Configuration

Configuration

Zinc reads workspace .zinc/config.yaml and global ~/.zinc/config.yaml.

packages:
  responses: openai-responses@0.1.5

runtime:
  parallel: true
  max_parallel: 0

models:
  default: local-llama

  local-llama:
    adapter: responses.responses
    params:
      endpoint: http://127.0.0.1:30000
      endpoint_kind: chat_completions
      model: local-gemma-4-e4b-it
      temperature: 0.2

max_parallel: 0 uses the host CPU count.

A model preset names an adapter asset and passes params to it. The adapter receives the part name, params, instructions, typed inputs, and expected outputs.