# Local (or VPS) zero-GPU OpenJev: llama.cpp CPU + official openjev-server.
#
#   docker compose up --build -d
#   curl http://localhost:7860/v1/systemone ...
#
# Optional overrides:
#   MODEL_FILE=OpenJev-Q8_0.gguf            bigger quant, same pipeline
#   OPENJEV_TOKEN=secret                    require Authorization: Bearer secret on /v1/*
#   LLAMA_THREADS=8 LLAMA_THREADS_BATCH=8  bound the CPU instead of using all cores
services:
  openjev:
    build: .
    ports:
      - "7860:7860"
    environment:
      OPENJEV_TOKEN: ${OPENJEV_TOKEN:-}
      MODEL_FILE: ${MODEL_FILE:-OpenJev-Q4_K_M.gguf}
      LLAMA_THREADS: ${LLAMA_THREADS:-}
      LLAMA_THREADS_BATCH: ${LLAMA_THREADS_BATCH:-}
      LLAMA_CTX: ${LLAMA_CTX:-8192}
    volumes:
      - openjev-data:/data
volumes:
  openjev-data: