Key Takeaways
- This is the companion review to Farfalle's entry in the Local LLM Software Directory
- Free and open source under Apache 2.0, verified 2026-09-18
- Installs via Docker Compose; runs locally at
http://localhost:3000once set up - LLM backends: fully local via Ollama (Llama 3, Mistral, Gemma, Phi-3), or cloud via OpenAI, Groq, or custom models through LiteLLM
- Search providers: SearXNG, Tavily, Serper, and Bing
- Tech stack: Next.js frontend, FastAPI backend, Redis for rate limiting, per the repository
- 3,541 GitHub stars as of this review (2026-09-18)
- A hosted demo using cloud models only is available at farfalle.dev without self-hosting
📍 In One Sentence
Farfalle is a free, open-source (Apache 2.0), self-hosted AI search engine modeled on Perplexity that pairs a search provider with an LLM — including fully local models via Ollama — to answer questions with cited sources.
💬 In Plain Terms
Instead of asking Perplexity or a similar hosted service to search the web and summarize the results, Farfalle lets you run that same kind of tool yourself, on your own server, and optionally point it at a local model so your queries and the model's reasoning never leave your infrastructure.
📌Note: This review is based on Farfalle's own GitHub README and repository metadata. It does not claim PromptQuorum has run hands-on benchmarks of Farfalle's answer quality against Perplexity or other hosted search engines.
What Is Farfalle?
Farfalle is a self-hosted, open-source search application that answers a query by retrieving results from a search provider and having an LLM synthesize a cited answer, in the style of Perplexity. Because you run it yourself, you choose which search provider and which LLM (local or cloud) power each search.
- Product type: a self-hosted web app (with an underlying API) — not a hosted SaaS product, though a limited cloud-model demo exists at farfalle.dev
- Creator/repository: github.com/rashadphz/farfalle
- License: Apache 2.0, per the repository
- Locality: hybrid — it can run entirely offline against a local LLM and a self-hosted SearXNG instance, or use cloud search/LLM providers, or mix both
- Scale: 3,541 GitHub stars as of this review (2026-09-18)
- Tech stack: Next.js frontend, FastAPI backend, Redis for rate limiting, shadcn/ui components, per the repository
What Does Farfalle Actually Do?
Farfalle takes a question, retrieves results from your chosen search provider, and has an LLM read those results and write a cited answer — an agent-style planning step decides how to run the search before the answer is generated.
- Search step: queries one of four supported providers — SearXNG (self-hostable, no API key needed), Tavily, Serper, or Bing
- Answer step: an LLM synthesizes a response from the retrieved results, citing sources, in a Perplexity-style format
- Local LLM support: connects to Ollama for fully local inference, with Llama 3, Mistral, Gemma, and Phi-3 named in the repository as supported local models
- Cloud LLM support: OpenAI models (GPT-4o, GPT-3.5-turbo) and Groq-hosted Llama 3, plus other providers routed through LiteLLM
- Agentic search: the README describes an agent that plans and executes the search rather than a single fixed retrieval step
- Browser integration: Farfalle can be set as a default search engine in a browser, per the repository
- Roadmap items: the repository lists chat history and local file search as planned, not-yet-shipped features as of this review — verify current status on the repository before relying on either
Usage Examples: Three Ways to Use Farfalle
These are concrete workflows built from Farfalle's documented setup, not hypothetical scenarios.
Run a fully local, self-hosted search stack
Pair Farfalle with a self-hosted SearXNG instance for retrieval and Ollama running a local model (for example Llama 3 or Mistral) for the answer step. Configured this way, a search query and its answer never leave your own infrastructure.
Use Farfalle as a private, cited research assistant
Connect Tavily or Serper for higher-quality web retrieval while keeping the answering LLM local via Ollama, giving you Perplexity-style cited answers without sending your queries to a hosted AI search product.
Try it hosted before self-hosting
Visit the cloud-model-only demo at farfalle.dev to see Farfalle's answer format and citation style before committing to a Docker Compose deployment with your own search and LLM provider choices.
