OpenClaw Cloud

Managed personal AI assistants with browser chat, Telegram, or Discord. Powered by OpenClaw (open source).

Overview

OpenClaw Cloud deploys a fully configured OpenClaw instance as a RailPush service. Users choose browser-only chat, Telegram, or Discord, then choose how OpenClaw will authenticate to a model provider. ChatGPT/Codex OAuth and OpenAI Platform API keys are separate paths. RailPush handles deployment, web chat, isolation, persistent storage, and a version-pinned managed release track.

Deep dive: Launch OpenClaw Without Telegram or Discord.

RailPush Integration

Each OpenClaw instance can connect directly to your RailPush workspace. When enabled, your assistant gets a scoped API key with access to the full RailPush API -- the same 235 MCP tools available to any agent. From a chat message, your assistant can:

• Deploy services -- create from Git repos, Docker images, or templates; trigger deploys; roll back
• Manage databases -- provision PostgreSQL, MySQL, MongoDB, Redis, or Meilisearch; run queries; create backups
• Configure infrastructure -- set environment variables, add custom domains, configure autoscaling
• Monitor and debug -- check service health, view logs, inspect metrics, diagnose failing deploys
• Manage storage -- create S3-compatible buckets, manage objects and credentials

Authorization options: During setup, choose to grant access immediately (API key) or later via the OAuth device flow -- your assistant sends you a link, you approve in your browser. API keys are scoped to read, write, and deploy permissions. You can revoke access anytime from Settings -> API Keys.

Billing safeguards: Your assistant is instructed to always ask for confirmation before creating paid services. OpenClaw Cloud itself is a paid agent instance; the full resource pricing (Nano $4/mo, Small $7/mo, Medium $14/mo, Large $28/mo, GPU $49/mo) is shown to your assistant so it can communicate costs clearly.

USDC payments: RailPush supports USDC workspace credit top-ups through Stripe Checkout. The checkout settles in USD on Stripe and the resulting credits automatically apply to future invoices. Today this is the stablecoin payment path RailPush supports publicly.

Sizes

Size	Price	RAM	Always On	Billing
Everyday	$5/mo	512 MB	Yes	Fixed monthly
Pro	$12/mo	2 GB	Yes	Fixed monthly

Model Auth Options

• ChatGPT/Codex OAuth -- openai-codex/gpt-5.5, connected after launch from the RailPush OpenClaw page. RailPush uses device-code sign-in when OpenAI allows it and falls back to browser sign-in with a localhost callback paste when OpenAI blocks the device-code request. No OPENAI_API_KEY is stored for this option.
• OpenAI Platform API -- openai/gpt-5.4, openai/gpt-4o, openai/gpt-4o-mini with OPENAI_API_KEY.
• Anthropic Claude API -- anthropic/claude-sonnet-4-20250514, anthropic/claude-opus, anthropic/claude-haiku with ANTHROPIC_API_KEY.
• Google Gemini API -- google/gemini-2.5-pro, Gemini Flash models with GOOGLE_API_KEY.
• xAI Grok API -- xai/grok-4, xai/grok-4-fast with XAI_API_KEY.
• OpenRouter API -- openrouter/auto or any OpenRouter model route with OPENROUTER_API_KEY.

If a user chooses an API-key provider without entering a key during setup, the instance still deploys with provider auth pending. Add the matching environment variable in Service Details, then redeploy or restart the service. If a user chooses ChatGPT/Codex OAuth, launch first and use the Connect ChatGPT panel after the service is running. The panel will show either a device code or a browser sign-in callback field, depending on what OpenAI accepts for that runtime.

Supported Channels

• Browser only -- Skip Telegram and Discord, then use the RailPush share link for web chat
• Telegram -- Create a bot via @BotFather, paste the token
• Discord -- Create an app at discord.com/developers, paste the bot token and application ID

Deploy via Dashboard

Go to railpush.com/openclaw and follow the guided setup wizard. It walks you through selecting ChatGPT/Codex OAuth or an API-key provider, optionally creating a bot, and launching -- all in about 2 minutes.

Deploy via MCP

MCP

// Create an OpenClaw instance
create_openclaw_instance({
  name: "my-assistant",
  plan: "openclaw",          // openclaw | openclaw-pro
  provider: "openai-codex",  // openai-codex | anthropic | openai | google | xai | openrouter
  api_key: "",               // API-key providers only; omit for openai-codex OAuth
  channel: "none",           // none | telegram | discord
  bot_token: "123456:ABC...", // only for telegram | discord
  enable_workspace_access: true
})

Environment Variables

Variable	Description
ANTHROPIC_API_KEY	Anthropic API key (for Claude)
OPENAI_API_KEY	OpenAI Platform API key (for openai/* models)
GOOGLE_API_KEY	Google API key (for Gemini)
XAI_API_KEY	xAI API key (for Grok)
OPENROUTER_API_KEY	OpenRouter API key (for openrouter/* models)
OPENCLAW_MODEL	Model identifier (e.g. anthropic/claude-sonnet-4-20250514)
OPENCLAW_GATEWAY_TOKEN	Password for the browser-based web chat gateway
TELEGRAM_BOT_TOKEN	Telegram bot token from @BotFather
DISCORD_BOT_TOKEN	Discord bot token
DISCORD_APP_ID	Discord application ID for the bot
RAILPUSH_API_KEY	Optional scoped RailPush API key for managing the workspace from chat
RAILPUSH_API_URL	RailPush API base URL used for workspace actions

Updates

Managed OpenClaw no longer tracks latest. RailPush pins the hosted image to a vetted OpenClaw release and only bumps it after smoke tests pass. New instances get the current managed release, and existing instances move forward only during an intentional OpenClaw rollout.

Customization

Each instance has a persistent workspace at /workspace. Users can customize their assistant by editing files via the RailPush file browser or shell:

• SOUL.md -- Personality, tone, and behavioral rules
• AGENTS.md -- Workspace conventions and memory strategy
• TOOLS.md -- Environment-specific notes
• memory/ -- Daily notes and long-term memory files