OpenClaw

Prerequisites

• Node.js 20.19+ (or 22.12+, or 23+) inside the container or host where OpenClaw runs bash.
• Shell access in the workspace directory.
• For OAuth device login: you can authorize the sign-in link on your phone when the agent asks you to log in.

Installation

cd /data/workspace

npx skills add https://github.com/Ratimon/openquok-monorepo/tree/main/agent --skill openquok-core -y

clawhub install openquok-core

npm install -g @openquok/auto-cli@latest
openquok --version

export OPENQUOK_API_KEY=opo_your_programmatic_token
openquok auth:status

openquok integrations:list

Host OpenClaw on Railway

You can run the OpenClaw Gateway on Railway with a one-click template.

After deploy, use Railway’s shell (or the Control UI at /openclaw) to run the install steps below.

Typical Railway settings from the Railway install guide:

• Attach a volume at /data (persists config and workspace across redeploys).
• Set OPENCLAW_WORKSPACE_DIR=/data/workspace so the project workspace matches the path used in this guide.
• Enable HTTP Proxy on port 8080 and set OPENCLAW_GATEWAY_PORT=8080 (see OpenClaw docs for OPENCLAW_GATEWAY_TOKEN and other variables).

Then open a shell in the Railway service, cd /data/workspace, and continue with Installation.

OpenClaw + Telegram notes

• Start a new chat session after installing the skill or upgrading the CLI so the agent reloads instructions and checks openquok --version once at session start.
• If the agent reports an old CLI version after you upgraded, run command -v openquok and which -a openquok — another binary may be earlier on PATH.
• For posting with images in chat, the user must provide a file or a direct https:// image URL for upload-from-url; ask before calling posts:create with media.

Real World Use case: Chat model vs image generation

OpenClaw uses two separate systems:

Do not set an image-only model (for example google/gemini-3-pro-image-preview or openrouter/google/gemini-3-pro-image-preview) as your main Telegram or session model. Those models are built for image output, not general chat — the gateway may fail before it can run tools or answer questions.

Set the main chat model

Fastest path on a fresh host:

openclaw onboard

Or reconfigure models only:

openclaw configure --section model

Pick a chat model such as anthropic/claude-sonnet-4-6. You can also set the default from the shell:

openclaw models set anthropic/claude-sonnet-4-6
openclaw models status

OpenClaw writes the result to gateway config (typically openclaw.json under your OpenClaw data directory):

{
  agents: {
    defaults: {
      model: { primary: "anthropic/claude-sonnet-4-6" },
    },
  },
}

In an existing Telegram or Control UI session, switch with /model anthropic/claude-sonnet-4-6 or /model for the numbered picker. Use /model default to return to the configured primary.

See Models for fallbacks, allowlists, and provider auth.

Configure image generation

Image generation is not part of the OpenQuok skill. Set it up once on the OpenClaw gateway host:

1. Add an API key for at least one image provider — for example OPENAI_API_KEY, GEMINI_API_KEY, FAL_KEY, or OPENROUTER_API_KEY (OpenAI Codex OAuth also works for openai/gpt-image-2).
2. Optionally set a default image model so image_generate does not rely on auto-detection alone.

openclaw config set agents.defaults.imageGenerationModel '{"primary":"openai/gpt-image-2","timeoutMs":180000}' --strict-json --merge

Example gateway config with fallbacks:

{
  agents: {
    defaults: {
      imageGenerationModel: {
        primary: "openai/gpt-image-2",
        timeoutMs: 180_000,
        fallbacks: [
          "openrouter/google/gemini-3.1-flash-image-preview",
          "google/gemini-3.1-flash-image-preview",
          "fal/fal-ai/flux/dev",
        ],
      },
    },
  },
}

Common routes (see the OpenClaw image generation guide for the full provider table):

The image_generate tool appears only when at least one image provider is available. Inspect registered providers at runtime with /tool image_generate action=list.

Then ask in Telegram or the Control UI:

Generate a square image of a product hero shot on a clean white background

OpenClaw routes that to image_generate using imageGenerationModel.primary (or the next fallback) — your main chat model stays on Sonnet.

Full model list, editing, aspect ratios, and provider deep dives: OpenClaw image generation.

Troubleshooting

Skill source on GitHub

Related Section(s)