agentic-deep-researcher

[](https://mseep.ai/app/shivamkgate-agentic-deep-researcher) # Agentic Deep Researcher This is a `MCP`-powered multi-agent deep researcher that performs real-time web research using agents through CrewAI, LinkUp API, and DeepSeek R1 as a local LLM. It uses a team of agents to web search, analyze the research, and finally synthesize the information into structured, cited insights. It orchestrates three roles: - **Web Searcher** Formulates and executes web searches via LinkUp API - **Research Analyst** Synthesizes raw search results into structured insights with citations of sources. - **Technical Writer** Produces clear, well-formatted Markdown responses with source links. Under the hood, the pipeline is defined in `backend/agents.py` and runs sequentially with built-in error handling and in-memory context storage. --- ## Features - Interactive chat UI (`Streamlit`) with history tracking and clear history functionality - Automated multi-step research pipeline: 1. Web searching process via LinkUp API 2. Structured analysis of collected results 3. Final drafting of Markdown responses - Configurable via environment variables - Extensible architecture: add new agents, tools, or output formats --- ## Installation 1. Clone the repository: ```powershell git clone https://github.com/ShivamKGate/agentic-deep-researcher.git cd agentic-deep-researcher ``` 2. Create and activate a Python virtual environment: ```powershell python -m venv .venv .\.venv\Scripts\Activate ``` 3. Install dependencies: ```powershell pip install -r requirements.txt ``` 4. Obtain your LinkUp API key from [LinkUp](https://app.linkup.so/sign-in) and configure API keys and model settings in a `.env` file: ```ini LINKUP_API_KEY=<your linkup api key> ``` 5. Download and install [Ollama](https://ollama.com/) locally, then pull the `deepseek-r1:7b` model by running: ```powershell ollama pull deepseek-r1:7b ``` > **Note:** Feel free to choose any DeepSeek model with higher or lower parameters (`1.5b`, `7b`, `8b`, `14b`, `32b`, `70b`, or `671b`) based on your available GPU resources and the desired level of reasoning quality during the model selection in the terminal. --- ## Usage ### `Streamlit` Chat UI ```powershell streamlit run backend/app.py ``` Open your browser at `http://localhost:8501` to chat with the multi-agent deep researcher! ### FastMCP RPC Server ```powershell python backend/server.py ``` Exposes an `async` tool `crew_research` for orchestration via FastMCP protocols. --- ## Configuration | Variable | Description | Default | |------------------|-----------------------------------|-------------| | `LINKUP_API_KEY` | API key for LinkUp search service | (required) | All other configuration values are defined in code and do not require manual setup. --- ## Project Structure ```text root/ ├── backend/ │ ├── agents.py # Agent definitions, tool wrappers, and pipeline orchestration │ ├── app.py # Streamlit UI application │ └── server.py # FastMCP RPC server endpoint ├── LICENSE ├── README.md └── requirements.txt # Python dependencies ``` --- ## Contributing Contributions, issues, and feature requests are welcome! Please open an issue or pull request on the GitHub repository. --- ## Future Implementations - **Full Async & Reactive Pipeline** Convert all research steps (web search, analysis, writing) to non-blocking async operations with event-driven scheduling and responsive I/O for improved throughput. - **Enhanced Error Handling & Retry Mechanisms** Implement granular retry strategies, circuit-breaker patterns, backoff policies, and comprehensive structured logging to handle transient and persistent failures at each step. - **Memory Agent** Introduce a dedicated agent to capture and persist conversational context and research state into the LLM’s memory store for long-term continuity through the agents' sequential run. - **Feedback Agent** Add an agent to validate and refine outputs from each agent in parallel by comparing raw results against quality criteria and user feedback loops before final drafting. - **Cleaner & Customizable Outputs** Introduce configurable Markdown templates, support for HTML/PDF exports, and user-defined formatting options for polished, presentation-ready reports. --- ## License This project is licensed under the MIT License. See `LICENSE` for details.