Server Management

Commands for starting, stopping, and updating the GuardClaw server.

start

Start the GuardClaw server and open the dashboard in your browser.

guardclaw start [options]

Options

Flag	Description	Default
-f, --foreground	Run in foreground (Ctrl-C to stop)	Daemon mode
--port <port>	Server port	3002
--no-open	Don't open browser on start	Opens browser
--no-onboarding	Skip first-run setup wizard (useful for CI/scripts)	Runs wizard
--openclaw-url <url>	OpenClaw gateway URL	From .env
--openclaw-token <token>	OpenClaw gateway token	From .env
--anthropic-key <key>	Anthropic API key	From .env

Legacy aliases --clawdbot-url and --clawdbot-token also work for --openclaw-url and --openclaw-token.

Daemon mode (default)

By default, guardclaw start runs the server as a background daemon. The process detaches from the terminal and continues running after you close the shell.

guardclaw start
# ✅ GuardClaw running in background (PID 12345)
# 📜 Logs: .guardclaw/server.log
# 🛑 Stop with: guardclaw stop

• Server logs are written to .guardclaw/server.log in the current directory.
• The PID file is not written; the stop command finds the process via process table and port inspection.

Foreground mode

Use -f or --foreground to run in the current terminal. The server's stdout/stderr is printed directly. Press Ctrl-C to stop.

guardclaw start -f

This is useful for development, debugging, or when running inside a container.

First-run onboarding

On first launch (when no .env file exists), the CLI runs a 4-step interactive setup wizard:

1. Evaluation mode — local-only, mixed, or cloud-only
2. LLM backend — LM Studio, Ollama, Built-in, or Fallback (+ cloud provider for mixed/cloud)
3. Response mode — auto or monitor-only
4. Agent connections — auto-detects and connects installed agents

Skip this with --no-onboarding or by running guardclaw setup manually later.

Working directory

The server runs in your current working directory, not the npm install directory. This means:

• .env is read from your current directory
• .guardclaw/ data directory (events.db, memory.db) is created in your current directory
• Each project can have its own GuardClaw configuration and event history

Examples

# Start with all defaults
guardclaw start

# Start on custom port, no browser
guardclaw start --port 4000 --no-open

# Start in foreground for debugging
guardclaw start -f --port 3002

# Start with OpenClaw token inline
guardclaw start --openclaw-token eyJ...

# Start in CI without interactive wizard
guardclaw start --no-onboarding --no-open

stop

Stop the running GuardClaw server.

guardclaw stop

The command finds GuardClaw processes in two ways:

1. Process table scan — searches ps ax for processes running server/index.js with "guardclaw" in the command, or whose working directory is the GuardClaw install directory.
2. Port listener scan — finds any process listening on the GuardClaw port (default 3002) that is running server/index.js.

All matched processes are terminated with SIGKILL.

guardclaw stop
# 🛑 Stopping GuardClaw...
# ✅ Stopped PID 12345
# ✅ Stopped 1 process(es)

If no GuardClaw process is found:

ℹ️  GuardClaw is not running.

restart

Stop the server and start it again. Accepts all the same options as start.

guardclaw restart [options]

Aliases: rs, r

# Quick restart
guardclaw rs

# Restart on a different port
guardclaw restart --port 4000

The command waits 300ms between stop and start to allow the OS to release sockets and PIDs.

status

Show a comprehensive overview of the running server state.

guardclaw status

Output includes:

• Running — PID of the server process
• Blocking — whether pre-execution blocking is enabled
• Fail-closed — whether the server blocks tool calls when the LLM is unavailable
• LLM — active backend and loaded model
• Agent Connections — each connected agent with type (hook/ws) and event count
• Events — total, safe, warning, and blocked counts
• Cache — hits, misses, and AI calls

Example output:

⛨  GuardClaw Status

  Running:     ✅ Yes  (PID 12345)
  Blocking:    🟢 OFF (monitor only)
  Fail-closed: No
  LLM:         lmstudio (qwen/qwen3-4b-2507)

  Agent Connections
  ✅ Claude Code       [hook]  42 events
  ⚪ Codex             [hook]
  ✅ OpenClaw          [ws]    18 events

  Events: 60 total  🟢 45 safe  🟡 12 warn  🔴 3 blocked
  Cache:  38 hits / 22 misses / 22 AI calls

TIP

guardclaw status requires the server to be running. If the server is offline, it will print an error message and exit.

update

Update GuardClaw to the latest version from npm.

guardclaw update

Alias: upgrade

The command:

1. Checks the current version against the latest on npm
2. If already up to date, prints a message and exits
3. Stops the running server (if any)
4. Runs npm install -g guardclaw@latest
5. Restarts the server if it was running before the update

guardclaw update
# 🔄 Updating GuardClaw...  0.1.5 → 0.1.6
# 🛑 Stopping server before update...
# ✅ Updated successfully
# 🔄 Restarting server...

version

Print the installed version.

guardclaw version
guardclaw --version
guardclaw -v