haive.agents.graphs¶

LangGraph Platform graph entry points for haive-agents.

This module exposes all workable agent graphs as callable factories for LangGraph Platform (langgraph.json). Each function returns a compiled LangGraph StateGraph served via langgraph dev or LangGraph Cloud.

Functions¶

`collaborative_conversation`()	Collaborative multi-agent conversation.
`debate_conversation`()	Structured debate between agents.
`document_grading_rag`()	Document Grading RAG: grades retrieved documents for relevance.
`dynamic_rag`()	Dynamic RAG with multiple retrieval sources.
`flare_rag`()	FLARE RAG: forward-looking active retrieval.
`fusion_rag`()	RAG Fusion: reciprocal rank fusion across multiple retrievals.
`lats`()	LATS: Monte Carlo Tree Search over LLM responses with reflection scoring.
`llm_compiler`()	LLM Compiler: structured DAG planner + parallel executor + joiner.
`llm_rag`()	LLM-based RAG agent.
`memory_agent`()	Memory-enhanced agent with vector store + knowledge graph.
`multi_agent_sequential`()	Sequential multi-agent pipeline (agent A -> B -> C).
`multi_query_rag`()	Multi-Query RAG: generates multiple query variants for retrieval.
`plan_and_execute`()	Plan-and-execute agent: creates a plan then executes each step.
`rag_agent`()	Base retrieval-augmented generation agent (HuggingFace embeddings).
`react_agent`()	ReAct agent with reasoning + acting loop and tool-use.
`reasoning_system`()	Logic-based reasoning system with multiple engines.
`reflection_agent`()	Self-reflection agent that generates then improves its answer.
`reflexion`()	Reflexion: draft -> reflect -> revise loop with ReactAgent + SimpleAgent.
`rewoo`()	ReWOO: plan all steps upfront, execute in parallel, synthesize answer.
`round_robin_conversation`()	Round-robin conversation: each agent speaks in fixed order.
`self_discover_adapter`()	Self-Discover adapter: adapts modules to the task.
`self_discover_executor`()	Self-Discover executor: executes the structured reasoning plan.
`self_discover_selector`()	Self-Discover selector: picks relevant reasoning modules.
`self_discover_structurer`()	Self-Discover structurer: structures the reasoning plan.
`self_route_rag`()	Self-Route RAG: analyzes query to choose optimal retrieval strategy.
`simple_agent`()	Basic conversational agent with hooks and structured output.
`simple_rag`()	Simple RAG = retriever + answer agent in sequence.
`speculative_rag`()	Speculative RAG: hypothesis generation + parallel verification.
`step_back_rag`()	Step-Back RAG: generates abstracted queries for better retrieval.
`supervisor`()	Dynamic supervisor that routes tasks to sub-agents.
`task_analysis`()	Task analysis and decomposition agent.

Module Contents¶

haive.agents.graphs.collaborative_conversation()¶: Collaborative multi-agent conversation.

haive.agents.graphs.debate_conversation()¶: Structured debate between agents.

haive.agents.graphs.document_grading_rag()¶: Document Grading RAG: grades retrieved documents for relevance.

haive.agents.graphs.dynamic_rag()¶: Dynamic RAG with multiple retrieval sources.

haive.agents.graphs.flare_rag()¶: FLARE RAG: forward-looking active retrieval.

haive.agents.graphs.fusion_rag()¶: RAG Fusion: reciprocal rank fusion across multiple retrievals.

haive.agents.graphs.lats()¶: LATS: Monte Carlo Tree Search over LLM responses with reflection scoring.

haive.agents.graphs.llm_compiler()¶: LLM Compiler: structured DAG planner + parallel executor + joiner.

haive.agents.graphs.llm_rag()¶: LLM-based RAG agent.

haive.agents.graphs.memory_agent()¶: Memory-enhanced agent with vector store + knowledge graph.

haive.agents.graphs.multi_agent_sequential()¶: Sequential multi-agent pipeline (agent A -> B -> C).

haive.agents.graphs.multi_query_rag()¶: Multi-Query RAG: generates multiple query variants for retrieval.

haive.agents.graphs.plan_and_execute()¶: Plan-and-execute agent: creates a plan then executes each step.

haive.agents.graphs.rag_agent()¶: Base retrieval-augmented generation agent (HuggingFace embeddings).

haive.agents.graphs.react_agent()¶: ReAct agent with reasoning + acting loop and tool-use.

haive.agents.graphs.reasoning_system()¶: Logic-based reasoning system with multiple engines.

haive.agents.graphs.reflection_agent()¶: Self-reflection agent that generates then improves its answer.

haive.agents.graphs.reflexion()¶: Reflexion: draft -> reflect -> revise loop with ReactAgent + SimpleAgent.

haive.agents.graphs.rewoo()¶: ReWOO: plan all steps upfront, execute in parallel, synthesize answer.

haive.agents.graphs.round_robin_conversation()¶: Round-robin conversation: each agent speaks in fixed order.

haive.agents.graphs.self_discover_adapter()¶: Self-Discover adapter: adapts modules to the task.

haive.agents.graphs.self_discover_executor()¶: Self-Discover executor: executes the structured reasoning plan.

haive.agents.graphs.self_discover_selector()¶: Self-Discover selector: picks relevant reasoning modules.

haive.agents.graphs.self_discover_structurer()¶: Self-Discover structurer: structures the reasoning plan.

haive.agents.graphs.self_route_rag()¶: Self-Route RAG: analyzes query to choose optimal retrieval strategy.

haive.agents.graphs.simple_agent()¶: Basic conversational agent with hooks and structured output.

haive.agents.graphs.simple_rag()¶: Simple RAG = retriever + answer agent in sequence.

haive.agents.graphs.speculative_rag()¶: Speculative RAG: hypothesis generation + parallel verification.

haive.agents.graphs.step_back_rag()¶: Step-Back RAG: generates abstracted queries for better retrieval.

haive.agents.graphs.supervisor()¶: Dynamic supervisor that routes tasks to sub-agents.

haive.agents.graphs.task_analysis()¶: Task analysis and decomposition agent.