DeerFlow: FULLY FREE Local DEEP Research Agent - Powerful AI Agent Can Do Anything!

Key Concepts

• Deep Research Agent: An AI system capable of performing in-depth research tasks, including web searching, data analysis, and report generation.
• Gearflow: An open-source, local deep research framework.
• Multi-Agent Architecture: A system where multiple AI agents collaborate to achieve a common goal.
• Ollama: A tool for running large language models locally.
• Open Router: An API that provides access to various open-source language models.
• MCP (Modular Computation Platform): A platform that allows for the integration of various tools and plugins to enhance the agent's capabilities.
• UV, NVM, PMPM: Package managers and version managers for Python and Node.js.
• API Key: A code used to authenticate and authorize access to an API.
• Web UI: A graphical user interface accessible through a web browser.
• Console UI: A text-based user interface accessible through a command prompt.

Gearflow: A Local, Open-Source Deep Research Agent

Overview

The video introduces Gearflow, a community-driven, open-source deep research framework designed to replicate the capabilities of advanced deep research agents like Manis AI and Gen Spark, but with the advantage of being fully local and customizable. Gearflow combines large language models (LLMs) with specialized tools for web search, crawling, and Python execution.

Key Features and Capabilities

• Open Source and Local: Gearflow is fully open-source, allowing users to inspect, modify, and contribute to the code. It runs locally, eliminating the need for cloud-based services and ensuring data privacy.
• Multi-Agent Architecture: Gearflow employs a multi-agent architecture consisting of a coordinator, planner, research team (research agent and coder), and reporter.
  • Coordinator: Engages with the user prompt.
  • Planner: Uses AI agents to update the plan with human feedback.
  • Research Team: Develops and executes the plan, using a research agent and coder (if coding is needed).
  • Reporter: Compiles the research findings into a report.
• LLM Integration: Gearflow supports open-source models via Ollama (e.g., DeepSeek R1, Qwen), OpenAI-compatible APIs, and multi-tier LM systems for different task complexities.
• Tool and MCP Integrations: Includes tools for search, retrieval, RAG (Retrieval-Augmented Generation), scraping, and MCPs for enhanced capabilities through plugins.
• Human Collaboration: Supports human-in-the-loop workflows and report post-editing.
• Content Creation: Enables the creation of podcasts and presentations, similar to Manis AI.

Example Use Case

The video demonstrates Gearflow's capabilities by asking it to write a brief on the top trending repositories on GitHub. Gearflow quickly identifies Cortex AI Suna AI as a top trending repository, researches it, and provides:

• Key points
• Overview
• Detailed analysis
• Purpose
• Images of the repository
• Key citations

Installation Methods

Gearflow can be installed using two methods:

1. Source Code: Cloning the repository and installing dependencies using UV, npm, and nvm.
2. Docker: Using Docker Desktop for a simpler installation process.

The video focuses on the source code installation method due to past user difficulties with Docker.

Source Code Installation Steps

1. Prerequisites:
   • Python (latest version)
   • Node.js
   • UV
   • NVM
   • PMPM
   • Git
   • VS Code
2. Clone the Repository: Use the command git clone [repository URL] in the command prompt.
3. Navigate to the Directory: Use the command cd deer-flow.
4. Configure API Keys:
   • Open the env.example file in VS Code and rename it to .env.
   • Configure API keys for Tavily, Brave Search, Vulk Engine (optional), and Languageflow (optional).
   • If using DuckDuckGo, replace Tavily with DuckDuckGo and provide the corresponding API key.
5. Configure LLM:
   • Open the conf.yaml.example file in VS Code and rename it to conf.yaml.
   • Configure the large language model settings, including the base URL, model name, and API key.
   • Options include OpenAI-compatible models, Ollama models (local), and Open Router models (free models available).
   • The video suggests using DeepSeek and provides instructions for configuring it with the base URL, model (V3), and API key.
6. Install MARP Ply:
   • Windows: npm install -g @marp-team/marp-cli
   • macOS/Linux: brew install marp-cli
7. Web UI (Optional):
   • Navigate to the web directory.
   • Install dependencies: npm install
   • Run the backend and frontend servers: uv run web (or the appropriate command for your OS).
8. Run the Agent:
   • Navigate back to the deer-flow directory.
   • Use the bootstrap command to start the servers.

Testing the Agent

The video demonstrates testing the agent by asking it to compare the height of the Eiffel Tower to the tallest building in the world. The agent uses the Tavily API key to search the web, find relevant information, and provide an answer.

Conclusion

Gearflow offers a powerful, open-source, and local alternative to proprietary deep research agents. Its multi-agent architecture, LLM integration, and tool support make it a versatile tool for various research tasks. The video provides a detailed guide to installing and configuring Gearflow, enabling users to leverage its capabilities for free.