Repolode
Submit
felladrin

felladrin/MiniSearch

A self-hosted metasearch engine with an AI assistant that can run entirely inside the browser, with no API key required.

Open-source alternative to:Perplexity

View on GitHubWebsite
Stars
608
Forks
76
Open issues
0
Last commit
22 hours ago

What is it?

MiniSearch is a self-hosted search engine that combines web results with an AI assistant that writes cited answers. Web results are aggregated by a bundled SearXNG instance, reranked locally by a cross-encoder model, and cached, all packaged as a single Docker container.

The AI can run inside the browser tab on WebGPU or CPU using curated models from about 135M to 4B parameters, so no API key, separate inference server, or third-party access to queries is needed. Users can alternatively connect an OpenAI-compatible API such as Ollama, LM Studio, vLLM, or llama.cpp, or let the server proxy a private API key without exposing it to clients.

Search history, cached results, and chats are stored in the browser's IndexedDB and are not sent to a server. The project can be set as a default browser search engine, triggered from Raycast, or embedded in other pages.

Key features

  • 01Runs curated language models in the browser on WebGPU or CPU, with no API key needed.
  • 02Aggregates text and image results through a bundled SearXNG instance.
  • 03Reranks search results locally with a cross-encoder model.
  • 04Connects to OpenAI-compatible APIs, AI Horde, or a server-side proxy for your own key.
  • 05Provides cited answers, follow-up questions with conversation memory, and read-aloud output.
  • 06Stores history, cache, and chats locally in the browser, with no accounts or telemetry.

Use cases

  • Running a private search engine for yourself or a small group with a local model.
  • Pairing the search UI with a local Ollama or llama.cpp server for cited answers.
  • Hosting a shared instance that uses your own API key without exposing it to users.
  • Setting it as the default search engine in a browser or launcher.

Quick start

  1. 1

    Run the published Docker image

    docker run -p 7860:7860 ghcr.io/felladrin/minisearch
  2. 2

    Open the instance in your browser and start searching

Things to know

  • Running the AI fully in the browser depends on WebGPU or CPU performance, and models must be downloaded first.

In collections

Self-hosted appsPrivacy & local-first tools

Topics

#web-search#metasearch#rag#llm#self-hosted#webgpu#searxng#privacy#ai#ai-search-engine

FAQ

What is MiniSearch?
A self-hosted metasearch engine with an AI assistant that can run entirely inside the browser, with no API key required. MiniSearch is a self-hosted search engine that combines web results with an AI assistant that writes cited answers. Web results are aggregated by a bundled SearXNG instance, reranked locally by a cross-encoder model, and cached, all packaged as a single Docker container.
Is MiniSearch free to use? What license does it have?
MiniSearch is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is MiniSearch still maintained?
Actively maintained. The last commit to MiniSearch was on Nov 6, 2026.
What can I use MiniSearch for?
Running a private search engine for yourself or a small group with a local model. Pairing the search UI with a local Ollama or llama.cpp server for cited answers. Hosting a shared instance that uses your own API key without exposing it to users. Setting it as the default search engine in a browser or launcher.
How do I install MiniSearch?
Run the published Docker image: docker run -p 7860:7860 ghcr.io/felladrin/minisearch Open the instance in your browser and start searching
Where can I get MiniSearch?
The source code and installation instructions are on GitHub: https://github.com/felladrin/MiniSearch

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects