agently-model-setup（Agent 模型设置）

Agently Model Setup

Use this skill for provider wiring and transport setup before request logic is discussed.

Native-First Rules

• default to async-first guidance when the configured model will be used from services, streaming paths, or concurrent workflows
• when settings live in files, prefer Agently.load_settings("yaml_file", path, auto_load_env=True)
• use Agently.set_settings(...) or agent.set_settings(...) for inline mappings or host-owned overrides
• prefer settings files with ${ENV.xxx} placeholders for base URL, model, and auth
• put provider settings under the namespace read by the owning plugin. For OpenAICompatible, prefer plugins.ModelRequester.OpenAICompatible.*
• call the matching settings loader with auto_load_env=True when the payload may rely on .env
• if the app must fail fast, validate required env names in the integration layer before calling Agently
• after loading, verify the effective provider activation, base URL, model, and auth presence instead of assuming the file shape was correct
• keep provider setup outside business workflow logic and prompt files

Anti-Patterns

• do not hardcode provider-specific parsing into request code
• do not bake secrets or environment-specific endpoints into committed Python code when settings plus env placeholders fit
• do not leave provider config at a root-level namespace that the active plugin will not read
• do not let sync-only samples dictate the architecture of async-capable services
• do not mix model setup with output parsing or workflow design

Read Next

• references/overview.md