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.