Skip to main content

agentwatch-cli

PyPI version Python 3.9+ License: MIT

Connect your local Moltbot (OpenClaw) gateway to AgentWatch cloud without exposing your local network.

Overview

agentwatch-cli is a secure bridge that allows AgentWatch to communicate with your locally-running Moltbot gateway. Instead of exposing your local network, the connector establishes an outbound Socket.IO connection to AgentWatch cloud and relays messages to your gateway via the OpenClaw WebSocket RPC protocol (chat.send).

Architecture:

AgentWatch Cloud <--Socket.IO--> agentwatch-cli <--WebSocket RPC--> Local Moltbot Gateway
     (cloud)                     (your machine)                       (your machine)

Key Benefits:

  • No port forwarding required - All connections are outbound
  • Secure - TLS encryption, HMAC challenge-response authentication, credentials never stored in plain text
  • Isolated sessions - Creates a dedicated connector session separate from the user's active session
  • Stateless measurement - Session history is cleared after each request to ensure fresh, unbiased context for every question
  • Automatic reconnection - Handles network interruptions and stale WebSocket connections gracefully
  • Service mode - Runs in background, starts on boot

Requirements

  • Python 3.9 or higher
  • A running Moltbot (OpenClaw) gateway on your local machine
  • An AgentWatch account with a custom agent configured

Installation

macOS

Option 1: Using pipx (Recommended)

# Install pipx if you don't have it
brew install pipx
pipx ensurepath

# Install agentwatch-cli
pipx install agentwatch-cli

Option 2: Using pip

pip install --user agentwatch-cli

# Add to PATH (add this line to ~/.zshrc for persistence)
export PATH="$(python3 -c 'import sysconfig; print(sysconfig.get_path("scripts", "posix_user"))'):$PATH"

Linux (Ubuntu/Debian)

Option 1: Using pipx (Recommended)

# Install pipx if you don't have it
sudo apt install pipx
pipx ensurepath

# Install agentwatch-cli
pipx install agentwatch-cli

Option 2: Using pip

pip install --user agentwatch-cli

The script is installed to ~/.local/bin/ which should already be in your PATH.

Linux (Fedora/RHEL)

# Install pipx
sudo dnf install pipx
pipx ensurepath

# Install agentwatch-cli
pipx install agentwatch-cli

Verify Installation

agentwatch-cli --version

If you get "command not found", ensure your PATH includes the installation directory:

# For pip --user installs:
# macOS: ~/Library/Python/3.X/bin
# Linux: ~/.local/bin

# For pipx installs:
# Usually ~/.local/bin (run `pipx ensurepath` to fix)

Quick Start

1. Create a Custom Agent in AgentWatch

  1. Go to AgentWatch and navigate to Custom Agents
  2. Click Add Agent and select Moltbot
  3. You'll receive an 8-character enrollment code (e.g., ABCD-1234)

2. Enroll the Connector

Quick install + enroll (single command):

pip install --user --upgrade agentwatch-cli && python3 -m agentwatch_cli enroll --code ABCD-1234

Test your installation first (no enrollment code needed):

pip install --user --upgrade agentwatch-cli && python3 -m agentwatch_cli enroll --dry-run

Or if already installed:

agentwatch-cli enroll --code ABCD-1234

This will:

  • Register your connector with AgentWatch
  • Auto-discover your gateway token from ~/.openclaw/openclaw.json
  • Create a dedicated connector session in Moltbot (separate from your interactive sessions)
  • Save configuration to ~/.agentwatch-cli/config.json

3. Install as a Service (Recommended)

# Linux (requires sudo)
sudo agentwatch-cli install-service

# macOS (no sudo needed - uses launchd)
agentwatch-cli install-service

That's it! The connector will now:

  • Start automatically on boot
  • Run in the background
  • Reconnect automatically if disconnected

Alternative: Run Manually

agentwatch-cli start

Commands

Command Description
enroll Enroll with an enrollment code from AgentWatch
start Start the connector manually
status Check connector status and gateway connectivity
config Update configuration settings
revoke Revoke enrollment and clear credentials
install-service Install as a system service (auto-start)
uninstall-service Remove the system service
service-status Check if the background service is running

enroll

Enroll the connector with an enrollment code from AgentWatch.

agentwatch-cli enroll --code ABCD-1234

start

Start the connector and begin relaying messages.

agentwatch-cli start

# With custom gateway URL (if not localhost)
agentwatch-cli start --gateway-url http://192.168.1.100:18789

# With explicit gateway token
agentwatch-cli start --gateway-token your-token-here

status

Check the connector status and test gateway connectivity.

agentwatch-cli status

Example output:

AgentWatch CLI Connector Status
========================================
Enrolled: Yes
Agent: My Moltbot
Agent ID: abc123...
Connector ID: def456...

