haive.core.schema.field_registry¶

Field Registry for standardized field definitions across Haive.

This module provides a centralized registry of commonly used field definitions that can be referenced by nodes, engines, and schema composers. This ensures consistency and allows for selective state schema composition.

Key benefits: - Standardized field definitions across the framework - Selective inclusion in state schemas (only what’s needed) - Type safety with proper generics - Token counting integration for messages - Backwards compatibility

Classes¶

`CommonFieldSets`	Pre-defined sets of fields for common use cases.
`FieldRegistry`	Dynamic field registry for custom field definitions.
`PrebuiltStates`	Registry of prebuilt state schemas for common use cases.
`StandardFields`	Registry of standard field definitions used across Haive.

Functions¶

get_standard_field(name, **kwargs)

Get a standard field definition by name.

Module Contents¶

class haive.core.schema.field_registry.CommonFieldSets¶: Pre-defined sets of fields for common use cases.

class haive.core.schema.field_registry.FieldRegistry¶

Dynamic field registry for custom field definitions.

This complements StandardFields by allowing registration of custom field definitions at runtime.

classmethod clear()¶

Clear the registry (mainly for testing).

Return type:: None

classmethod get(name)¶

Get a registered field definition.

Parameters:: name (str)
Return type:: haive.core.schema.field_definition.FieldDefinition | None

classmethod list_fields()¶

List all registered field names.

Return type:: list[str]

classmethod register(field_def)¶

Parameters:: field_def (haive.core.schema.field_definition.FieldDefinition)
Return type:: None

class haive.core.schema.field_registry.PrebuiltStates¶

Registry of prebuilt state schemas for common use cases.

Hierarchy: - MessagesState (basic, no tokens) - MessagesStateWithTokenUsage (with token tracking)

LLMState (single engine + tokens + thresholds) - ToolState (tools + LLM features)

classmethod base_messages_state()¶

Get basic MessagesState without token tracking.

Return type:: Any

classmethod llm_state()¶

Get LLMState for single-engine LLM agents with token tracking and model awareness.

Return type:: Any

classmethod messages_with_tokens()¶

Get MessagesStateWithTokenUsage for token-aware conversations.

Return type:: Any

classmethod tool_state()¶

Get ToolState for tool-using agents with LLM features, tools, and token tracking.

Return type:: Any

class haive.core.schema.field_registry.StandardFields¶

Registry of standard field definitions used across Haive.

Each field is defined with: - name: field name in snake_case - type: proper Python type annotation - description: human-readable description - default: default value or factory - metadata: additional field metadata

classmethod ai_message()¶

Single AI message output field.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod available_nodes()¶

Available graph nodes.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod context()¶

Retrieved context documents.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod documents()¶

Retrieved documents with metadata.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod engine_name()¶

Name of the engine being used.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod human_message()¶

Single human message input field.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod messages(use_enhanced=True)¶

Standard messages field for conversation history.

Parameters:: use_enhanced (bool) – Whether to use the enhanced MessageList with token counting and metadata
Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod observations()¶

Agent observations from tools/environment.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod plan_steps()¶

Generated plan steps.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod query()¶

User query string.

Return type:: haive.core.schema.field_definition.FieldDefinition

classmethod structured_output(model_class, field_name=None)¶

Create a structured output field for a Pydantic model.

Parameters:

model_class (type[pydantic.BaseModel]) – The Pydantic model class
field_name (str | None) – Optional custom field name (defaults to snake_case model name)

Return type: