Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
RAG & Document Chat

SearXNG Review: Self-Hosted Search Backend for Local LLMs

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

SearXNG (build 2026.10.24-ef0ec03da, updated on 24 October 2026) is a self-hosted metasearch engine that a local LLM front end can query for live web results. It is free under the AGPL-3.0 licence, contains no AI model of its own and runs as a container. Every statement here comes from the project's documentation, its repository and its Docker Hub tags.

Key Takeaways

  • Maintenance: a rolling release whose commits and container builds are as recent as 24 October 2026; the repository has no GitHub releases, so no stale-project warning applies.
  • Licence and cost: AGPL-3.0, free.
  • Role: not an AI app. It supplies search results; the model and the chat interface come from other software.
  • Setup: Docker or Podman plus edits to a YAML settings file, so a terminal is required.
  • Maker: a community project that began in mid-2021 as a fork of searx.

📍 In One Sentence

SearXNG is an open-source metasearch engine that you host yourself and that local LLM front ends can use as a web-search backend.

💬 In Plain Terms

It asks many search engines for results on your behalf and hands them back as a web page or as JSON, so your local model can read live web results through a service you host.

📌Note: This review is based on the documentation, the repository and the container tags, checked on 24 October 2026. PromptQuorum has not installed SearXNG or measured result quality, so the privacy statements are the project's own.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

SearXNG is a free, open-source metasearch engine that you run yourself, and local LLM front ends such as Open WebUI can call it as their web-search backend. This review covers it in that role only: it is not an AI app and contains no language model. It is based on the project documentation, the GitHub repository and the Docker Hub tag list, checked on 24 October 2026; PromptQuorum has not run it hands-on.

SearXNG Review: Self-Hosted Search Backend for Local LLMs

What Is SearXNG?

SearXNG is a free metasearch engine: it sends your query to many search services at once and merges the answers into one result list. The documentation describes it as search without being tracked, lists about 70 public instances on searx.space and says it began in mid-2021 as a fork of the searx project.

For local AI its value is the search API: a front end requests /search?q=...&format=json and gets structured results it can pass to a model. The software contains no model and no answer generation, which is why answer engines such as Vane and Farfalle, which can use it as their search step, add the model on top.

Get It

SearXNG is distributed as a container image and as install scripts; there is no desktop installer, so a terminal is required.

Option
How to get it
Container imagesearxng/searxng on Docker Hub, mirror ghcr.io/searxng/searxng
Docker Compose (recommended)Container installation guide: docker compose up -d
Websitedocs.searxng.org
Privacy policyAbout page with the privacy statements; no separate policy document found
Source codegithub.com/searxng/searxng, AGPL-3.0
Public instancessearx.space lists them; many disable the JSON format

Version reviewed: 2026.10.24-ef0ec03da, the newest Docker Hub tag on 24 October 2026. SearXNG is a rolling release whose version number is based on the git commit, and the repository has no GitHub releases. The documentation lists no hardware requirement and no host operating systems.

How to Get Started

These steps follow the project's container guide and the Open WebUI documentation for SearXNG; PromptQuorum has not run them.

  1. 1
    Start the container
    Why it matters: Download the compose file and `.env.example` from the repository, copy the second to `.env` and run `docker compose up -d`; the stack also starts a Valkey database and uses port 8080 by default.
  2. 2
    Enable the JSON format
    Why it matters: The default `search.formats` list in `settings.yml` contains only `html`; add `json` and restart. A request for a format that is not enabled returns 403 Forbidden.
  3. 3
    Test the API
    Why it matters: Request `/search?q=searxng&format=json` on your instance, for example with curl, and check that structured results come back.
  4. 4
    Connect the front end
    Why it matters: In Open WebUI, set the web-search engine to searxng and the query URL to your instance's `/search` address. That guide is a community contribution that the Open WebUI team does not officially support, and it notes that web search must be switched on per chat.

Data Flow and Privacy

The documentation says SearXNG does not store information about its users, does not generate a profile and never shares anything with a third party. Those are the project's statements; the About page does not describe what an instance operator can see, and its answer for readers who care is to run the software yourself.

On a self-hosted instance the operator is you, so the open question is the outbound side: the About page says queries are mixed with searches on other platforms, meaning the upstream engines still receive them. The default settings file also names DuckDuckGo as the autocomplete backend; leaving that setting blank turns it off.

📌Note: These are the project's declarations, not audit results. PromptQuorum has not inspected network traffic or logs, and which queries leave your machine depends on the engines you enable.

Who Should Use It

  • Open WebUI and similar front-end users who want web search for a local model from an instance they control.
  • Self-hosters already running Docker who are comfortable editing a YAML file.
  • Developers building agents who need a JSON search endpoint they can call from code.

What We Could Not Verify

  • Privacy in practice. The no-tracking and no-sharing statements are unaudited, and which upstream engines see your queries depends on the engines you leave enabled.
  • Result quality for LLMs. PromptQuorum did not compare SearXNG results with other search providers or measure whether a model answers better with them.
  • Resistance to blocking. The default settings suspend an engine for a period after a CAPTCHA or an access-denied reply (24 hours for a CAPTCHA), so engines can drop out of the results; how often that happens was not measured.
  • Hardware and operating systems. The documentation states no RAM or CPU requirement and lists no host operating systems; the platforms on the directory tile are inferred from the container route.

Competitors and Alternatives

Vane

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
Answer engine that adds LLM-written, cited answers; its directory entry describes a self-hosted SearXNG backend, so it builds on SearXNG instead of replacing it

Farfalle

Platforms:
Windows, macOS, Linux
Price / license:
Free / Apache-2.0
Key difference:
Perplexity-style search app whose search step can use SearXNG, Tavily, Serper or Bing, per its review
Articles about Farfalle (2)

Local Deep Research

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
Multi-step research agent across arXiv, PubMed, the web and your own documents; a research tool, not a search backend
Articles about Local Deep Research (4)

Web Search MCP

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
MCP server that gives local LLMs live web search and page content in tools such as LM Studio

Vane and Farfalle can use SearXNG as their search step rather than compete with it; confirm each app's current price, license and platforms on its own listing.

Compare in directory

Frequently Asked Questions

Can I use a public SearXNG instance instead of hosting one?

Often not: the API documentation notes that many public instances disable the JSON format and answer 403, and a public instance routes your queries through its operator, about whom the About page gives no details. The searx.space list is the place to look.

Does SearXNG need a GPU or a language model?

No. It contains no model and the documentation states no hardware requirement; any GPU is used by the separate LLM that reads the results.

How do I update it or pin a version?

Pull a new image and restart the stack; the guide advises reviewing changes to the compose template first. The compose file reads the image tag from a SEARXNG_VERSION variable that defaults to latest, so setting a dated tag pins a build.

Do I need the limiter on a personal instance?

Not necessarily: it is off by default and needs a Valkey database, which the recommended compose file already starts. The compose file publishes the port on all interfaces unless SEARXNG_HOST is set, so check that setting before exposing the instance beyond your machine.

Does it work over Tor or without JavaScript?

The documentation says it can be used over Tor and that JavaScript and cookies are optional; PromptQuorum tested neither.

Verdict

SearXNG is a long-running, actively maintained way to give a local LLM front end live web results from a service you host, at no cost. Against that, it is a container-and-YAML setup rather than an installer, its JSON output must be switched on by hand, queries still reach the upstream engines, and its privacy and quality claims could not be checked here. It suits people who already run Open WebUI or a similar tool and are comfortable with Docker; anyone who needs strict privacy guarantees should read the configuration and the engines in use rather than rely on the project's summary.

Sources

← Back to Power Local LLM

MCP