Multi-Agent-Research-Assistant/README.md

Multi-Agent Research Assistant

A multi-agent research assistant built with the OpenAI Agents SDK, Olostep, and Reflex.

Enter a research question and a team of AI agents collaborates to produce a polished, source-backed Markdown research report. The original notebook is included, and the same logic is also available as a Reflex web app.

Flow

User question
    |
    v
Manager agent
    |
    +--> Olostep Answer API
    |        |
    |        v
    |    Judge agent
    |        |
    |        +--> Good enough --> Analyst agent --> Markdown report
    |        |
    |        +--> Needs more evidence
    |                 |
    |                 v
    |          Source research agent
    |                 |
    |                 +--> Search with Scrape
    |                 +--> Targeted Search
    |                 +--> Scrape selected URLs
    |                 |
    |                 v
    |          Analyst agent
    |                 |
    |                 v
    +----------> Markdown research report + sources

Agents

Agent	Role
Manager	Orchestrates the workflow: answer, judge, research if needed, then write.
Judge	Evaluates whether the initial answer is good enough or needs deeper research.
Source Researcher	Gathers evidence using Olostep search, scrape, and targeted web queries. Prioritizes the most recent sources.
Analyst	Writes the final Markdown research report from the gathered evidence.

Setup

Install dependencies:

pip install -r requirements.txt

Create a .env file from .env.template:

OPENAI_API_KEY=your_openai_api_key
OLOSTEP_API_KEY=your_olostep_api_key
OPENAI_MODEL=gpt-5.4-mini

Run the Reflex app

reflex run

Then open the local URL printed by Reflex, usually:

http://localhost:3000

The app files live in app/:

• app/app.py — Reflex UI with styled Markdown report rendering and download.
• app/research_assistant.py — OpenAI Agents SDK multi-agent workflow with Olostep tools.

Features

• Multi-agent workflow — Manager, Judge, Source Researcher, and Analyst agents collaborate automatically.
• Live progress logs — Watch each agent step in real time.
• Styled Markdown report — Headings, bullets, tables, code blocks, and more render properly in the browser.
• Download report — Export the full Markdown report with one click.
• Recent results — The source research agent is aware of the current date and prioritizes up-to-date sources.

Run the notebook

Open and run:

multi_agent_research_assistant_openai_agents_olostep.ipynb

Set RUN_LIVE_EXAMPLE=true to run the live example. The notebook prints an OpenAI trace URL for inspecting the agent workflow.