OpenClawMU vs Lobu — which multi-tenant OpenClaw fork?

OpenClaw is single-user by design, so the moment you want to host it for more than one person a fork has to add a tenancy layer. Lobu and OpenClawMU are two open-source, self-hosted forks that do exactly that. They start from the same place — the upstream OpenClaw agent runtime and its channel adapters — and diverge on how much of the “run it as a service” problem they solve for you.

What both share

• The upstream OpenClaw agent runtime (the Pi agent) and its skill ecosystem.
• The same channel adapters — WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and more.
• Open source and self-hosted — your data and credentials stay on hardware you control, and you pay your LLM provider directly with no platform spread.
• A multi-tenant surface that upstream OpenClaw does not ship.

Where Lobu focuses

Lobu is the closest direct rival: an open-source multi-tenant OpenClaw gateway centered on per-user sandboxing and isolated filesystems. If your requirement is “give each user their own isolated agent and workspace,” that’s the problem Lobu is built around.

Where OpenClawMU focuses — the operator stack

OpenClawMU treats isolation as the starting point, then adds the pieces an operator needs to charge for the service:

• Tenant token auth in front of every JSON-RPC and HTTP API.
• Isolated per-tenant directories for sessions, memory, plugins, sandbox roots, cron, channels, and config.
• Per-tenant quotas — tokens/day, USD/day, requests/minute.
• Per-tenant cost accounting with a billing report generator.
• S3-compatible backup and restore of full tenant state, with path-traversal hardening.
• Browser web terminal — xterm.js + node-pty, sandboxed per tenant.
• Control-plane HTTP API for programmatic tenant lifecycle management.
• Admin / tenant key separation — a tenant config overlay can’t override admin keys.

How to choose

Pick the fork by how far past raw isolation you need to go. If you need per-user sandboxing and filesystem isolation and will handle metering, billing, and backups yourself, Lobu is a lean fit. If you’re building a SaaS or agency business on OpenClaw and want per-tenant billing, quotas, portable backups, and web terminals bundled and maintained together, OpenClawMU is built for that operator use case.

OpenClawMU is an independent community fork maintained by Neul Labs. It is not affiliated with or endorsed by the upstream OpenClaw project or by Lobu.

When to pick OpenClawMU

• You're running OpenClaw for paying customers and need per-tenant cost accounting, quotas (tokens/day, USD/day, requests/minute), and billing reports out of the box.
• You need portable tenant snapshots — S3-compatible backup and restore of full tenant state.
• You want a browser-based, per-tenant sandboxed web terminal (xterm.js + node-pty) to give tenants a shell without SSH.
• You want a control-plane HTTP API to create, rotate, back up, and delete tenants programmatically.
• You need admin/tenant key separation so a tenant config overlay can never override admin keys.

When to pick Lobu instead

• Your requirement is squarely per-user sandboxing and filesystem isolation, and you don't need bundled billing or backup.
• You prefer Lobu's isolation model or are already running it and it fits.
• You want the smallest possible multi-tenant surface over upstream OpenClaw.

Frequently asked

What's the core difference between OpenClawMU and Lobu?

Both are open-source, self-hosted forks that make the single-user OpenClaw gateway multi-tenant. Lobu concentrates on per-user sandboxes and isolated filesystems. OpenClawMU adds the operator stack on top of isolation: per-tenant billing and quotas, S3 backup/restore, a per-tenant web terminal, and a control-plane API — the pieces you need to actually run OpenClaw as a paid service.

Which multi-tenant OpenClaw fork should I pick to build a SaaS?

If your product needs to meter, bill, and back up tenants, OpenClawMU bundles that (per-tenant cost accounting, tokens/USD/request quotas, S3-portable snapshots) so you don't build it yourself. If you only need per-user isolation and will handle metering and backups separately, a leaner fork like Lobu can be enough.

Do both forks track upstream OpenClaw?

Both are forks of the same upstream OpenClaw agent runtime, so channels (WhatsApp, Telegram, Slack, Discord, Signal, and more), the Pi agent, and the skill ecosystem come from upstream on both sides. OpenClawMU keeps its additions behind fork-marker comments to keep the rebase surface small; it rebases on upstream every 2–4 weeks.

Are OpenClawMU and Lobu affiliated with upstream OpenClaw?

No. Both are independent community forks of openclaw/openclaw. OpenClawMU is maintained by Neul Labs and is not endorsed by or affiliated with the upstream OpenClaw project.