Build Your Own AI Search Engine: Fireplexity v2 with Firecrawl and Groq
You've probably used Perplexity or similar AI search tools that pull live web results and summarize them with an LLM. They're slick, but what if you want that same experience for yourself—with your own API keys, your own deployment, and your own customization? That's exactly what Fireplexity v2 gives you: a self-hosted AI search engine that combines web, news, and image search with a Groq-powered LLM to synthesize answers.
Fireplexity v2 is an open-source project from the team at Mendable (the folks behind Firecrawl) that stitches together two powerful services: Firecrawl for scraping and searching the web, and Groq for ultra-fast LLM inference. The result is a fully functional AI search engine you can run locally or deploy to Vercel in minutes.
What It Does
Fireplexity v2 is, at its core, an AI search engine with three search modalities: web, news, and images. You type a query, and the app fetches relevant results from the web, then uses a Groq-powered LLM to generate a synthesized answer based on those results.
The architecture is straightforward. Firecrawl handles the heavy lifting of crawling and extracting content from search results—it's a scraping API that turns web pages into clean, LLM-ready text. Groq provides the inference engine, and it's notably fast because it runs on custom hardware designed for LLM workloads. The app itself is a Next.js project (you'll see npm run dev in the setup), so you're getting a modern React frontend with API routes in the same codebase.
The demo GIF in the README shows the interface in action—a clean search box that returns both a generated answer and the source links, similar to what you'd see on Perplexity or Bing's AI mode. But the key difference is that all of this runs under your control, with your API keys and your configuration.
Why It's Cool
There are a few things about Fireplexity v2 that make it genuinely interesting beyond just being another AI demo.
It's a reference implementation for a real product category. If you've ever wondered how AI search engines actually work under the hood—how they decide what to scrape, how they feed results to an LLM, how they structure the response—this project shows you the whole pipeline in a few hundred lines of code. It's like having a senior engineer walk you through their architecture.
The tech stack is intentionally minimal. Two API keys, one npm install, and you're running. There's no vector database, no fine-tuned model, no complex orchestration layer. Firecrawl handles retrieval, Groq handles generation, and Next.js glues it together. That simplicity makes it an excellent learning tool and a solid starting point for your own experiments.
It's built on two genuinely impressive services. Firecrawl is quickly becoming the standard for turning web content into LLM-ready data—it handles JavaScript rendering, extracts clean markdown, and manages rate limits so you don't have to. Groq, meanwhile, is famous for its inference speed; their LPU hardware can run Llama 3 at thousands of tokens per second. Combining the two gives you an AI search experience that feels nearly instant.
The deployment path is frictionless. The one-click Vercel button in the README means you can go from nothing to a deployed AI search engine in about five minutes. For developers who want to iterate on the UI or add features, that's a huge advantage over trying to wire up these services from scratch.
How to Try It
Getting started is genuinely quick. First, clone the repository and install dependencies:
git clone https://github.com/mendableai/fireplexity.git
cd fireplexity
npm install
Next, you'll need to set up your environment variables. The README shows you exactly what to copy:
cp .env.example .env.local
Then add your API keys to .env.local:
FIRECRAWL_API_KEY=fc-your-api-key
GROQ_API_KEY=gsk_your-groq-api-key
You'll need to grab keys from Firecrawl and Groq—both offer free tiers, so you can experiment without spending anything.
Finally, run it locally:
npm run dev
Then open http://localhost:3000 and start searching. If you want to deploy it for the world to see, the README includes a Vercel deploy button that handles the whole process with one click.
Final Thoughts
Fireplexity v2 is best suited for developers who want to understand how AI search works, or who want a solid foundation to build their own search-powered application. It's not trying to compete with Perplexity as a product—it's a starting point, a reference, and a proof that you can assemble a surprisingly capable AI search engine with just two well-chosen APIs and a bit of glue code.
The honest assessment: this is a simple project, and that's exactly its strength. It demonstrates the modern pattern of composing specialized services rather than building everything from scratch, and it gives you a working baseline you can extend in any direction—custom prompts, different models, additional search sources, whatever your use case demands. If you've been curious about how AI search tools are built, this is a great place to start poking around.
Follow @githubprojects for more developer tools and open source projects.