ChatGPT
AI conversational assistant for answering questions, writing, and coding help
Farfalle
Self-hosted AI search engine with local and cloud LLMs
Farfalle combines multiple search APIs and LLMs—local via Ollama or cloud via OpenAI, Groq, LiteLLM—to deliver AI-enhanced answers, all deployable via Docker and Vercel.
Overview
Overview
Farfalle is a self‑hosted AI‑powered search platform that merges web search providers (SearXNG, Tavily, Serper, Bing) with large language models. Users can run local models through Ollama (llama3, gemma, mistral, phi3) or tap cloud services such as OpenAI GPT‑4o, Groq Llama‑3, and any custom endpoint via LiteLLM. The system answers queries with a blend of live web results and LLM reasoning, offering a richer, more contextual response than standard search.
Audience & Deployment
Designed for developers, teams, and privacy‑conscious users who need control over data and model costs. The stack—Next.js frontend, FastAPI backend, Redis rate limiting, and Logfire logging—runs in Docker containers, with a pre‑built image for quick start. After launching the backend, the frontend can be deployed on Vercel by pointing it to the API URL. Optional API keys enable additional search sources, but the core functionality works entirely offline with local models.
Highlights
Multi‑search provider integration (SearXNG, Tavily, Serper, Bing)
Support for local LLMs via Ollama and cloud models (OpenAI, Groq)
Custom LLM routing through LiteLLM
Agent‑driven search planning for higher relevance
Pros
- Runs locally for privacy‑focused users
- Flexible model selection (local, cloud, custom)
- Dockerized deployment simplifies setup
- Rich UI built with Next.js and shadcn/ui
Considerations
- Requires Docker and Ollama for local models
- Performance depends on chosen LLM and API limits
- File‑based chat feature is still in roadmap
- Configuration of multiple API keys can be complex
Managed products teams compare with
When teams consider Farfalle, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Developers wanting a self‑hosted AI search platform
- Teams needing control over model privacy and cost
- Projects that require integration with multiple search APIs
- Users who prefer a customizable UI built with Next.js
Not ideal when
- Non‑technical users uncomfortable with Docker setup
- Environments without internet access to fetch models
- Scenarios demanding instant zero‑config deployment
- Use cases needing built‑in document ingestion (not yet released)
How teams use it
Research assistant for developers
Provides concise answers sourced from web and LLM reasoning, accelerating code exploration.
Enterprise knowledge base search
Combines internal search APIs with private LLMs to answer employee queries while keeping data on‑prem.
Educational Q&A portal
Students ask questions and receive AI‑generated explanations backed by live web results.
Prototype AI chatbot
Leverages LiteLLM to route requests to preferred cloud provider, enabling rapid experimentation.
Tech snapshot
TypeScript69%
Python28%
Dockerfile2%
CSS1%
JavaScript1%
Mako1%
Frequently asked questions
Do I need an OpenAI API key to run Farfalle?
No, you can operate entirely with local models via Ollama; API keys are only required for cloud providers.
Which Docker command starts the application?
Run `docker-compose -f docker-compose.dev.yaml up -d` after configuring the .env file.
Can I add my own search provider?
Farfalle is built to integrate additional providers; you can extend the FastAPI backend to call custom APIs.
Is there a pre‑built Docker image?
Yes, a ready‑to‑use image is published as part of the repository’s releases.
What license governs Farfalle?
The project is released under the Apache‑2.0 license.
Project at a glance
Dormant- Stars
- 3,499
- Watchers
- 3,499
- Forks
- 323
LicenseApache-2.0
Repo age1 year old
Last commitlast year
Self-hostingSupported
Primary languageTypeScript
Last synced 2 days ago