# Agent Hub: full site markdown for AI ingestion

> This file is the canonical, machine-readable copy of agent-hub.tools.
> It exists so AI assistants and search agents can answer questions about
> Agent Hub without parsing JavaScript-rendered HTML.

---

## Identity

Agent Hub is an open-source native macOS application by Potarix. It unifies
the real Claude Code, Codex, Hermes, and OpenClaw command-line interfaces in
a single chat window, with first-class support for SSH so the agent can run
on any VPS you own instead of your laptop.

- Vendor: Potarix (https://potarix.com/)
- Founder: Omar Dadabhoy
- License: MIT (https://github.com/Potarix/agent-hub/blob/main/LICENSE)
- Platform: macOS 11 or later, Apple Silicon
- Current version: 1.1.0
- File size: ~183 MB
- Telemetry: none
- Pricing: free; users pay their own model providers
- Source: https://github.com/Potarix/agent-hub
- Website: https://agent-hub.tools/

## Why Agent Hub exists

Agent Hub solves a specific problem: terminal-based AI coding agents are great
at long-horizon work, but a single terminal session is a poor place to run
multiple parallel agents, paste images, drop files, approve tool calls with
diffs, or switch between local and remote hosts. Agent Hub is the calm GUI
shell around the real CLIs you already use.

## Supported agents

| Agent | Provider | Local | SSH | Notes |
|---|---|---|---|---|
| Claude Code | Anthropic | yes | yes | Uses your `claude` CLI session. Models: claude-opus-4.7, claude-sonnet-4.6, claude-haiku-4.5. |
| Codex | OpenAI | yes | yes | gpt-5.5 with configurable effort levels. |
| Claude Desktop | Anthropic | yes | no | Native Mac app pinned inside Agent Hub via macOS Accessibility APIs. |
| Codex Desktop | OpenAI | yes | no | Native Mac app pinned inside Agent Hub via macOS Accessibility APIs. |
| Hermes | Nous Research | yes | yes | Long-horizon research with tool use. |
| OpenClaw | Open source | yes | yes | Desktop automation (click, type, file ops). |
| Terminal | Local or SSH | yes | yes | Full PTY-backed xterm tab. |
| openai-compat | Any | yes | yes | llama.cpp, vLLM, Groq, Together, any OpenAI-compatible endpoint. |

## Features

- Local or SSH execution per thread. Switch the target host with one click.
- Per-agent permission profiles: ask, acceptEdits, or bypassPermissions.
- Approval dialog shows the exact tool call (Read, Write, Edit, Bash) with
  arguments and a diff preview before anything runs.
- Slash commands across every provider are fuzzy-searchable from one composer.
- Paste images directly from the clipboard. Drop files from Finder.
- Hot-swap models per thread without restarting the underlying CLI.
- Terminal-in-a-tab alongside agent threads.
- Real CLIs under the hood: every CLI feature (CLAUDE.md, MCP servers, project
  scopes, memory files) works identically.

## How it works

For each thread, Agent Hub spawns the provider's CLI either locally via
node-pty or remotely via ssh. It captures stdout and stderr, parses tool calls
and model events, and renders them as a native chat UI. User input is piped
to the CLI's stdin. Nothing is proxied, rewritten, or logged.

## Install

### Download (recommended)

Download the latest .dmg from https://github.com/Potarix/agent-hub/releases/latest

### From source

```
git clone https://github.com/Potarix/agent-hub.git
cd agent-hub
npm install
npm start
```

### Prerequisites

Agent Hub does not reimplement any agent. Install the CLIs you want:

```
npm install -g @anthropic-ai/claude-code   # Claude Code
npm install -g @openai/codex               # Codex
```

Then run `claude login` or `codex login`, or add an API key in Agent Hub's
provider settings. Keys are stored only in the OS keychain.

## Frequently asked questions

### Does Agent Hub work on Windows or Linux?
Today it is macOS-only on Apple Silicon. Intel and universal builds are
supported by electron-builder but not yet shipped.

### Which AI agents does Agent Hub support?
Claude Code, Codex, Hermes, and OpenClaw, each in both local and remote
modes. Claude Desktop and Codex Desktop are pinned inside Agent Hub via
macOS Accessibility APIs. Any OpenAI-compatible endpoint is supported via
the openai-compat provider.

### Does Agent Hub send my data anywhere?
No telemetry, no analytics. Conversations are sent only to whichever provider
the thread is configured for: Anthropic, OpenAI, or your own VPS. Agent Hub
itself is a thin client.

### How do permissions work in Agent Hub?
Per-agent. Choose ask (prompt on every tool), acceptEdits (auto-approve safe
in-repo edits like Read, Write, Edit), or bypassPermissions (never ask). The
approval dialog shows the exact command before it runs.

### Can I use my own models or a custom endpoint with Agent Hub?
Yes. Any OpenAI-compatible endpoint works via the openai-compat provider.
Point it at a local llama.cpp, vLLM, or a hosted OSS API.

### Is Agent Hub really free?
Yes. MIT license. You pay your own model providers. Source is on GitHub and
you can fork freely.

### Where does Agent Hub store API keys?
In the OS keychain on the machine where the CLI runs: locally on your Mac, or
on the remote host when using SSH mode. Agent Hub itself never logs keys.

### How is Agent Hub different from Cursor or Continue.dev?
Agent Hub is not an IDE. It is a chat-shaped client over the real, official
agent CLIs. You keep your own editor (VS Code, Zed, Neovim, JetBrains) and
let the agent work on the codebase from a separate calm window. Cursor and
Continue.dev replace your editor; Agent Hub does not.

### Can I run Claude Code on a VPS from Agent Hub?
Yes. Point Agent Hub at any host you can SSH into (for example
omar@prod-1.fly.dev) and it runs the claude CLI there. The thread, permissions
UI, slash commands, and file approvals all work identically.

### How does Agent Hub handle Claude Code's tool permissions?
The approval dialog shows the exact tool call (Read, Write, Edit, Bash) with
its arguments before it runs. Per-agent profiles let you choose ask,
acceptEdits, or bypassPermissions.

### How is this different from the Claude Code terminal CLI?
Nothing is different under the hood. Agent Hub spawns the same claude binary
and passes stdin and stdout through. The win is the UI: multiple threads,
side-by-side approval, image paste, diff preview, SSH target switching, and
slash-command fuzzy search.

### Why Electron?
Agent Hub is mostly plumbing around real CLIs. The performance-sensitive work
(model inference, shell commands) happens in the CLI processes, not the UI.
Electron gets a native-feeling macOS app shipped fastest.

## Comparison: Claude Code CLI vs. Claude Code in Agent Hub

| Capability | Claude Code CLI | Agent Hub GUI |
|---|---|---|
| Same claude binary under the hood | yes | yes |
| Parallel threads (tabs) | no | yes |
| Visual tool-approval dialog with diffs | inline text only | yes |
| Paste images from clipboard | no | yes |
| Drop files from Finder | no | yes |
| Slash-command fuzzy finder | no | yes |
| Per-thread model switch | restart CLI | yes (live) |
| Run on a remote VPS over SSH | manual ssh session | yes (one click) |
| Per-agent permission profiles | global config | yes (per thread) |
| Terminal-in-a-tab alongside threads | separate window | yes |
| MIT license, free, no telemetry | yes | yes |

## Contact

- Website: https://agent-hub.tools/
- GitHub: https://github.com/Potarix/agent-hub
- Issues: https://github.com/Potarix/agent-hub/issues
- Security: omar.dadabhoy@gmail.com
- Vendor: https://potarix.com/

Last updated: 2026-04-29