OpenClawMU is the multi-tenant fork of OpenClaw — the self-hosted AI gateway across WhatsApp, Telegram, Slack, Discord, Signal, iMessage and 6 more. Add tenant isolation, sandboxed agents, per-tenant tokens & quotas, web terminals, S3 backup. Apache-2.0, self-hosted, no SaaS lock-in.
What the MU adds
Each tenant gets its own sessions, memory, plugins, sandboxes, cron jobs, device pairings, and config — keyed by a SHA-256-hashed token that never leaves the box.
xterm.js in the browser, attached to a per-tenant sandbox via terminal.spawn / write / resize. Real shells, real ptys, real isolation.
Lightweight bubblewrap (no root) or full Docker with seccomp + cgroups. Pick your isolation level per workload, not per cluster.
Token usage, cost snapshots, quota enforcement, billing history — accounted at the tenant boundary. Hand your finance team a CSV.
Channels bridged
OpenClaw upstream bridges 12 channels via vetted SDKs. OpenClawMU keeps every one of them — and routes each inbound message to the right tenant.
Isolation model
Every state-bearing surface — sessions, vector memory, plugins, cron jobs, sandboxes, device pairings, channel credentials — lives in a per-tenant directory. The token never leaves the box (SHA-256 hashed at rest, constant-time verified), and the dispatcher will not route a message to the wrong tenant even on a malformed input.
.. segments and absolute paths.Comparison
| Capability | OpenClawMU | OpenClaw (upstream) | Botpress | Chatwoot |
|---|---|---|---|---|
| Multi-tenant isolation | Built-in (token + isolated dirs) | Single-user only | Closed cloud SaaS | Multi-account, not multi-tenant |
| Self-hosted | Yes (Node 22+ or Docker) | Yes | Cloud-first | Yes |
| License | Apache-2.0 | Apache-2.0 | Proprietary | MIT (Community) · Commercial features |
| Channels supported | 12+ (WA, Telegram, Slack, etc.) | 12+ | WhatsApp, Telegram, Web | Email, Web, WhatsApp, FB, Twitter |
| Sandboxed code execution | bubblewrap or Docker | bubblewrap or Docker | Cloud sandbox | Not a feature |
| Browser web terminal | xterm.js per tenant | Local only | Studio in browser | Not a feature |
| Per-tenant cost tracking | Built-in usage + billing | Single bucket | Plan-tiered | Per-account |
| Bring your own LLM | Anthropic, OpenAI, any OpenAI-compatible | Same | OpenAI + curated | Plug your own LLM |
| S3-compatible backup/restore | Built-in | Not bundled | Cloud-managed | Manual |
From the blog
A multi-tenant AI gateway is the layer between your messaging channels and your LLM that isolates per-customer state. Here's how the pattern works, why it matters now, and what a defensible implementation looks like.
Read articleRun a multi-tenant WhatsApp bot platform on your own infra without the WhatsApp Business API contract. Open-source, BYO-LLM, with per-customer isolation and cost tracking. Here's the architecture.
Read articleOpenClaw is a brilliant single-user AI gateway. OpenClawMU is the multi-tenant fork. Here's exactly what differs, when the fork is worth it, and how to migrate cleanly without losing state.
Read articleFrequently asked
Every question here is also emitted as FAQPage JSON-LD — so Google AI Overviews, Perplexity, and ChatGPT browsing extract the answers cleanly. Crawler-friendly fallback below for screen readers and bots that don't run JS.
OpenClawMU is the multi-tenant fork of OpenClaw — a self-hosted AI gateway that bridges WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Teams, Matrix, LINE, Lark, Google Chat and WebChat. The MU fork adds tenant isolation, per-tenant SHA-256 tokens, sandboxed agents (bubblewrap or Docker), browser-based web terminals, per-tenant quotas and billing, and S3-compatible backup/restore — so one gateway can safely serve many teams or customers.
Upstream OpenClaw is a personal AI assistant — one user, one device, one session store. OpenClawMU keeps every upstream channel and agent feature, then layers tenant token auth, isolated session/memory/plugin/sandbox/cron directories per tenant, tenant-scoped HTTP APIs (OpenAI- and OpenResponses-compatible), per-tenant quotas, S3 backup, and a control-plane HTTP API for admin resource management. Fork commits are labelled OPENCLAWMU ADDITION so you can rebase against upstream cleanly.
Platform builders (the bot/agent platform you're shipping needs multi-tenant isolation without inventing it from scratch), hosting providers (you want to offer self-hosted AI assistants to multiple customers), enterprise IT (you need one gateway your departments share without sharing data), and bot developers ready to graduate from a single-user prototype to a multi-customer product.
Yes. OpenClawMU exposes OpenAI-compatible /v1/chat/completions and OpenResponses /v1/responses endpoints, scoped to a tenant by token. You can point any OpenAI SDK at your gateway and the tenant's sessions, memory, and quotas apply automatically — no SDK changes for your end users.
Yes. The entire codebase is Apache-2.0 on GitHub. You self-host the gateway on your own infrastructure — Tailscale Serve/Funnel, SSH tunnel, your own reverse proxy, or a managed VM. Anthropic/OpenAI/your-LLM-of-choice keys live in your environment. We don't run servers between you and your users.
npm install -g openclaw@latest, then openclaw onboard --install-daemon for an interactive wizard. Or use the Docker image. Once the gateway is running, openclaw tenants create acme provisions your first tenant and mints a SHA-256 token. From there, point your messaging channel webhooks at the gateway and you're routing traffic. Full instructions on the install page.
EXFOLIATE!
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.