Skip to content

Provider models

In addition to the regular pydantic-ai models, AgentPool supports all model types from llmling-models through YAML configuration. Each model is identified by its type field. These models often are some kind of "meta-models", allowing model selection patterns as well as human-in-the-loop interactions.

Basic Configuration

agents:
  my_agent:
    model:
      type: string           # Basic string model identifier
      identifier: gpt-5      # Model name

Available Model Types

See Config section to see the available types.

Model Settings

You can set common model settings to fine-tune the LLM behavior:

agents.yml

All settings are optional and providers will use their defaults if not specified.

Setting pydantic-ai models by identifier

AgentPool also extends pydantic-ai functionality by allowing to define more models via simple string identifiers. These providers are

  • OpenRouter (openrouter:provider/model-name, requires OPENROUTER_API_KEY env var)
  • Grok (X) (grok:grok-2-1212, requires X_AI_API_KEY env var)
  • DeepSeek (deepsek:deepsek-chat, requires DEEPSEEK_API_KEY env var)

For detailed model documentation and features, see the llmling-models repository.