Run private coding assistants against an LLM and Git server you control. MTCode maps remote services to local ports so your tools work as if everything were running on your laptop.
AI coding assistants need access to code context. For closed-source projects, contractor work, security-sensitive codebases, or NDA-protected repositories, sending that context to hosted models may be unacceptable.
At the same time, self-hosted AI coding is becoming much more realistic. Open-weight coding models are improving quickly, and more capable local AI hardware is reaching developers, small teams, and offices β from high-VRAM gaming GPUs and workstations to unified-memory desktops and personal AI systems. For many daily coding tasks, a private model server is no longer just an experiment; it can be a practical development tool.
A common solution is to run a private LLM on a high-VRAM GPU at home or in the office, and to keep source code in a self-hosted Git server such as Gitea, Forgejo, or GitLab CE. But those services are tied to one location while your laptop moves between home, office, client sites, and travel.
Opening ports exposes services to the public internet. VPNs add network-level complexity. Tunnel services add a third-party relay. MTCode DirectLink lets you access selected private services from anywhere while keeping them off the public internet.
Register your Ollama, LM Studio, vLLM, or MTCode-LLM endpoint with MTCode Server. MTCode Portal maps it to a local address on your laptop, so compatible coding tools can connect as if the model were local.
Publish your self-hosted Git service through MTCode Server. Clone, pull, and push through a local Portal mapping without exposing the Git server publicly.
MTCode shares only the services you register β for example, an LLM endpoint and a Git web interface β not the entire computer or local network.
11434 or the web port for Gitea.localhost:11434.