MultiUserClaw

What MultiUserClaw does

MultiUserClaw adds multi-tenancy around Hermes Agent. A Vite and React frontend talks to a FastAPI platform gateway, which handles JWT authentication, starts or finds the user's container and proxies WebSocket traffic to the Hermes API Server inside it. Each dedicated user gets a separate Docker container with 2GB of RAM and 4 CPUs, created on demand and paused when idle. A shared runtime mode also exists.

When the agent needs a model, requests go back through the gateway at /llm/v1/chat/completions, which checks the container token and user quota, matches a provider and injects the real API key. Providers named in the README include DashScope, Anthropic, OpenAI, DeepSeek, AiHubMix, Evolink and OpenRouter. The web app covers agent management, chat, cron jobs, a per-agent knowledge base, a skills store sourced from skills.sh, and channels such as Telegram, Discord, email, WhatsApp Web, Signal, Slack and iMessage. Upgrading Hermes means rebuilding the Hermes image. The README is in Chinese.

Key features

• One isolated Docker container per user running the Hermes API Server
• Platform gateway with JWT auth, quota checks and real API key injection
• Web UI for agents, chat, cron jobs, knowledge base files and the skills store
• Channel support for Telegram, Discord, email, WhatsApp Web, Signal, Slack and iMessage
• Multiple model providers, including DashScope, Anthropic, OpenAI, DeepSeek and OpenRouter
• Hermes upgrades by rebuilding the Hermes image

When to use it

• Offering Hermes-based assistants to many users with isolated runtimes
• Building a commercial AI assistant SaaS with per-user quotas
• Hosting an internal team assistant with channel access and scheduled tasks

Who it is for: Developers who want to host Hermes Agent for many users on shared infrastructure.

How it fits with Hermes Agent

The main branch uses Hermes Agent as its default runtime, and it is deployed as a container per user whose API Server the platform calls.

Requirements: Docker for the per-user containers, which use 2GB of RAM and 4 CPUs each, and an LLM provider API key.

Note: The main branch is based on a late-June Hermes version, and the earlier OpenClaw-based branch is archived.

FAQ