent.

04 — Architecture

How it works

The stack is Tauri v2 + React + Rust. Core logic lives in Rust; the desktop UI and the CLI are the same entry points. Each agent has its own install path, config format, and limits. AgentHub takes those differences in.

Desktop UITauri + React

CLIagenthub command

agenthub-coreRust core · shared by the GUI and the CLI

Agent adaptersInstall paths · config · capability differences

SQLite~/.agenthub/

Local FilesSessions · logs · Skills

The GUI and CLI are thin entries; logic lives in agenthub-core

Runs locally
Tauri + React + Rust
GUI and CLI
Windows / macOS / Linux
Data stays on the machine

05 — Local-first

Data stays on the machine

AgentHub handles local data by default. It is not a cloud console, and it does not collect usage by proxying requests.

Local sessions / logsExisting agent files

Read-only parseNo proxy, no request intercept

~/.agenthubSQLite · settings · backups

No cloud upload · credentials follow existing storage and are redacted

Usage parses local data read-only and does not redirect requests

~/.agenthub/: State, settings, logs, and backups
~/.agenthub/backups/: Config snapshots taken before a switch or import
~/.agents/skills/: Shared user-skills library
Project .agents/skills/: Skills that apply only to that project
Logins: Follow existing local storage; the UI, CLI, and logs redact secrets

06 — Quick start

Install and use

Download a desktop installer for your platform, or run from source using the repository instructions.

Windows

NSIS .exe / MSI

Installer and updater are signed. Run from source with run.ps1.

Download

macOS

.dmg

Updater is signed; Apple notarization is not promised. Run from source with run.sh.

Download

Linux

.deb / AppImage

Packages may be unsigned. Auto-update is enabled only when the release manifest includes a signed Linux entry. Run from source with run.sh as well.

Download

Run from source

Requires Node.js LTS, Rust, Git, and pnpm.

Node.js LTS
Rust
Git
pnpm

Open GitHub Releases

The CLI command is agenthubCommon subcommands:doctorenvagentprovideraccountskillusagebackuprunconfigFull steps are in the repository README.

07 — FAQ

Frequently asked questions

What is AgentHub?

AgentHub is a local-first multi-agent desktop tool. It uses one GUI and one CLI to manage AI coding agents: install environments, logins and connections, Skills, usage, and local sessions. It does not replace Claude Code, Codex, Kimi Code, or the other tools themselves.

Which agents are supported?

It currently manages Claude Code, Codex, Kimi Code, Grok, Pi, WorkBuddy, ZCode, DeepSeek Harness, Kiro, and Cursor Agent. Cursor Agent is hidden from the sidebar and Connections by default, and can be unhidden on the Agents page. Per-agent capabilities follow the repository README and the source.

Is data uploaded to the cloud?

No. AgentHub handles local data by default. Usage is parsed read-only from local sessions or logs. It does not intercept requests through a proxy, and it does not upload to the cloud. Local routing does not record request or response bodies.

Can I install MCP servers or plugins?

MCP cannot be installed here. The MCP page only scans and lists services already found; it does not install them or change each tool’s settings. The Plugins page lists packs already installed for Claude, Grok, and Pi; Claude and Grok packs can be installed, uninstalled, enabled, or disabled. Pi remains list-only.

What happens when connecting to another tool is not possible?

Prefer a direct connection or reuse this login. Local routing is used only when the two sides speak different protocols and a working rule already exists; otherwise the UI shows “Not supported” and does not silently forward.

What is Sub2API?

Sign in to a Sub2API site to view and manage API keys by group, then import usable keys into installed agents. The sidebar entry can be shown or hidden in Settings → Features; new installs hide it by default.

How do I install it?

Download a Windows (NSIS / MSI), macOS (.dmg), or Linux (.deb / AppImage) installer. You can also open GitHub Releases. Running from source requires Node.js LTS, Rust, Git, and pnpm.

What is the license?

MIT License. Source: https://github.com/nicechencs/AgentHub.

Download AgentHub

Windows, macOS, and Linux are first-class desktop platforms. The source is released under the MIT License; issues and PRs are welcome on the repository.

MIT LicenseWindows / macOS / LinuxLatest v0.4.14

Download Open GitHub Releases

Download for macOS Download for Linux