Deep Dog 2: A completely open source deep research agent which beats OpenAI Deep Research (#5 overall and #1 OSS on DeepResearch Bench) using Deepseek v4 flash as the base model. Easy to install an...

Repository: github.com/beneadie/deep_dog_2 The quickest setup is: python -m pip install "git+github.com/beneadie/deep_dog_2.git" Add your provider keys to a .env file: DEEPSEEK_API_KEY=your-deepseek-key EXA_API_KEY=your-exa-key Then import it directly into Python: import asyncio from pathlib import Path from dotenv import load_dotenv load_dotenv() from deep_research.integration import run_research async def main(): result = await run_research( "What are the main benefits and limitations of sodium-ion batteries?" ) print(result.status) if result.status == "completed": Path("report.md").write_text(result.final_report, encoding="utf-8") print("Saved report.md") else: print(result.failure) asyncio.run(main()) The default setup uses DeepSeek V4 Flash for the supervisor, research sub-agents, and drafting, with Exa for web search. The result is returned as a Markdown string, so developers can print it, save it, send it to another application, or process it however they want. The more configurable quickstart lets you choose the models, search engine, enabled agents, research time, iteration limits, search budgets, read limits, and output behavior. Available specialist agents include Web, PubMed, Reddit, Substack, SEC Edgar, Arxiv, and others. The code is designed to be modified. Developers can add agents, change prompts, swap providers, alter the supervisor and sub-agent behavior, adjust budgets, or integrate the result into their own application. The engine is packaged so you can use the integration layer without having to rebuild the orchestration system from scratch. This project is completely free and released under the MIT License. I’m not building a business around it or offering a hosted service. The only potential costs are the provider APIs you choose to use, such as DeepSeek or Exa.