Introduction: From Amateur Coding to an AI Lab
Since my university days in Assembly and then C++, I’ve always struggled with writing clean code. Recently, agentic coding tools like Claude Code have helped me progress quickly, but each machine still required tedious configuration. My goal was simple: centralise everything on my Proxmox server and access it from any browser.
1. Why Proxmox for an AI Lab?
Proxmox offers lightweight virtualization thanks to LXC containers, ideal for hosting multiple agents without memory overhead. Its web interface also simplifies resource management and image updates.
Integrating an AI hub inside a container keeps isolation while sharing local models via OpenCode.
Resource Management
Each agent can be limited in CPU and RAM, preventing conflicts between heavy tasks such as fine‑tuning or code generation.
2. Setting Up the Jean Container
Jean, the open‑source tool from openLabs, doesn’t ship a model but orchestrates agent CLIs (Claude Code, Codex, Cursor…). It creates a dedicated worktree for each task so that changes stay isolated.
To deploy it:
- Create an LXC on Proxmox.
- Install Java and Git.
- Clone the Jean repository.
- Configure OpenCode to point at your local models.
Secure Connection with Caddy
Caddy acts as a HTTPS reverse proxy, using your personal domain. With an access token, the interface stays protected while remaining reachable from any browser.
3. Choosing the Right Local Models
The choice depends on the type of tasks: code generation, documentation or research. Open‑source models such as CodeLlama or Llama 2 Code offer a good balance between performance and cost.
It’s crucial to tweak prompt parameters to avoid the agent getting “lost” in infinite loops. A best practice is to set a token limit and an explicit timeout.
Prompt Optimization
Use structured prompts with clear examples, then let the agent generate code from a functional description.
4. Automating Updates and Maintenance
Proxmox lets you create regular snapshots of the Jean container. If corruption occurs or a rollback is needed, you can quickly return to a stable state.
Automation via Ansible or Terraform can trigger these snapshots at defined intervals, ensuring continuity for your AI lab.
“AI isn’t just about power; it’s also about organisation. A well‑structured environment lets you get the most out of agents without wasting time on configuration.”
Conclusion: An Accessible and Masterable AI Hub
By centralising your AI agents in Proxmox via Jean and Caddy, you gain a robust, secure, and easy‑to‑maintain platform. No more configuring each machine separately; everything is managed from your browser.
Ready to turn your server into a high‑performance AI lab? Download the Jean repository today, follow this guide, and start coding your own agents with no extra effort!
