Farfalle
Farfalle is an open-source AI-powered answer engine. The principle functionality of Farfalle is to provide detailed and accurate responses to user queries. Root...
Last verified:
What is Farfalle?
Farfalle is an open-source AI-powered search engine and answer engine positioned as a Perplexity-style alternative. It lets users ask questions and get synthesized answers rather than just keyword search results, and it can be used as a live demo at farfalle.dev with cloud models.
The tool combines multiple search providers with large language models to gather information from external sources and turn it into concise answers. Its search flow is agent-based, meaning it can plan and execute search steps to improve result quality instead of relying on a single search query.
Farfalle is aimed at developers, self-hosting enthusiasts, and teams that want a private, customizable search experience. It is especially relevant for users who want to run local models on their own infrastructure, connect custom models through LiteLLM, or use cloud models when convenient.
The project emphasizes flexibility in deployment and model choice. The website and repository position it as a modern stack built around Next.js on the frontend, FastAPI on the backend, and support for several external search APIs and rate-limited backend services.
Farfalle pricing
Pricing model: Free
The website presents Farfalle as free and open-source, with a live demo available at farfalle.dev that currently uses cloud models only. For self-hosting, the project can run without API keys if you use local models through Ollama, but optional keys may be used for Tavily, Serper, OpenAI, Bing, and Groq. No paid plans are listed on the website.
Farfalle pros
- Open-source codebase
- Perplexity-style answer engine
- Supports multiple search providers
- Uses agent-based search planning
- Can run local LLMs
- Works with cloud LLMs
- Supports custom LLMs via LiteLLM
- Has a live web demo
- Designed for self-hosting
- No API keys required for local use
- Flexible deployment options
- Modern Next.js frontend
- FastAPI backend
- Can use several search APIs
- Built for concise synthesized answers
- Open-source licensing
- Supports browser as search engine
Farfalle cons
- Cloud demo only uses cloud models
- Local model setup requires Ollama
- Docker is needed for local startup
- External API keys may be needed for some providers
- Not all roadmap features are shipped yet
- No releases published
- Local file chat is only on roadmap
- Chat history is only on roadmap
- Docker deployment setup is still on roadmap
- Can depend on third-party search services
Frequently asked questions about Farfalle
What is Farfalle?
Farfalle is an open-source AI-powered search engine and answer engine. It is designed to take a question, search across multiple sources, and return a concise synthesized answer instead of a simple list of links.
How is Farfalle different from a normal search engine?
Instead of only matching keywords, Farfalle uses an agent-based workflow that plans search steps, queries multiple search providers, and then uses an LLM to assemble an answer. That makes it closer to an answer engine than a traditional search box.
Can Farfalle run locally?
Yes. The project supports local LLMs, and the repository says you can run it with models like llama3, mistral, gemma, and phi3 through Ollama. Local use does not require API keys if you are using those local models.
Does Farfalle support cloud models?
Yes. The project supports cloud models such as OpenAI gpt-4o, OpenAI gpt-3.5-turbo, and Groq/Llama3. The live demo on farfalle.dev is described as cloud-model-only.
What search providers does Farfalle use?
The project integrates multiple search providers, including Tavily, Searxng, Serper, and Bing. This gives it several ways to gather source material before generating an answer.
Can I use my own LLM with Farfalle?
Yes. Farfalle supports custom LLMs through LiteLLM, so it is not limited to the built-in local or cloud model options. That makes it easier to connect a preferred model stack to the search workflow.
What do I need to run Farfalle locally?
The repository lists Docker as a prerequisite, and Ollama if you want to run local models. It also suggests optional API keys for services like Tavily, Serper, OpenAI, Bing, and Groq, depending on the setup you choose.
How do I start using it quickly?
The repository shows a quick-start flow of cloning the repo, copying the environment template, setting any optional keys, and launching the development stack with docker-compose. After the app starts, it is available on a local web address.
Can I use Farfalle as my browser search engine?
Yes. The repository includes instructions for adding a local Farfalle URL such as http://localhost:3000/?q=%s as a custom browser search engine entry.
What features are planned but not yet complete?
The roadmap lists several upcoming items, including better local LLM support through Ollama, Docker deployment setup, searxng support to reduce external dependencies, a pre-built Docker image, chat history, expert search, and chat with local files.