Install Farfalle
Farfalle installs free via Docker Compose — its source code is on GitHub.
Source | Link |
|---|---|
| GitHub repository (source code, Apache 2.0) | github.com/rashadphz/farfalle |
| Clone the repository | git clone https://github.com/rashadphz/farfalle.git |
| Copy the environment template | cp .env-template .env |
| Start via Docker Compose | docker-compose -f docker-compose.dev.yaml up -d |
| Hosted demo (cloud models only) | farfalle.dev |
After startup the app is available at http://localhost:3000. Optional API keys for search or cloud LLM providers go in the .env file — they are not required if you run entirely against a local Ollama model and a self-hosted SearXNG instance. Verify the current setup steps on the GitHub repository before running them, since instructions can change between releases.
Farfalle Pricing and Licensing
Farfalle itself is free with no paid tier. It is licensed Apache 2.0, per the repository — a permissive license that allows commercial use, modification, and redistribution.
- No subscription, no paid tier, no usage limits imposed by Farfalle itself
- No account or sign-up required to self-host it
- Running it fully locally (self-hosted SearXNG plus a local Ollama model) has no per-query cost beyond your own hardware and electricity
- If you connect Tavily, Serper, Bing, OpenAI, or Groq instead, those providers' own pricing applies — Farfalle does not add a markup
- The hosted demo at farfalle.dev is a separate, limited, cloud-models-only instance, not a paid product tier
Farfalle vs. Local Deep Research
Farfalle and Local Deep Research both combine web search with an LLM, but they target different jobs: Farfalle is built for fast, cited, single-turn search answers, while Local Deep Research is built for longer, multi-step research reports.
Primary use case
- Farfalle:
- Fast, Perplexity-style cited Q&A
- Local Deep Research:
- Multi-step deep-research reports
Interface
- Farfalle:
- Web app + CLI
- Local Deep Research:
- Web app + CLI + Python API
Local LLM support
- Farfalle:
- Via Ollama
- Local Deep Research:
- Via Ollama and other local backends
Search providers
- Farfalle:
- SearXNG, Tavily, Serper, Bing
- Local Deep Research:
- Multiple, including SearXNG
Install method
- Farfalle:
- Docker Compose
- Local Deep Research:
- pip or Docker
If you want a quick, single-answer search replacement, Farfalle's scope is the closer fit. If you need a longer research report synthesized across many sources, review Local Deep Research directly. Both are free and open-source; verify current feature details on each project's own repository before deciding.
Who Should Use Farfalle?
Farfalle fits people who want a Perplexity-style search experience they control end-to-end, including the option to keep the whole stack local.
Use Farfalle if you
- Want a self-hosted, open-source alternative to a hosted AI search product, so you control the search provider and the model
- Want the option to run the answering LLM fully locally via Ollama for privacy or cost reasons
- Are comfortable running a Docker Compose stack and configuring a
.envfile - Want cited, source-linked answers rather than a plain chat response with no references
- Want to set a self-hosted search engine as your browser's default search
Look elsewhere if you
- Want a zero-setup, fully hosted product — the farfalle.dev demo is cloud-models-only and not positioned as a full product tier
- Need chat history or local file search today — the repository lists both as roadmap items, not shipped features, as of this review
- Want a deep, multi-step research report rather than a fast single-turn answer — see Local Deep Research instead
- Need vendor support with an SLA — this review found no evidence of commercial backing or a support contract behind the project
What Farfalle Is Not Good For
Farfalle is not a good fit if you want a zero-setup hosted product, chat history today, or deep multi-step research reports.
