dromerosm/crewai-multiagent-research

CrewAI Multiagent Research Tool

This Space runs a multi-agent research workflow built with CrewAI and Gradio. It searches the web, extracts relevant content, and returns a final report.

Current stack

• —crewai with multi-agent orchestration.
• —gradio as UI runtime in Hugging Face Spaces.
• —duckduckgo-search, newspaper4k, and PyMuPDF for retrieval.
• —OpenAI, Azure OpenAI, and Groq model options in the interface.

CrewAI alignment

This repository is aligned with the current CrewAI docs:

• —Intro: <https://docs.crewai.com/en/introduction>
• —Core concepts used here:
• —Agents
• —Tasks
• —Crews
• —Tools

Recent CrewAI versions may evolve output metadata. The app handles token metrics defensively to support both token_usage and usage_metrics structures when available.

How it works

1. 1.The Researcher agent performs search and source collection.
2. 2.The Editor agent refines findings into a structured final report.
3. 3.The app streams verbose execution logs in the UI.
4. 4.The generated report can be exported to outputs/output.md.

Local run

1. 1.Create a virtual environment:

python3 -m venv .venv
source .venv/bin/activate

1. 1.Install dependencies:

pip install -r requirements.txt

1. 1.Run the app:

python3 app.py

Hugging Face Space config

The Space is currently functional and running. For deployment operations, this repo uses:

• —HF_TOKEN
• —HF_SPACE_ID

Both values must exist in .env locally for deployment scripts/API calls.

Notes

• —.env is gitignored and must never be committed.
• —Secrets are managed in Hugging Face Space settings/API.
• —Variables can be listed via API; secret values are never retrievable.