LLM-Box Code — Private Coding Assistant (Qwen2.5-Coder-7B)

LLM-Box Code is a ready-to-run private coding assistant for teams that must not send source code to external services (banks, public sector, NDA development).

— Qwen2.5-Coder-7B-Instruct model: code generation, explanation, refactoring, bug finding and fixing, test writing; 40+ programming languages;
— OpenAI-compatible API (/v1) that plugs into IDE extensions and CI, plus a web chat (Open WebUI);
— runs on a single NVIDIA T4 GPU, fully offline; code and data stay inside your perimeter;
— ready image: launch the VM and get a working coding assistant in minutes.

1. Create a VM from this image on an NVIDIA T4 platform (8 vCPU, 32 GB RAM), 60 GB disk.
2. Assign a public IP; open ports 22 (SSH) and 80 in the security group.
3. Add your SSH key when creating the VM.
4. After start, wait 6-8 minutes while the model loads into the GPU (“No models” until then is normal).
5. Open http://<public IP>/ — the first login creates the administrator. Model: qwen2.5-coder-7b.
   The URLs and API key are shown in the SSH login banner.

— Coding assistant in the IDE via an OpenAI-compatible API, with no source-code leakage;
— code generation, refactoring and test coverage inside a protected perimeter;
— legacy-code explanation and review in a private environment;
— automation of routine development in CI without external cloud LLMs.

Technical support by email: info@papadata.ru.
We assist with deployment and operation, responding within 1–2 business days.