Farfalle
Open-source AI search engine you self-host with local or cloud LLMs, using SearXNG or paid search APIs.
Farfalle is a Perplexity-style search engine with a Next.js front end and FastAPI back end.
Farfalle is an open-source AI search engine, described by its own README as a Perplexity clone. It runs local models through Ollama, any model through LiteLLM, or cloud models from OpenAI and Groq. It suits hobbyists and developers who want a simple self-run search app rather than a polished product.
What it is
You start the app with Docker Compose, then ask questions in the browser. It searches through SearXNG, Tavily, Serper or Bing and answers with the model you choose. An expert-search mode uses an agent that plans and executes the search, and chat history is supported.
Who it's for
It suits developers and tinkerers who want a small, readable AI search codebase to run or modify. It does not suit anyone who needs a maintained, feature-rich product, since the README's roadmap leaves chat with local files unchecked and its model list names older models.
Key features
Local model support
Run llama3, mistral, gemma or phi3 through Ollama.
Custom LLMs via LiteLLM
Connect any model LiteLLM supports.
Multiple search providers
Search through SearXNG, Tavily, Serper or Bing.
Expert search agent
An agent plans and executes the search for better results on harder questions.
Chat history
Keeps your earlier searches available.
Browser search engine
Add http://localhost:3000/?q=%s as a custom search engine in your browser.
Up and running in 3 steps
Our honest verdict
What's great
- Ollama support means answers can be generated without sending prompts to a cloud model.
- SearXNG support removes the need for a paid search API.
- The Next.js plus FastAPI split is a small codebase that is easy to read and adapt.
- You can register it as your browser's default search engine with a single URL pattern.
Where it falls short
- You need Docker, and Ollama with a downloaded model if you want fully local answers, so you need a machine able to run those models.
- Roadmap item 'Chat with local files' is still unchecked, so you cannot ask questions about your own documents.
- The README lists models such as gpt-4o, gpt-3.5-turbo and llama3, and the project page shows no releases, so expect to update configuration yourself.
- Rate limiting uses Redis and logging uses Logfire, which adds moving parts if you deploy beyond a local trial.
Pricing & license
Farfalle is free under Apache-2.0 and its README shows no paid plan. The only cost is running it yourself, plus any cloud model or search API usage.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Does Farfalle need an API key?
Not if you use Ollama with a local model. The README marks all API keys (Tavily, Serper, OpenAI, Bing, Groq) as optional.
Is there a hosted demo?
The README links farfalle.dev as a live demo that uses cloud models only.
Can I deploy it publicly?
The README shows Render for the backend and Vercel for the frontend, with NEXT_PUBLIC_API_URL pointing at the backend.
