CLI automation

Use openclaw onboard --non-interactive to script setup. It requires --accept-risk: non-interactive setup can write credentials and daemon config without a confirmation prompt, so the flag is the explicit risk acknowledgement. Each command can install a managed Gateway with --install-daemon, require an already-running compatible Gateway by omitting daemon flags, explicitly leave the Gateway stopped with --skip-daemon, or use --skip-health for config-only setup. The explicit skip still probes for an existing Gateway and reports whether one is reachable, but an absent listener is informational rather than a setup failure.

--json does not imply non-interactive mode. Pass --non-interactive --accept-risk explicitly for scripts.

Baseline non-interactive example

openclaw onboard --non-interactive --accept-risk \
  --mode local \
  --auth-choice apiKey \
  --anthropic-api-key "$ANTHROPIC_API_KEY" \
  --secret-input-mode plaintext \
  --gateway-bind loopback \
  --install-daemon \
  --daemon-runtime node \
  --skip-bootstrap \
  --skip-skills

Add --json for a machine-readable summary.

--gateway-port defaults to 18789; only pass it to override.
--skip-bootstrap skips creating default workspace files, for automation that pre-seeds its own workspace.
--secret-input-mode ref stores new credentials as env-backed references ({ source: "env", provider: "default", id: "<ENV_VAR>" }); set the provider env var when adding a credential or passing an inline key flag. Existing resolvable named profiles and their env, file, exec, or store references are reused unchanged, without a new credential write or additional provider env var. Existing plaintext is not migrated; run openclaw secrets configure --apply, then openclaw secrets audit --check. See Secrets management.
The gateway token follows the same mode. Setup generates that value itself, so reference mode has no env var to point at unless you supply one: with OPENCLAW_GATEWAY_TOKEN exported, gateway.auth.token becomes an env ref to it; otherwise the token goes into the SQLite secret store as OPENCLAW_GATEWAY_TOKEN and config keeps a store ref. Either way openclaw.json holds no plaintext gateway token. Inspect the entry with openclaw secrets store list.
In reference mode, explicit --gateway-password and --remote-password must match OPENCLAW_GATEWAY_PASSWORD, and --remote-token must match OPENCLAW_GATEWAY_TOKEN. Missing or mismatched environment values fail before setup changes state; matching credentials are stored as env SecretRefs.

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice openai-api-key \
  --secret-input-mode ref

Provider-specific examples

Anthropic API key example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice apiKey \
  --anthropic-api-key "$ANTHROPIC_API_KEY" \
  --gateway-bind loopback

Cloudflare AI Gateway example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice cloudflare-ai-gateway-api-key \
  --cloudflare-ai-gateway-account-id "your-account-id" \
  --cloudflare-ai-gateway-gateway-id "your-gateway-id" \
  --cloudflare-ai-gateway-api-key "$CLOUDFLARE_AI_GATEWAY_API_KEY" \
  --gateway-bind loopback

Gemini example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice gemini-api-key \
  --gemini-api-key "$GEMINI_API_KEY" \
  --gateway-bind loopback

Mistral example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice mistral-api-key \
  --mistral-api-key "$MISTRAL_API_KEY" \
  --gateway-bind loopback

Moonshot example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice moonshot-api-key \
  --moonshot-api-key "$MOONSHOT_API_KEY" \
  --gateway-bind loopback

Ollama example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice ollama \
  --custom-model-id "qwen3.5:27b" \
  --gateway-bind loopback

OpenCode example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice opencode-zen \
  --opencode-zen-api-key "$OPENCODE_API_KEY" \
  --gateway-bind loopback

Swap to --auth-choice opencode-go --opencode-go-api-key "$OPENCODE_API_KEY" for the Go catalog.

Synthetic example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice synthetic-api-key \
  --synthetic-api-key "$SYNTHETIC_API_KEY" \
  --gateway-bind loopback

Vercel AI Gateway example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice ai-gateway-api-key \
  --ai-gateway-api-key "$AI_GATEWAY_API_KEY" \
  --gateway-bind loopback

Z.AI example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice zai-api-key \
  --zai-api-key "$ZAI_API_KEY" \
  --gateway-bind loopback

Custom provider example

openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice custom-api-key \
  --custom-base-url "https://llm.example.com/v1" \
  --custom-model-id "foo-large" \
  --custom-api-key "$CUSTOM_API_KEY" \
  --custom-provider-id "my-custom" \
  --custom-compatibility anthropic \
  --custom-image-input \
  --gateway-bind loopback

--custom-api-key is optional; some endpoints do not require auth. If omitted, onboarding checks CUSTOM_API_KEY in env. --custom-provider-id is optional and auto-derived from the base URL when omitted. --custom-compatibility defaults to openai (other values: openai-responses, anthropic).OpenClaw infers image-input support from known vision model-id patterns (gpt-4o, claude-3/4, gemini, -vl/vision suffixes, and similar). Add --custom-image-input to force it on for an unrecognized vision model, or --custom-text-input to force text-only.Ref-mode variant, storing apiKey as { source: "env", provider: "default", id: "CUSTOM_API_KEY" }:

export CUSTOM_API_KEY="your-key"
openclaw onboard --non-interactive --accept-risk --skip-health \
  --mode local \
  --auth-choice custom-api-key \
  --custom-base-url "https://llm.example.com/v1" \
  --custom-model-id "foo-large" \
  --secret-input-mode ref \
  --custom-provider-id "my-custom" \
  --custom-compatibility anthropic \
  --custom-image-input \
  --gateway-bind loopback

Anthropic setup-token auth remains supported, but OpenClaw prefers Claude CLI reuse when a local Claude CLI login is available. For production, prefer an Anthropic API key.

Add another agent

openclaw agents add <name> creates a separate agent with its own workspace, sessions, and auth profiles. Running it without --workspace (and no other flags) launches the interactive wizard; passing any of --workspace, --model, --agent-dir, --bind, or --non-interactive runs it non-interactively and then requires --workspace.

openclaw agents add work \
  --workspace ~/.openclaw/workspace-work \
  --model openai/gpt-5.6-sol \
  --bind whatsapp:biz \
  --non-interactive \
  --json

Config keys it writes (agents.entries.* entry for the new agent id):

name
workspace
agentDir
model (only when --model is passed)

Notes:

Default workspace (when --workspace is omitted in the interactive wizard): ~/.openclaw/workspace-<agentId>.
--bind <channel[:accountId]> is repeatable; add bindings to route inbound messages to the new agent (the wizard can also do this interactively).
The agent name is normalized to a valid agent id. main is allowed, but an existing named installation may require openclaw doctor --fix to finish legacy-session and shared-auth ownership migrations before creating it.

Onboarding hub: Onboarding (CLI)
Full reference: CLI Setup Reference
Command reference: openclaw onboard

Overview

First steps

Guides

Baseline non-interactive example

Provider-specific examples

Add another agent

​Baseline non-interactive example

​Provider-specific examples

​Add another agent

​Related docs

Baseline non-interactive example

Provider-specific examples

Add another agent

Related docs