Aetheron

What does Aetheron do?

Aetheron's README calls it "an advanced, fully agentic AI search engine built to deliver real-time, highly accurate answers sourced directly from the web", inspired by Perplexity. A query goes through a short pipeline: a fast model rewrites the question into optimised search queries, Aetheron fans out to public SearXNG instances in parallel with a DuckDuckGo HTML Lite fallback, pulls the top three unique URLs through the Jina Reader API to get clean text, scores chunks against the question locally, and streams a markdown answer with inline citations from Groq's inference engine.

The stack is Next.js 14, Tailwind CSS and the Vercel AI SDK. The README says the whole thing runs without a vector database, scoring chunks "in milliseconds" against query keywords. Chat history stays in the browser's Local Storage; the README calls the project "100 percent local-first memory" on that point.

Key facts

• Licence: MIT, verified from the LICENSE file.
• Stack: Next.js 14, Tailwind CSS, Vercel AI SDK, served by Node.js 18 or newer.
• Needs a free Groq API key, per the README.
• Default answer model, per the README: meta-llama/llama-4-scout-17b-16e-instruct. Query rewriting runs on llama-3.1-8b-instant.
• Search: concurrent requests to public SearXNG instances with a DuckDuckGo HTML Lite fallback.
• Content extraction uses the Jina Reader API, which the README says "strips away cookie banners, complex navigation menus, and ads, leaving only pure, clean Markdown text."
• Top three URLs are deduped from an aggregated set of six.
• No vector database; chunk scoring runs locally against keywords.
• Chat history stays in the browser's Local Storage.

What does it replace, and where does it fall short?

Aetheron replaces Perplexity for a reader who wants a cited, streamed answer against the open web, without a paid subscription, and does not need a vector database or an enterprise-scale index. It is a candidate on the open-source alternatives to Perplexity page alongside Vane (formerly Perplexica) and Morphic.

Where it falls short: Aetheron depends on public SearXNG instances and the Jina Reader API, both of which are third-party services with their own rate limits and uptime. The Groq API is also required; there is no local-model path in the README. The chat history is in Local Storage, so it is tied to the browser and not shared across devices. For a self-contained Perplexity alternative that bundles SearXNG, Vane is closer to that pattern.

How does Aetheron run?

Clone the repository, run npm install, create a .env.local with your Groq API key and other variables named in the README, then start the Next.js server. The README's setup section ends with "Follow these steps to deploy Aetheron locally for testing or development."

The project has no hosted or paid version. It is a self-hosted web app.

Who is Aetheron for?

A developer who wants a Perplexity-style answer engine on a small Next.js codebase they can read, extend and deploy themselves, and is comfortable with Groq as the model provider. For a bundled search-and-answer stack, Vane is a nearby pick.

What limits does the README state?

• Groq API key is required.
• Search depends on third-party SearXNG instances with a DuckDuckGo fallback.
• Content extraction depends on the Jina Reader API.
• Only the top three URLs are read fully; the aggregator keeps the top six.
• Chat history stays in Local Storage.

Questions people ask

Is Aetheron open source?

Yes. The LICENSE file is MIT, verified on the date on this page. MIT is an OSI-approved licence with no added conditions, so you can read, change and redistribute the code.

Does Aetheron need an API key?

Yes. A free Groq API key is required. The README names Groq's models and endpoints as the backend; there is no local-model path. SearXNG and the Jina Reader API do not require keys for the described flow.

How does Aetheron compare with Perplexity?

Perplexity is a hosted answer engine with its own search index, models and ranking. Aetheron is an open-source Perplexity-style pipeline on a small Next.js codebase: it rewrites a question, calls SearXNG, scrapes with Jina Reader, scores chunks locally and streams an answer from Groq. The open-source alternatives to Perplexity page compares Aetheron with Vane and Morphic.

Does Aetheron use a vector database?

No. The README calls the design "vector-less" and says keyword chunk scoring runs locally in milliseconds. If you want vector search for a bigger corpus, pick a different project.

Where does my chat history live?

In the browser's Local Storage, per the README. Clearing site data clears the history, and switching browsers or devices means a fresh history.

Sources

• Aetheron README and MIT LICENSE: github.com/Smilin01/Aetheron, read 2026-10-19.