agents.rag.step_back.agent

Step-Back Prompting RAG Agents.

from typing import Any Implementation of step-back prompting for abstract reasoning. Generates broader conceptual queries for enhanced context retrieval.

Classes

DualRetrievalAgent

Agent that performs both original and step-back retrieval.

StepBackQuery

Step-back query generation result.

StepBackQueryGeneratorAgent

Agent that generates step-back queries for abstract reasoning.

StepBackRAGAgent

Complete Step-Back RAG agent with abstract reasoning.

StepBackResult

Combined results from step-back retrieval.

Functions

create_step_back_rag_agent(documents[, llm_config, ...])

Create a Step-Back RAG agent.

get_step_back_rag_io_schema()

Get I/O schema for Step-Back RAG agents.

Module Contents

class agents.rag.step_back.agent.DualRetrievalAgent(documents, embedding_model=None, max_docs_each=5, **kwargs)

Bases: haive.agents.base.agent.Agent

Agent that performs both original and step-back retrieval.

Initialize dual retrieval agent.

Parameters:

  • documents (list[langchain_core.documents.Document]) – Documents for retrieval

  • embedding_model (str | None) – Embedding model

  • max_docs_each (int) – Max docs to retrieve for each query

  • **kwargs – Additional arguments

build_graph()

Build dual retrieval graph.

Return type:

haive.core.graph.state_graph.base_graph2.BaseGraph

class agents.rag.step_back.agent.StepBackQuery(/, **data)

Bases: pydantic.BaseModel

Step-back query generation result.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

Parameters:

data (Any)

class agents.rag.step_back.agent.StepBackQueryGeneratorAgent(llm_config=None, abstraction_level='moderate', **kwargs)

Bases: haive.agents.base.agent.Agent

Agent that generates step-back queries for abstract reasoning.

Initialize step-back query generator.

Parameters:

  • llm_config (haive.core.models.llm.base.LLMConfig | None) – LLM configuration

  • abstraction_level (str) – Level of abstraction (“low”, “moderate”, “high”)

  • **kwargs – Additional agent arguments

build_graph()

Build step-back query generation graph.

Return type:

haive.core.graph.state_graph.base_graph2.BaseGraph

class agents.rag.step_back.agent.StepBackRAGAgent(config)

Bases: haive.agents.multi.base.SequentialAgent

Complete Step-Back RAG agent with abstract reasoning.

Init .

Parameters:

config (SequentialAgentConfig) – [TODO: Add description]

classmethod from_documents(documents, llm_config=None, embedding_model=None, abstraction_level='moderate', max_docs_each=5, **kwargs)

Create Step-Back RAG agent from documents.

Parameters:

  • documents (list[langchain_core.documents.Document]) – Documents to index

  • llm_config (haive.core.models.llm.base.LLMConfig | None) – LLM configuration

  • embedding_model (str | None) – Embedding model for retrieval

  • abstraction_level (str) – Level of abstraction for step-back queries

  • max_docs_each (int) – Max docs to retrieve for each query type

  • **kwargs – Additional arguments

Returns:

StepBackRAGAgent instance

class agents.rag.step_back.agent.StepBackResult(/, **data)

Bases: pydantic.BaseModel

Combined results from step-back retrieval.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

Parameters:

data (Any)

agents.rag.step_back.agent.create_step_back_rag_agent(documents, llm_config=None, reasoning_depth='moderate', **kwargs)

Create a Step-Back RAG agent.

Parameters:

  • documents (list[langchain_core.documents.Document]) – Documents for retrieval

  • llm_config (haive.core.models.llm.base.LLMConfig | None) – LLM configuration

  • reasoning_depth (str) – Depth of reasoning (“shallow”, “moderate”, “deep”)

  • **kwargs – Additional arguments

Returns:

Configured Step-Back RAG agent

Return type:

StepBackRAGAgent

agents.rag.step_back.agent.get_step_back_rag_io_schema()

Get I/O schema for Step-Back RAG agents.

Return type:

dict[str, list[str]]