Gateway URL: http://127.0.0.1:18789
AgentWatch URL: wss://agentwatch.helivan.io
Gateway Token: <auto-discovered>

Testing gateway connection...
Gateway Status: ONLINE

config

Configure connector settings.

# Set gateway URL (for remote gateways)
agentwatch-cli config --gateway-url http://192.168.1.100:18789

# Set gateway token explicitly
agentwatch-cli config --gateway-token your-token-here

# Set AgentWatch cloud URL (for self-hosted deployments)
agentwatch-cli config --agentwatch-url wss://your-agentwatch-instance.com

revoke

Revoke enrollment and clear credentials.

agentwatch-cli revoke

# Skip confirmation prompt
agentwatch-cli revoke --force

install-service

Install as a system service for automatic startup.

Linux (systemd):

# Requires sudo
sudo agentwatch-cli install-service

# Specify user (default: current user)
sudo agentwatch-cli install-service --user myuser

macOS (launchd):

# No sudo needed - installs to ~/Library/LaunchAgents
agentwatch-cli install-service

uninstall-service

Remove the system service.

# Linux
sudo agentwatch-cli uninstall-service

# macOS
agentwatch-cli uninstall-service

service-status

Check if the service is running.

agentwatch-cli service-status

Configuration

Configuration is stored in ~/.agentwatch-cli/config.json:

{
  "connector_id": "uuid",
  "secret": "encrypted-secret",
  "agent_id": "uuid",
  "agent_name": "My Moltbot",
  "gateway_url": "http://127.0.0.1:18789",
  "gateway_token": null,
  "agentwatch_url": "wss://agentwatch.helivan.io"
}

Gateway Token Auto-Discovery

If gateway_token is null, the connector automatically discovers the token from ~/.openclaw/openclaw.json:

{
  "gateway": {
    "auth": {
      "token": "your-gateway-token"
    }
  }
}

This is the recommended approach - the token is read fresh each time, so if OpenClaw rotates it, the connector automatically uses the new one.

How It Works

  1. Enrollment: When you run enroll, the CLI exchanges the one-time enrollment code for permanent credentials (connector_id + secret). The gateway token is encrypted (AES-256-GCM) before being stored on the cloud side.

  2. Connection: On start, the connector establishes a Socket.IO connection to AgentWatch cloud and authenticates via HMAC challenge-response (the secret is never sent over the wire). It also connects to your local Moltbot gateway over WebSocket and completes the OpenClaw handshake with operator.admin scope.

  3. Session Isolation: The connector creates a dedicated session (agent:main:agentwatch-connector) in your Moltbot's ~/.openclaw/agents/main/sessions/sessions.json. This session is separate from your normal interactive sessions, so connector activity never interferes with your own conversations.

  4. Message Relay: When AgentWatch needs to measure your agent's behavior:

    • AgentWatch sends a job request through Socket.IO to your connector
    • The connector forwards the question to your local Moltbot gateway using the chat.send WebSocket RPC method
    • The connector waits for the chat event with state: "final" and extracts the response text
    • The response flows back to AgentWatch through the same Socket.IO channel
  5. History Clearing: After each request completes, the connector deletes the session's .jsonl history file. This ensures every question is answered with a clean, empty context — no prior conversation history can influence the response. This is critical for accurate behavior measurement, as each question must be evaluated independently.

