Engine and model registry for Candil.Llm.

Stores engine and model definitions in ETS so they can be looked up by alias throughout the application. Definitions can be loaded at startup from application configuration or registered programmatically at runtime.

Application config

config :apero, Candil.Config,
  engines: [
    %{
      alias: :llama_server,
      use_precompiled: true,
      precompiled_version: :latest,
      host: "127.0.0.1",
      port: 8080,
      start_args: ["--n-gpu-layers", "35"]
    }
  ],
  models: [
    %{
      alias: :llama3,
      type: :local,
      model_dir: "/models",
      filename: "llama-3-8b-q4_k_m.gguf",
      context_size: 8192,
      engine: :llama_server,
      usage: [:chat, :completion]
    },
    %{
      alias: :gpt4o,
      type: :remote,
      name: "gpt-4o",
      context_size: 128_000,
      provider: :openai,
      usage: [:chat, :completion, :embeddings]
    }
  ],
  providers: [
    %{
      alias: :openai,
      type: :openai,
      base_url: "https://api.openai.com",
      api_key: {:system, "OPENAI_API_KEY"}
    }
  ]

Programmatic registration

Candil.Config.register_engine(%Candil.Engine{alias: :llama_server, ...})
Candil.Config.register_model(%Candil.Model{alias: :llama3, ...})
Candil.Config.register_provider(%Candil.Provider{alias: :openai, ...})

Values for :api_key can be:

• A plain string "sk-..." — used as-is.
• {:system, "ENV_VAR"} — resolved from the environment at lookup time.