node_id: "node-1"
listen_host: "0.0.0.0"
listen_port: 8091

# Set to the dispatcher gateway URL if you want auto-registration/heartbeat.
dispatcher_base_url: "http://127.0.0.1:8080"
# Optional auth key presented to dispatcher for /v1/nodes/* endpoints
dispatcher_node_key: "change-me-node-key-1"
dispatcher_roles: ["planner", "coder"]
heartbeat_interval_sec: 5

llama_server_bin: "llama-server"

model_roots:
  - "/models"

models:
  - model_id: "planner-gguf"
    path: "/models/SomePlannerModel.Q5_K_M.gguf"
    roles: ["planner"]
    default_ctx: 8192
    server_args:
      # Examples (llama.cpp flags differ by build/version):
      # c: 8192
      # n_gpu_layers: 60
      # threads: 8
      # parallel: 1
      # keep: true
      c: 8192