- Not zero-setup — self-hosting requires Docker Compose and, for a fully local stack, your own SearXNG and Ollama setup
- Not a chat-history tool yet — the repository lists this as a roadmap item, not a shipped feature, as of this review
- Not a local-file-search tool yet — also listed as roadmap, not shipped, as of this review
- Not fully local unless you choose it that way — Tavily, Serper, and Bing are cloud search providers; only SearXNG can be self-hosted alongside a local Ollama model for an entirely offline stack
- Not backed by a funding round or company that this review could verify — treat it as an independently maintained, community-supported project
Common Mistakes When Evaluating Farfalle
Most confusion about Farfalle comes from assuming it is automatically fully private, or expecting features still on the roadmap.
Mistake 1: Assuming Farfalle is fully local by default
A default setup often uses cloud search providers or cloud LLMs. Fully local requires deliberately choosing Ollama for the LLM and self-hosting SearXNG for search — Tavily, Serper, and Bing remain cloud services regardless of which LLM you connect.
Mistake 2: Expecting chat history or local file search out of the box
The repository lists both as roadmap items rather than shipped features as of this review — check the repository's current README before assuming either is available.
Mistake 3: Confusing the farfalle.dev demo with a full self-hosted deployment
The hosted demo at farfalle.dev uses cloud models only and is a limited preview, not a substitute for reviewing the self-hosting setup if local LLM support matters to you.
Mistake 4: Expecting the same feature set as a deep-research tool
Farfalle is built for fast, single-turn, cited search answers, not multi-step research synthesis. For longer research reports, compare it against Local Deep Research rather than expecting Farfalle to do that job.
Competitors and Alternatives
Farfalle is most often compared to other self-hosted local-search and research tools such as Local Deep Research, Onyx, and Hyperlink — its main differentiator is a fast, Perplexity-style cited answer format with a simple Docker Compose setup.
Tool | Best known for | Link |
|---|---|---|
| Local Deep Research | Multi-step, deep-research report generation over local and web sources | Local Deep Research review |
| Onyx | Self-hosted enterprise search and RAG platform (formerly Danswer) | Onyx review |
| Hyperlink (Nexa AI) | Local search and personal-assistant app over your own files | Hyperlink review |
This list reflects tools commonly compared to Farfalle, not an independent PromptQuorum ranking — verify each tool's current feature set before choosing.
Frequently Asked Questions
What is Farfalle?
Farfalle (github.com/rashadphz/farfalle) is a free, open-source (Apache 2.0), self-hosted AI search engine, modeled on Perplexity, that pairs a search provider with an LLM to produce cited answers.
Is Farfalle free?
Yes. Farfalle is free and open source under Apache 2.0, with no paid tier. If you connect a paid search or LLM provider (Tavily, Serper, Bing, OpenAI, Groq), that provider's own pricing applies — Farfalle itself does not charge anything.
Can Farfalle run fully offline with local models?
Yes, if you choose it that way: connect Ollama for a fully local LLM and self-host SearXNG for search. Tavily, Serper, and Bing remain cloud services, so avoid those if you want a fully offline stack.
How do I install Farfalle?
Clone the repository, copy .env-template to .env, and run docker-compose -f docker-compose.dev.yaml up -d. The app is then available at http://localhost:3000. Verify current setup steps on the GitHub repository before running them.
Which local LLMs does Farfalle support?
Through Ollama, the repository names Llama 3, Mistral, Gemma, and Phi-3 as supported local models.
Which search providers does Farfalle support?
Four: SearXNG (self-hostable, no API key needed), Tavily, Serper, and Bing.
Does Farfalle have a hosted version I can try first?
Yes, a limited demo using cloud models only runs at farfalle.dev, separate from the self-hosted Docker Compose deployment.
Does Farfalle support chat history or searching my own local files?
Not yet, as of this review — the repository lists both as roadmap items rather than shipped features. Check the current README for updated status.
What license does Farfalle use?
Apache 2.0, per the repository — a permissive license that allows commercial use, modification, and redistribution.
How is Farfalle different from Local Deep Research?
Farfalle is built for fast, single-turn, Perplexity-style cited answers, while Local Deep Research is built for longer, multi-step research reports. See the dedicated comparison section above for details.