CLI Reference

IronClaw ships as a single binary named ironclaw. Every command is a subcommand — running ironclaw on its own prints usage and exits without starting anything.

ironclaw --help
ironclaw --version

Running the Agent

Command	What it does
`ironclaw serve`	Start the web interface on `127.0.0.1:3000`
`ironclaw repl`	Start an interactive terminal session with the agent
`ironclaw run -m "<message>"`	Send a single message and exit
`ironclaw status`	Print a runtime status snapshot
`ironclaw doctor`	Check configuration and driver health without creating state

ironclaw run -m "What changed in my repo today?"

doctor is the first thing to run when something isn’t working. It reports your home directory, active profile, whether configuration files were found, and whether each driver initialized.

Setup and Configuration

Command	What it does
`ironclaw onboard`	One-time setup — see Onboard
`ironclaw config path`	Show resolved configuration paths without creating state
`ironclaw config init`	Write a commented `config.toml` and `providers.json` stub
`ironclaw config list`	List every configuration key and its current value
`ironclaw config get <key>`	Read one value by dot-separated key
`ironclaw config set <key> [value]`	Write one value; secret keys omit it and prompt
`ironclaw profile list`	List the available boot profiles

ironclaw config get llm.default.provider_id
ironclaw config set google.client_id <your-client-id>

config init refuses to overwrite existing files unless you pass --force. config list and config get read every key. config set is narrower: it accepts only <provider>.api_key, google.client_id, google.client_secret, google.redirect_uri, slack.enabled, and webui.token --rotate, routing each to the configuration file, the encrypted secret store, or the web token file. Everything else is edited in config.toml directly. config set does not restart the running instance — it prints to apply: ironclaw service restart. See Configuration.

Models and Providers

Command	What it does
`ironclaw models list`	List available inference providers, or show one
`ironclaw models status`	Show which model route is active
`ironclaw models set <model>`	Set the default model for the active provider
`ironclaw models set-provider <provider>`	Set the default provider

ironclaw models list
ironclaw models set-provider anthropic --model claude-sonnet-4-20250514
ironclaw models status

See Inference Providers for the full catalog.

Extensions and Skills

Command	What it does
`ironclaw extension search <query>`	Search available extension packages
`ironclaw extension install <name>`	Install an extension package
`ironclaw extension activate <name>`	Activate an installed extension
`ironclaw extension remove <name>`	Remove an installed extension
`ironclaw skills list`	List configured skills
`ironclaw hooks list`	Not implemented yet — reports that and exits
`ironclaw channels list`	Not implemented yet — reports that and exits
`ironclaw logs`	Not implemented yet — reports that and exits

Add --verbose to skills list for extra detail. Activating an extension that needs credentials starts its setup flow. Complete that from Extensions in the web interface, which is also where you connect channels while channels list remains unimplemented. For service logs, read the files directly — see Run as a Service.

Running as a Service

Command	What it does
`ironclaw service install`	Install the OS service (launchd or systemd)
`ironclaw service start`	Start the installed service
`ironclaw service stop`	Stop the running service
`ironclaw service restart`	Restart, or start if stopped
`ironclaw service status`	Show service status
`ironclaw service uninstall`	Remove the service and its unit file

See Run as a Service.

service subcommands do not work on a NEAR AI hosted instance — there is no user service manager for them to drive. Restart a hosted agent from the Agent Dashboard instead.

Trace Commons

ironclaw traces manages optional, redacted contribution of your agent traces. It is off until you explicitly opt in.

Command	What it does
`traces status`	Show your current contribution policy
`traces opt-in` / `traces opt-out`	Turn contribution on or off
`traces preview <file>`	Preview a redacted contribution envelope
`traces submit` / `traces enqueue` / `traces flush-queue`	Submit contributions
`traces queue-status`	Show local queue diagnostics
`traces credit`	Show credit totals and explanations
`traces list-submissions` / `traces revoke`	Review or withdraw past contributions
`traces enroll-instance`	Enroll an entire instance with an operator invite link
`traces profile set` / `withdraw` / `token`	Manage the optional public community profile
`traces ingest-health`	Check an ingestion service health endpoint

See Trace Commons.

Shell Completion

ironclaw completion zsh > "${fpath[1]}/_ironclaw"

Bash, fish, and PowerShell are also supported.

Using IronClaw

Capabilities

Channels

API

Extensions (Tools)

Tutorials

Running the Agent

Setup and Configuration

Models and Providers

Extensions and Skills

Running as a Service

Trace Commons

Shell Completion

​Running the Agent

​Setup and Configuration

​Models and Providers

​Extensions and Skills

​Running as a Service

​Trace Commons

​Shell Completion

Running the Agent

Setup and Configuration

Models and Providers

Extensions and Skills

Running as a Service

Trace Commons

Shell Completion