Getting Started

Getting Started

This guide takes you from zero to a running agent — either an OpenClaw bot inside your group chat, or a Hermes personal assistant in your terminal — connected to an LLM provider.

The walkthrough below uses OpenClaw as the worked example because it has the deepest Dashboard integration today. If you'd rather start with Hermes, the path is identical through step 4; step 5 differs and is described in Runtime-Hermes.

Prerequisites

• macOS (Apple Silicon recommended) or Linux
• A ChatGPT Plus/Pro subscription (recommended) or an API key from an LLM provider (see provider guides in the sidebar)

Docker is installed automatically by the install script if not already present.

1. Install

curl -fsSL https://clawfleet.io/install.sh | sh

This single command will:

1. Install Docker if needed (Colima on macOS, Docker Engine on Linux)
2. Download and install the clawfleet CLI
3. Pull the pre-built sandbox image (~1.4 GB)
4. Start the Dashboard as a background daemon
5. Open http://localhost:8080 in your browser

Manual install (advanced)

git clone https://github.com/clawfleet/ClawFleet.git
cd ClawFleet
make build
sudo make install
clawfleet dashboard start

This requires Go 1.25+ and Docker to be installed.

2. Dashboard management

The Dashboard runs as a background daemon with auto-restart:

clawfleet dashboard status    # Check if running
clawfleet dashboard stop      # Stop the daemon
clawfleet dashboard start     # Start the daemon
clawfleet dashboard restart   # Restart the daemon

On macOS the daemon uses launchd (auto-starts on login). On Linux it uses systemd user services (or a PID-based fallback).

3. Get the sandbox image

The install script automatically pulls the pre-built image. You can also manage images from the Dashboard:

1. In the Dashboard sidebar, click System > Image
2. Click Pull Image (recommended) or Build Image

Alternatively, from the CLI:

clawfleet build    # Build locally (offline use or customization)

4. Add an LLM provider

Each instance needs an LLM to power it. The easiest way is to use your existing ChatGPT subscription:

Option A: ChatGPT (Codex) — Recommended

1. In the Dashboard sidebar, click Assets > Model Config
2. Click + Add Model Config — ChatGPT (Codex) is selected by default
3. Choose a model (e.g. gpt-5.4)
4. Click Login with ChatGPT — authenticate in the popup window
5. Done — config is saved automatically

No API key needed. Uses your ChatGPT Plus ($20/mo) or Pro ($200/mo) subscription.

Option B: API Key providers

1. In the Dashboard sidebar, click Assets > Model Config
2. Click + Add Model Config
3. Select your provider (OpenAI, Anthropic, Google AI Studio, or DeepSeek)
4. Paste your API key, choose a model
5. Click Test — you must see "Validated" before you can save
6. Click Configure to save

Models are shared — multiple instances can use the same model config simultaneously. You only need one config per provider + model combination.

ChatGPT (Codex) is OpenClaw-only. Hermes uses standard API-key providers; pick Anthropic, Google AI Studio, OpenAI, or DeepSeek for Hermes instances. See Runtime-Hermes for the full provider list.

For step-by-step instructions, see the provider guides: ChatGPT (Codex) ★ | OpenAI | Anthropic | Google AI Studio | DeepSeek

5. Create and configure your first instance

Create

1. In the sidebar, click Fleet > Instances
2. Click + Create Instances
3. Select a Runtime: OpenClaw (default) or Hermes Agent
4. Enter the number of instances (start with 1) and click Create

Instances are named by runtime: openclaw-1, openclaw-2, ... or hermes-1, hermes-2, ... and are grouped under their runtime's section header on the Fleet page.

Each OpenClaw instance is bigger than each Hermes instance — OpenClaw bundles a full XFCE desktop and Chromium for browser-driving skills, Hermes is headless. See the FAQ for measured numbers.

Configure (OpenClaw path)

1. On the instance card, click Configure
2. Select a Model Config from your asset pool
3. Optionally select a Character Config (assigns a persona) and Channel Config
4. Click Configure

ClawFleet will automatically:

• Set up the LLM provider and model
• Render the character into SOUL.md
• Enable and connect the messaging channel (if selected)
• Set DM and group policies to "open" so the bot responds to all messages
• Start the OpenClaw gateway

Configure (Hermes path)

For Hermes instances, ClawFleet's Configure dialog offers a quick model + channel setup; deeper configuration happens in Hermes's own native Dashboard:

1. Click Configure on the Hermes instance card to set model + channel (Discord, Telegram, or Slack)
2. For credentials beyond a single API key, scheduled cron jobs, personality (SOUL.md), terminal backend selection, or a channel beyond those three — click the ⚕ Dashboard button on the instance card to open Hermes's native UI on the host port shown on the card
3. To chat with Hermes from your terminal (TUI mode), run:

   clawfleet shell hermes-1

See Runtime-Hermes for the full Hermes setup walkthrough.

Access the desktop (OpenClaw only)

Click Desktop on an OpenClaw instance card to open the embedded noVNC desktop in your browser — useful for skills that browse, draft on the Canvas, or run cron jobs. Hermes instances are headless; their native Dashboard takes the place of a desktop.

Next steps

• Pick the right runtime for the next instance — read Runtime-OpenClaw vs Runtime-Hermes to understand when each shines, then create instances accordingly.
• Add a messaging channel — connect an OpenClaw instance to Telegram, Discord, Slack, Lark, or any of 24+ channels. See the channel guides in the sidebar.
• Scale up — run OpenClaw and Hermes side by side. The Dashboard groups them by runtime so you can manage a mixed fleet.
• Explore the Dashboard — see the Dashboard Guide for a full walkthrough.
• Visit the website — clawfleet.io for project updates and community links.