haive.agents.research.deep_research_agent

Deep Research Agent — advanced multi-stage research with RAG + search + analysis.

Uses multiple specialized agents composed via MultiAgent: 1. QueryPlanner (SimpleAgent) — decomposes question into sub-queries 2. WebResearcher (ReactAgent) — Tavily search + store findings 3. RAGAnalyzer (ReactAgent) — retrieves + analyzes stored findings with RAG tools 4. FactChecker (SimpleAgent) — validates consistency, flags contradictions 5. ReportWriter (SimpleAgent) — synthesizes into cited report

More thorough than perplexity_agent — designed for deep research tasks.

Usage:

from haive.agents.research.deep_research_agent import create_deep_research_agent agent = create_deep_research_agent() result = agent.run(“Compare transformer vs state space model architectures for LLMs”)

Classes

DeepResearchAgent

Advanced multi-stage research agent.

ResearchStore

In-session document store for research findings.

Functions

create_deep_research_agent([name, tools, ...])

Create a deep research agent with 5-stage pipeline.

Module Contents

class haive.agents.research.deep_research_agent.DeepResearchAgent

Bases: haive.agents.multi.agent.MultiAgent

Advanced multi-stage research agent.

Pipeline: Planner → Researcher → Analyzer → FactChecker → Writer

class haive.agents.research.deep_research_agent.ResearchStore

In-session document store for research findings.

haive.agents.research.deep_research_agent.create_deep_research_agent(name='deep_research', tools=None, max_search_iterations=10, include_fact_check=True)

Create a deep research agent with 5-stage pipeline.

Parameters:

  • name (str) – Agent name

  • tools (list | None) – Custom search tools (defaults to Tavily or mock)

  • max_search_iterations (int) – Max search iterations for researcher

  • include_fact_check (bool) – Include fact-checking stage

Returns:

DeepResearchAgent ready for execution

Return type:

DeepResearchAgent