Host a private AI model server on one powerful GPU computer, and let friends or family members connect their own OpenClaw agents to it from anywhere. Each person keeps their own agent workspace, documents, email, and schedules separate — while sharing the same private AI compute.
AI agents are becoming useful for everyday tasks: managing email, organizing schedules, summarizing family documents, drafting messages, planning trips, and helping with work notes. Tools such as OpenClaw make these workflows practical by giving each user their own browser-based agent workspace, or by connecting messaging apps such as Telegram, WhatsApp, or Discord through the OpenClaw gateway.
At the same time, self-hosted AI is becoming much more realistic. Open and open-weight models are improving quickly, and more capable local AI hardware is reaching individuals, small teams, and families — from gaming GPUs and workstations to unified-memory desktops and personal AI systems. For many daily assistant tasks, a privately hosted model server is no longer just a hobby experiment; it can be practical infrastructure.
But running AI agents still requires access to a capable model server. Cloud AI subscriptions can become expensive when every friend or family member needs a separate account. And with cloud-based AI, sensitive prompts leave your own infrastructure and are processed by the provider’s servers, subject to that provider’s retention, logging, and data-use policies.
Meanwhile, one person in the group may already own a powerful computer with enough GPU memory or unified memory to run a useful local model. The challenge is sharing that AI server safely with others — without exposing it to the public internet, opening router ports, or forcing everyone onto a VPN.
One person hosts the AI model server — for example, Ollama, vLLM, LM Studio, or MTCode-LLM — on a GPU computer. Each friend or family member runs their own OpenClaw instance and connects it to that shared model through MTCode DirectLink.
Everyone gets their own agent workspace, while the expensive GPU compute is shared.
OpenClaw agents often perform scheduled or background tasks — checking messages, summarizing documents, drafting replies, or preparing reminders. Friends and family members do not always need the model at the exact same moment, so a shared AI server can serve multiple personal agents efficiently.
Each person keeps a separate OpenClaw workspace, while scheduled jobs can use the shared model server at different times.
Send invitations to friends or family members. Each person signs in with their own account, and you can grant or revoke access without sharing passwords or changing router settings.
MTCode coordinates authentication, authorization, and connection setup. Once connected, model requests flow directly between each user and the shared AI server over an end-to-end encrypted peer-to-peer channel.
localhost:8000. The port number can be changed to any specific value if needed.The scenario above places each person's OpenClaw agent on their own computer. There is another arrangement that can be even more practical for groups that want around-the-clock AI coverage: all OpenClaw agents run on the same shared host computer, which stays on 24 hours a day.
Each group member's OpenClaw instance runs on the shared machine with its own separate account credentials, working directory, and a unique gateway port number for its Control UI. The administrator publishes each gateway port through MTCode Server under a distinct name — for example, "Alice's Agent" or "Bob's Agent" — and each member uses MTCode Portal to remotely access and manage their own agent from any device, without needing physical access to the host machine.