One place to talk to all your agents
SaferSkills independently audited agent-hub (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<div align="center"> <img src="assets/icon.png" alt="Agent Hub app icon" width="128" />
<h1>Agent Hub</h1>
<p><strong>One place to talk to all your AI agents.</strong><br/> Claude Code, Codex, Hermes, and OpenClaw in a single native macOS app — local or over SSH.</p>
<p> <a href="https://agent-hub.tools/"><strong>agent-hub.tools</strong></a> · <a href="https://agent-hub.tools/claude-code-gui">Claude Code GUI</a> · <a href="https://github.com/Potarix/agent-hub/releases/latest">Download</a> · <a href="CONTRIBUTING.md">Contributing</a> </p>
<p> <img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-blue.svg" /> <img alt="Platform: macOS 11+ (Apple Silicon)" src="https://img.shields.io/badge/platform-macOS%2011%2B%20%7C%20arm64-black.svg" /> <img alt="Built with Electron" src="https://img.shields.io/badge/Electron-32-47848F.svg" /> <img alt="No telemetry" src="https://img.shields.io/badge/telemetry-none-success.svg" /> <img alt="GitHub stars" src="https://img.shields.io/github/stars/Potarix/agent-hub?style=social" /> </p> </div>
Agent Hub is an open-source native macOS app that unifies the AI agent CLIs you already use — [Claude Code](https://docs.claude.com/en/docs/claude-code) (Anthropic), [Codex](https://developers.openai.com/codex) (OpenAI), Hermes (Nous Research), and OpenClaw — into one calm, keyboard-first chat window.
Point it at your laptop and it spawns the agent CLI locally. Point it at [email protected] and it runs over SSH. Same UI, same thread history, same slash commands, same permission model. The terminal version of these agents is great; Agent Hub is what you reach for when you want multiple parallel threads, visual tool-approval with diffs, image paste, and a sidebar of remote hosts.
claude, codex, hermes, and openclaw binaries. Every feature the CLI has — CLAUDE.md, MCP servers, memory files, slash commands — works identically.ask, acceptEdits, or bypassPermissions. The approval dialog shows the exact tool call and arguments before anything runs./plan, /review, /compact, /memory, /cost — every command registered by every provider, keyboard-searchable.claude-opus-4.7 → claude-sonnet-4.6 → gpt-5.5 without restarting.openai-compat provider.<p align="center"> <a href="https://agent-hub.tools/"> <img src="https://agent-hub.tools/hero-app.png" alt="Agent Hub showing Claude Code, Codex, Hermes and OpenClaw threads in one native macOS window" width="840" /> </a> </p>
| Agent | Provider | Local | SSH | Notes |
|---|---|---|---|---|
| Claude Code | Anthropic | ✓ | ✓ | Uses your claude CLI session. claude-opus-4.7, sonnet-4.6, haiku-4.5. |
| Codex | OpenAI | ✓ | ✓ | gpt-5.5 with effort levels. |
| Claude Desktop | Anthropic | ✓ | Native Mac app pinned inside Agent Hub's content area. Glued to window minimize/restore. | |
| Codex Desktop | OpenAI | ✓ | Native Mac app pinned inside Agent Hub's content area. Glued to window minimize/restore. | |
| Hermes | Nous Research | ✓ | ✓ | Long-horizon research with tool use. |
| OpenClaw | Open source | ✓ | ✓ | Desktop automation (click, type, file ops). |
| Terminal | Local/SSH | ✓ | ✓ | Full PTY-backed xterm tab. |
openai-compat | Any | ✓ | ✓ | Point at llama.cpp, vLLM, Groq, etc. |
Grab the latest .dmg from GitHub Releases. Requires macOS 11+ on Apple Silicon. ~183 MB.
git clone https://github.com/Potarix/agent-hub.git
cd agent-hub
npm install
npm startConvenience launcher (detached, closes terminal):
./launch.sh
# or
npm run start-bgCreate an unsigned local DMG/ZIP build:
CSC_IDENTITY_AUTO_DISCOVERY=false npm run dist:macRelease signing, notarization, and in-app update publishing are documented in RELEASE.md.
Agent Hub doesn't re-implement any agent — it spawns the real CLIs. Install the ones you want:
npm install -g @anthropic-ai/claude-code # Claude Code (Anthropic)
npm install -g @openai/codex # Codex (OpenAI)Then claude login / codex login, or drop an API key into Agent Hub's provider settings.
Optional desktop apps. If you'd rather use Anthropic's Claude.app or OpenAI's Codex.app, install them and Agent Hub will pin them inside its window automatically. First launch will prompt for System Settings → Privacy & Security → Accessibility access (required to position the app's window).
New thread: ⌘N, then pick a provider and a target host (local or any SSH host you've configured).
Slash commands: type / in the composer. Fuzzy-search to narrow.
Tool approvals: when an agent wants to run a tool, a dialog shows the exact call. ⌘↵ approves, ⌘⌫ denies. Per-agent profiles let you auto-approve safe edits.
Switch models: /model in any thread, or click the model pill in the header.
Per-agent permission modes:
ask — prompt for every tool call.acceptEdits — auto-approve safe in-repo edits (Read, Write, Edit).bypassPermissions — never ask. Use sparingly.┌──────────────────┐ ┌─────────────────────┐
│ Agent Hub (UI) │ stdio │ claude / codex / │
│ native macOS app │ ─────► │ hermes / openclaw │
└──────────────────┘ pty │ (real CLIs) │
│ └─────────────────────┘
│ ssh
▼
┌──────────────────┐
│ any VPS you own │ ← agent runs HERE, not on your laptop
└──────────────────┘For each thread, Agent Hub spawns the provider's CLI (locally via node-pty, or remotely via ssh), captures stdout/stderr, parses tool calls and model events, and renders them as a native chat UI. Your messages go into the CLI's stdin. Nothing is proxied, nothing is rewritten, nothing is logged.
Does it work on Windows or Linux? Today it's macOS-only (Apple Silicon). Intel and universal builds are supported by electron-builder but not yet shipped.
Does Agent Hub store my API keys? No. Keys live in the OS keychain on the machine where the CLI runs — locally on your Mac, or on the remote host in SSH mode.
Does it send any data to a server? No. Zero telemetry. Conversations go only to whichever provider that thread is configured for.
Can I use my own models? Yes. Any OpenAI-compatible endpoint works via the openai-compat provider.
Can I add a new provider? One file in providers/ — see CONTRIBUTING.md. providers/openai-compat.js is a good starting point.
Why Electron? Agent Hub is mostly plumbing around real CLIs — the perf-sensitive work (model inference, shell commands) happens in the CLI processes, not the UI. Electron gets us native-feel on macOS fastest.
Pull requests welcome. See CONTRIBUTING.md for the full guide. If you're adding a new provider, look at providers/openai-compat.js first — it's the cleanest template.
Found a bug? Open an issue.
Security issue? Please email [email protected] instead of opening a public issue. Details in SECURITY.md.
MIT © Omar Dadabhoy / Potarix
<p align="center"><sub>Made for the terminal-folk.</sub></p>
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.