For hosting providers

Sell AI assistants without inventing a multi-tenant gateway.

You already know how to run servers. OpenClawMU is the multi-tenant assistant gateway you would have built — Apache-2.0, no per-customer licensing, your margin is your own.

The opportunity, plainly

Hosted AI assistants are a product category your customers want and don't want to build. They want a WhatsApp-fronted, Slack-fronted, or Telegram-fronted bot. They don't want to set up Anthropic accounts, manage tokens, configure sandboxes, or deal with the auth dance for each channel.

The cloud bot platforms charge $5–$50 / active user / month plus model fees. That's a margin you could keep if you ran the stack yourself — and that's exactly the gap OpenClawMU fills.

What you provide

  • Hardware — a server (or three) with predictable bandwidth.
  • Operational reliability — uptime SLA, backups, on-call rotation.
  • Customer support — first-line response, troubleshooting, training.
  • Compliance assurances — data residency, SOC 2, ISO 27001 if your customers need them.
  • Branding — your skin on the assistant UI, your domain on the gateway.

What OpenClawMU provides

  • One Node process, many isolated tenants.
  • Token-based per-tenant auth, hashed at rest.
  • Sandbox isolation (bubblewrap or Docker) per agent run.
  • Per-tenant cost accounting → bill your customers cleanly.
  • S3 backup so you can run nightly snapshots, restore on incident.
  • Channel pairing — WhatsApp / Telegram / Slack / Discord / 8 more.
  • Web terminal for support to actually help customers when something needs hands-on attention.

Deployment topology

A small hosting provider can run OpenClawMU on a single VM with nightly S3 backups and Tailscale or Cloudflare Tunnel for ingress. A medium provider runs N gateways behind a hash-based tenant router (your own nginx Lua or HAProxy stick-table) for horizontal scale. A larger provider can co-locate gateways near LLM providers' POPs to minimize latency.

There's no central database, no shared message bus, no orchestration requirement. Each gateway is a self-contained unit; backups make them portable; quotas make them billable.

Margin math

Anthropic's claude-sonnet-4-6 is roughly $3 / $15 per million input/output tokens. A typical assistant interaction is ~3 K in, ~500 out — call it $0.018 per turn at list price. Mark up 30% and you're charging $0.025 / turn. A heavy SMB user sending 100 turns/day costs you $1.80 in LLM fees; you charge $2.50; you make $0.70 / day / user. Multiply by tenants. You still need to cover infrastructure (mid-2-figure $ / mo per gateway) and your time.

The point is: every dollar of margin survives because there's no platform fee between you and your customer.

EXFOLIATE!

Run your own gateway today.

Apache-2.0, self-hosted, no SaaS layer between you and your users. Install the CLI, create your first tenant, mint a token — you're routing traffic in 60 seconds.