Skip to content

OpenClaw Hosted

OpenClaw Hosted is a managed control plane for running OpenClaw agent gateways in the cloud. You get isolated deployments, billing and usage metering, a web dashboard, and a programmatic API — without operating Kubernetes or Cloud Run yourself.

This documentation describes customer-visible behavior at a tagged release. It is published at docs.clusterclaw.ai.

What you get

Component What it does
Control plane API Connect RPC at https://api.clusterclaw.ai — create deployments, manage billing, fetch status
Web dashboard www.clusterclaw.ai — sign in with Google, launch deployments, connect channels
CLI (claw) Terminal workflow for signup, deploy, status, and usage
Per-deployment gateway Your OpenClaw runtime — chat channels, tools, skills, and agent loops
Inference proxy Optional platform-managed model access (platform-managed inference mode) via https://proxy.clusterclaw.ai

How it works

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
You (dashboard, CLI, or API)
        │
        ▼  Bearer cc_* API key
https://api.clusterclaw.ai          ← control plane (Connect RPC)
        │
        ├── Account & billing (Stripe)
        ├── Deployment lifecycle
        └── Provisioning orchestrator
                │
                ▼
Your deployment (default: Cloud Run + persistent storage)
        │
        ├── OpenClaw gateway runtime
        ├── Chat channels (Slack, Discord, Telegram, …)
        └── Inference (platform-managed or BYOK)

Each deployment is an isolated OpenClaw gateway instance with its own storage, configuration, and optional public hostname (for example my-bot.clusterclaw.ai). The control plane tracks status from queued through healthy, surfaces logs, and meters compute, token, and storage usage for billing.

Key concepts

Customer account

One customer record owns all deployments, API keys, Stripe billing, and usage limits. Sign up via Google OAuth on the dashboard or programmatically with AccountService.Signup. See Account.

There is no multi-user organization model today — one account maps to one billing identity. Team SSO and seat management are planned, not shipped.

Deployment

A named gateway instance in a GCP region (default us-central1). You choose:

  • Target class — cloud_run (default, cost-efficient) or autopilot (GKE Autopilot, enterprise tier only)
  • Inference mode — platform-managed keys, BYOK, self-hosted, or external-hosted endpoints
  • Channels — Discord, Slack, Telegram, and others via the launch catalog
  • Public ingress — optional HTTPS hostname for the gateway Control UI

See Deployments.

Gateway vs control plane

The gateway is your running OpenClaw agent runtime. The control plane is ClusterClaw's API and dashboard. API keys authenticate to the control plane only. Each deployment also receives separate runtime tokens (ccgw_* for gateway SSO, ccproxy_* for inference) that are never stored in logs or echoed by routine API calls.

Inference modes

Mode Description
platform-managed ClusterClaw provides model access through the inference proxy; token usage is metered
byok You supply provider API keys; stored securely and injected at provision time; no token charges from ClusterClaw
self-hosted Your own Ollama, vLLM, or similar endpoint
external-hosted Third-party hosted models (Together, Replicate, etc.)

Tiers and entitlements

Self-serve tiers (starter, hobbyist, developer, team, business) set base subscription price, deployment limits, and usage rate cards. The free tier (Google sign-in default) can browse the dashboard but cannot create deployments until you upgrade.

Enterprise customers with the enterprise tier may use GKE Autopilot (target_class: autopilot). Contact sales for custom arrangements.

Documentation map

Guide Topics
Getting started First account, API key, first deployment, health checks
Authentication API keys, dashboard OAuth, programmatic access
API reference Connect RPC services, errors, examples
Account Signup, tiers, entitlements
Deployments Lifecycle, channels, hostnames, runtime
Session management Gateway sessions, list/abort, platform differences
Billing Stripe checkout, usage, limits
AIFA add-on Enterprise orchestration (not GA)

Support surfaces

API contract

Canonical RPC definitions live in proto/openclaw/v1/ in the OpenClaw Hosted repository. Generated OpenAPI is bundled for interactive reference in the Backstage customer portal (yarn start:customer in apps/backstage).