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:
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, requiresOPENROUTER_API_KEYenv var) - Grok (X) (
grok:grok-2-1212, requiresX_AI_API_KEYenv var) - DeepSeek (
deepsek:deepsek-chat, requiresDEEPSEEK_API_KEYenv var)
For detailed model documentation and features, see the llmling-models repository.