.env configurator

Builds a ready-to-use .env for the on-premise stack. Everything runs in your browser — the generated secrets never leave this page. Use the file with ./install.sh --server-host <host> --env-file llmedi.env.

Server

Server host (hostname or IP that browsers will use — not localhost)

This must be the LAN name/IP clients use — it becomes the TLS certificate subject and every app URL.

Release version to pin (the version you fetch — shown on the downloads page) Transcription language

Hardware profile

Two GPUs (x86 server) — LLM on its own card (>40 GB), ASR on the other. Qwen3.6-27B-FP8 · VLLM_GPU_UTIL 0.9 / BATCH 0.4 One shared GPU — LLM and ASR split one card. Qwen3.6-27B-FP8 · VLLM_GPU_UTIL 0.5 / BATCH 0.4 GB10 / DGX Spark — NVFP4 on the stock aarch64 nightly image. nvidia/Qwen3.6-27B-NVFP4 via Marlin · fetch the qwen3.6-27b-nvfp4 model group and --skip-group vllm GB10 / DGX Spark — 35B-A3B MoE + DFlash (fastest, custom image). Run ./build-vllm-spark.sh on the box first · fetch with --skip-group vllm --skip-group vllm-aarch64

HTTPS

Caddy internal CA (default) — install the auto-exported llmedi-ca.crt on each client. Own certificate in ./certs/ — pair with the cert-builder from the downloads page (one command; renewals never touch clients).

Result

CONVEX_ADMIN_KEY stays empty — the installer derives and fills it on the box.