Environment Variables

Variable Description
AGENTWATCH_ENROLLMENT_URL Override the enrollment API URL (for testing/self-hosted)

Security

  • Credentials protected: Config file has restricted permissions (0600)
  • HMAC authentication: The connector authenticates using HMAC-SHA256 challenge-response — the secret is never transmitted over the wire
  • Encrypted gateway token: Your Moltbot gateway token is encrypted with AES-256-GCM before being stored on the cloud
  • TLS encryption: All Socket.IO and WebSocket communication uses TLS
  • Outbound only: No inbound ports required — the connector initiates all connections

Troubleshooting

"Cannot connect to local gateway"

Ensure your Moltbot gateway is running and accepting WebSocket connections on port 18789:

# Check if gateway is listening
agentwatch-cli status

If the gateway shows as offline, start your Moltbot/OpenClaw gateway.

"missing scope: operator.admin"

The connector requires operator.admin scope to use the chat.send method. Ensure your gateway token has sufficient permissions. The token auto-discovered from ~/.openclaw/openclaw.json should have this scope by default.

"Invalid enrollment code"

  • Enrollment codes expire after 24 hours
  • Each code can only be used once
  • Generate a new code in AgentWatch if yours has expired

"Authentication failed"

Your connector credentials may have been revoked. Re-enroll:

agentwatch-cli revoke --force
agentwatch-cli enroll --code NEW_CODE

"command not found: agentwatch-cli"

The CLI isn't in your PATH. Add the installation directory:

macOS (pip --user):

echo 'export PATH="'"$(python3 -c 'import sysconfig; print(sysconfig.get_path("scripts", "posix_user"))')"':$PATH"' >> ~/.zshrc
source ~/.zshrc

Linux (pip --user):

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc

pipx:

pipx ensurepath
# Then restart your terminal

Service won't start

Check the service logs:

Linux:

sudo journalctl -u agentwatch-cli -f

macOS:

cat ~/Library/Logs/agentwatch-cli.log

Upgrading

pipx:

pipx upgrade agentwatch-cli

pip:

pip install --user --upgrade agentwatch-cli

After upgrading, restart the service if installed:

# Linux
sudo systemctl restart agentwatch-cli

# macOS
launchctl unload ~/Library/LaunchAgents/io.agentwatch.cli.plist
launchctl load ~/Library/LaunchAgents/io.agentwatch.cli.plist

Uninstalling

# Remove service first (if installed)
agentwatch-cli uninstall-service  # or with sudo on Linux

# Remove package
pipx uninstall agentwatch-cli
# or
pip uninstall agentwatch-cli

# Optionally remove config
rm -rf ~/.agentwatch-cli

License

MIT License - see LICENSE file for details.

Support

Release files for agentwatch-cli 0.2.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agentwatch-cli 0.2.0
File Size Uploaded
agentwatch_cli-0.2.0.tar.gz 30.2 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for agentwatch-cli 0.2.0
File Interpreter ABI Platform
agentwatch_cli-0.2.0-py3-none-any.whl Python 3 none any Details

Total release size: 59.2 kB

Release files / agentwatch_cli-0.2.0.tar.gz

Download URL agentwatch_cli-0.2.0.tar.gz
Size 30.2 kB
Tags Source
SHA-256 checksum
How to use checksums
8a8a6e27a3fe6750db6091a4b5669cb08cd6e1bddd6ba9c50d0db3cceeabf16b
BLAKE2b-256 checksum
How to use checksums
135d2744283e5eadece787c20968b4be2ab4b4a36a0d787bb86890abedb6df78
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.10.12

Release files / agentwatch_cli-0.2.0-py3-none-any.whl

Download URL agentwatch_cli-0.2.0-py3-none-any.whl
Size 28.9 kB
Tags Python 3
SHA-256 checksum
How to use checksums
0207e287f153e7eefda00f28fd8ccd6e7681ff25620bff2fbc1d77f1daedb5e3
BLAKE2b-256 checksum
How to use checksums
ac8783416f064b40917cab82cf293d0b9a1d359ac81db2d46c2d4e3aea94aa54
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.10.12

Release history Release notifications | RSS feed

This release

0.2.0 This release

2 release files

0.1.9

2 release files

0.1.8

2 release files

0.1.7

2 release files

0.1.6

2 release files

0.1.5

2 release files

0.1.4

2 release files

0.1.3

2 release files

0.1.2

2 release files

0.1.1

2 release files

0.1.0

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page