Firesearch is an AI-powered deep research tool designed for users needing comprehensive, cited web research. It automates the process of breaking down complex queries, finding relevant sources, extracting content, and synthesizing answers with citations, significantly reducing manual research effort.

How It Works

Firesearch employs a sophisticated workflow orchestrated by LangGraph. It begins by decomposing a user's query into smaller, manageable sub-questions. These sub-questions are then fed into the Firecrawl API's /search endpoint, which not only finds relevant URLs but also extracts their content directly into Markdown format. The extracted content is validated for relevance, and if insufficient, the system automatically retries with alternative search strategies and keywords. Finally, GPT-4o synthesizes the validated information into a coherent, cited answer.

Quick Start & Requirements

• Install: Clone the repository, create a .env.local file with FIRECRAWL_API_KEY and OPENAI_API_KEY, then run npm install or yarn install.
• Run: Execute npm run dev or yarn dev.
• Prerequisites: OpenAI API key, Firecrawl API key.
• Docs: firecrawl.dev/app/api-keys

Highlighted Details

• AI-driven research: Utilizes GPT-4o for query decomposition, strategy, and synthesis.
• Integrated scraping: Leverages Firecrawl's /search endpoint for simultaneous search and Markdown content extraction.
• Answer validation: Scores sources for relevance (0.7+ confidence) before synthesis.
• Automated retries: Employs multiple search strategies to overcome insufficient or irrelevant results.
• Full citations: Provides links to sources for every piece of information in the synthesized answer.

Maintenance & Community

The project is associated with Firecrawl. Further community or roadmap details are not explicitly provided in the README.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license allows for commercial use and integration with closed-source projects.

Limitations & Caveats

The effectiveness of the research is dependent on the quality and availability of data accessible via the Firecrawl API. The configuration includes parameters like MIN_CONTENT_LENGTH and MIN_ANSWER_CONFIDENCE which may require tuning for specific research tasks.