graph TD %%{init: {'theme': 'mc','layout': 'elk'}}%% ChatInput-im722[<div><img src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Input] style ChatInput-im722 stroke:#a170ff LanggraphReactAgent-yf4ap[Asistente de Investigación] style LanggraphReactAgent-yf4ap stroke:#a170ff ChatOutput-7h4uy[<div><img src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output] style ChatOutput-7h4uy stroke:#a170ff planner_optimizer_tool-5qmm8[Planner and Optimizer Tool2] style planner_optimizer_tool-5qmm8 stroke:#a170ff ThinkToolComponent-2cn4v[Think Tool] style ThinkToolComponent-2cn4v stroke:#a170ff SearXng-xyva2[Web Search ] style SearXng-xyva2 stroke:#a170ff OpenAIModel-xyjkl[<div><img src="/_astro/openAI.BhmuxEs3.svg" style="height: 20px !important;width: 20px !important"/></div>OpenAI] style OpenAIModel-xyjkl stroke:#a170ff ChatInput-im722 -.- LanggraphReactAgent-yf4ap linkStyle 0 stroke:#a170ff OpenAIModel-xyjkl -.- LanggraphReactAgent-yf4ap linkStyle 1 stroke:#a170ff LanggraphReactAgent-yf4ap -.- ChatOutput-7h4uy linkStyle 2 stroke:#a170ff planner_optimizer_tool-5qmm8 -.- LanggraphReactAgent-yf4ap linkStyle 3 stroke:#a170ff ThinkToolComponent-2cn4v -.- LanggraphReactAgent-yf4ap linkStyle 4 stroke:#a170ff SearXng-xyva2 -.- LanggraphReactAgent-yf4ap linkStyle 5 stroke:#a170ff

Structured Research Workflow

The Structured Research Workflow is an autonomous research agent that follows a strict four‑phase protocol—Plan, Act, Process, Synthesize—to respond to complex user queries. It automatically decomposes a question, conducts web searches, extracts insights, and synthesises a comprehensive report, all while leveraging a large‑language model and specialised tools.

🧩 Overview

This workflow transforms a user’s natural‑language question into a detailed, evidence‑based answer. By orchestrating planning, searching, deep‑analysis, and synthesis steps, the agent eliminates manual research effort and produces high‑quality, structured insights with minimal user intervention.

⚙️ Main Features

Orchestrates an end‑to‑end research cycle that never ends until all plan items are addressed.
Uses a large‑language model to guide tool selection and reasoning.
Implements a planner‑optimizer tool to generate and refine research plans.
Performs live web searches with a dedicated search engine.
Applies a thinking tool to transform raw search results into structured knowledge.
Synthesises collected data into a final, coherent report.
Provides a chat‑style interface for input and output.

🔄 Workflow Steps

Component Name	Role in the Workflow	Key Inputs	Key Outputs
Chat Input	Captures the user’s query and conversation context.	User message (question)	Message object
OpenAI Model	Generates natural‑language responses for the agent’s prompts.	Prompt from the agent	Generated text
Langgraph React Agent	Orchestrates the entire research process, selecting tools and aggregating results.	User query, language model, planner & optimizer tool, web search tool, think tool	Final answer message
Planner & Optimizer Tool	Creates a detailed research plan from the user’s question and progress updates.	Original question and any partial findings	Structured plan (markdown)
Web Search	Retrieves up‑to‑date information from the web for each plan item.	Search query	Search results data
Think Tool	Processes raw search results into structured summaries and key insights.	Raw search results	Structured analysis (markdown)
Chat Output	Displays the agent’s final answer to the user.	Final answer message	Visible chat message

All tool components are invoked internally by the Langgraph React Agent; the table shows their logical order in the research cycle.

🧠 Notes

The agent follows the P.A.P.S. protocol: Plan → Act → Process → Synthesize. It will not skip any step, ensuring a thorough investigation.
The OpenAI Model is the core language model used; the workflow can be adapted to other LLMs by swapping the model component.
Web Search is performed using the SearXng engine, which supports a wide range of file types and queries.
The Think Tool is purely a processing step; it does not access external resources beyond the data supplied by the web search.
The planner‑optimizer tool is the only component that requires contextual reasoning about the user’s original question; it refines the plan as the workflow progresses.
The workflow relies on external APIs (OpenAI, SearXng) and thus requires network connectivity and valid credentials.
All text and data are treated as metadata; the agent stores messages only if explicitly configured to do so.
The final answer is generated in natural language, but the agent can be extended to output structured JSON by providing an appropriate schema.