# enterprise-deep-research **Repository Path**: liyan24/enterprise-deep-research ## Basic Information - **Project Name**: enterprise-deep-research - **Description**: No description available - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-03-02 - **Last Updated**: 2026-03-03 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README

EDR Logo

Python 3.11+ arXiv HF Dataset DeepWiki Documentation License: Apache 2.0 GitHub stars Follow us on X Follow us on LinkedIn

## πŸ“° News - **2025/12/12**: πŸ₯‡ EDR **ranks #1** on the [LiveResearchBench leaderboard](https://livedeepresearch.github.io/#leaderboard)! - **2025/11/15**: Our technical report is accepted at AAAI Agentic AI Benchmarks and Applications Workshop. - **2025/11/13**: EDR is officially on the **[DeepResearchBench leaderboard](https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard)** --- We present **Enterprise Deep Research (EDR)**, a multi-agent system that integrate: - Master Planning Agent for adaptive query decomposition. - Four specialized search agents (General, Academic, GitHub, LinkedIn). - Extensible MCP-based tool ecosystem supporting NL2SQL, file analysis, and enterprise workflows. - Visualization Agent for data-driven insights. - Reflection mechanism that detects knowledge gaps and updates research direction with optional human-in-the-loop steering guidance. - Real-time steering commands for continuous research refinement. > [!Note] > These components enable automated report generation, real-time streaming, and seamless enterprise deployment, as validated on internal datasets. ![Architecture Overview](./assets/edr_ppl.png) ## πŸŽ₯ Demo We present a video demo of using EDR in web application for enterprise deep data analysis. #### EDR: Web Application > [!Note] > Multi-provider LLM support β€’ Slack agent β€’ Real-time streaming β€’ Document analysis β€’ Citation management β€’ Parallel processing β€’ Specialized benchmarking β€’ Human-in-the-loop steering ## πŸš€ Quick Start **Requirements**: Python 3.11+ β€’ Node.js 20.9.0+ ### Installation & Setup ```bash # Clone and setup git clone https://github.com/SalesforceAIResearch/enterprise-deep-research.git cd enterprise-deep-research # Python environment python -m venv venv source venv/bin/activate # Windows: venv\Scripts\activate pip install -r requirements.txt # Configure environment cp .env.sample .env # Edit .env with your API keys # Frontend setup cd ai-research-assistant && npm install && npm run build && cd .. ``` ### Environment Configuration **Required Variables:** - `TAVILY_API_KEY` - Tavily search API key - **One LLM provider key:** - `OPENAI_API_KEY` - OpenAI API key - `ANTHROPIC_API_KEY` - Anthropic API key - `GROQ_API_KEY` - Groq API key - `GOOGLE_CLOUD_PROJECT` - Google Cloud project ID - `SAMBNOVA_API_KEY` - SambaNova API key **Optional Settings:** - `LLM_PROVIDER` - Default provider (default: `openai`) - `LLM_MODEL` - Model name (provider-specific defaults) - `MAX_WEB_RESEARCH_LOOPS` - Max iterations (default: `10`) ### Supported Models | Provider | Default Model | Available Models | |----------|---------------|------------------| | **OpenAI** | `o4-mini` | `o4-mini`, `o4-mini-high`, `o3-mini`, `o3-mini-reasoning`, `gpt-4o` | | **Anthropic** | `claude-sonnet-4` | `claude-sonnet-4`, `claude-sonnet-4-thinking`, `claude-3-7-sonnet`, `claude-3-7-sonnet-thinking` | | **Google** | `gemini-2.5-pro` | `gemini-2.5-pro`, `gemini-1.5-pro-latest`, `gemini-1.5-flash-latest` | | **Groq** | `deepseek-r1-distill-llama-70b` | `deepseek-r1-distill-llama-70b`, `llama-3.3-70b-versatile`, `llama3-70b-8192` | | **SambaNova** | `DeepSeek-V3-0324` | `DeepSeek-V3-0324` | ### Running the Application **Full Stack (Recommended) - Single Command:** ```bash python -m uvicorn app:app --host 0.0.0.0 --port 8000 ``` The application will serve both the backend API and pre-built frontend at [http://localhost:8000](http://localhost:8000) **Backend API Documentation**: [http://localhost:8000/docs](http://localhost:8000/docs) ## πŸ’» Usage ### Command Line ```bash python benchmarks/run_research.py "Your research question" \ --provider openai --model o3-mini --max-loops 3 ``` ### Web Interface Navigate to [http://localhost:8000](http://localhost:8000) for interactive research with real-time progress tracking. ## πŸ“š Benchmarking & Development ### Supported Benchmarks ![Benchmarking Results](./assets/leaderboard.png) - **DeepResearchBench**: Comprehensive research evaluation - **ResearchQA**: Question-answering with citation verification - **DeepConsult**: Consulting-style analysis tasks ### EDR-200 Dataset The **[EDR-200 dataset](https://huggingface.co/datasets/Salesforce/EDR-200)** contains 201 complete agentic research trajectories generated by Enterprise Deep Researchβ€”99 queries from DeepResearch Bench and 102 queries from DeepConsult. Unlike prior benchmarks that only capture final outputs, these trajectories expose the full reasoning process across search, reflection, and synthesis steps, enabling fine-grained analysis of agentic planning and decision-making dynamics. ### Running Benchmarks Refer to our detailed [benchmarking guide](benchmarks/README.md). ### Development Setup ```bash # Testing python -m pytest tests/ python test_agents.py # Code quality black src/ services/ benchmarks/ mypy src/ services/ flake8 src/ services/ benchmarks/ # Development server python -m uvicorn app:app --reload --host 0.0.0.0 --port 8000 cd ai-research-assistant && npm run dev ``` ## πŸ“ Project Structure ```text enterprise-deep-research/ β”œβ”€β”€ ai-research-assistant/ # React frontend β”œβ”€β”€ benchmarks/ # Evaluation framework β”œβ”€β”€ src/ # Core research engine β”‚ β”œβ”€β”€ agent_architecture.py # Multi-agent orchestration β”‚ β”œβ”€β”€ graph.py # LangGraph workflow definitions β”‚ β”œβ”€β”€ state.py # Research state management β”‚ β”œβ”€β”€ simple_steering.py # Steering & task management β”‚ β”œβ”€β”€ steering_integration.py # Steering integration layer β”‚ β”œβ”€β”€ prompts.py # Agent prompts & templates β”‚ β”œβ”€β”€ configuration.py # Agent configuration β”‚ β”œβ”€β”€ utils.py # Utility functions β”‚ β”œβ”€β”€ visualization_agent.py # Visualization generation β”‚ └── tools/ # Research tools & MCP integration β”œβ”€β”€ services/ # Backend services (research, analysis, parsing) β”œβ”€β”€ routers/ # FastAPI endpoints β”œβ”€β”€ models/ # Data schemas β”œβ”€β”€ app.py # Main FastAPI application β”œβ”€β”€ llm_clients.py # LLM provider clients β”œβ”€β”€ session_store.py # Session management └── requirements.txt # Python dependencies ``` ## Star History Star History Chart ## πŸ“œ License & Citation Licensed under [Apache 2.0](./LICENSE.txt). ```bibtex @article{prabhakar2025enterprisedeepresearch, title={Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics}, author={Prabhakar, Akshara and Ram, Roshan and Chen, Zixiang and Savarese, Silvio and Wang, Frank and Xiong, Caiming and Wang, Huan and Yao, Weiran}, journal={arXiv preprint arXiv:2510.17797}, year={2025} } ``` ## πŸ“¬ Contributions We would love to hear from the broader AI community, and we welcome any contributions, pull requests, or issues! Please refer to [contributing guidelines](./CONTRIBUTING.md). For questions, suggestions, or contributions, please contact: [Akshara Prabhakar](https://aksh555.github.io/) **Acknowledgments**: Built on [LangGraph](https://github.com/langchain-ai/langgraph), [Tavily](https://tavily.com), [React](https://reactjs.org/), [Tailwind CSS](https://tailwindcss.com/), and [FastAPI](https://fastapi.tiangolo.com/).