Open source · Rust + Tokio · MIT

Query large language models over DNS.

A high-performance DNS server that answers TXT queries with AI-generated responses. No HTTP, no SDK, no special client — just dig. Works on every device, even where HTTP is blocked.

bash — dig over llm-over-dns
Try a query ↓
Why it's neat

A whole AI gateway hiding inside a DNS query.

DNS reaches everywhere and slips through almost any firewall. llm-over-dns turns that ubiquity into an LLM endpoint you can hit with tools you already have.

Live Demo

Try it live now. No setup needed.

A live Nameserver is running publicly at 178.18.253.241 mapped to llm-over-dns.duyet.net. Run these queries directly from your own terminal.

Option A: Query Server Directly

Query the live DNS server directly by pointing to its domain:

dig +short TXT "What is 15 + 30?" @llm-over-dns.duyet.net

Option B: NS Delegation (Magical)

Query from any machine globally without specifying the server address:

dig +short TXT "hello.llm-over-dns.duyet.net"
Quick start

Running in about two minutes.

Clone, drop in an API key, and run on a non-privileged port. Then ask anything from a second terminal.

Use cases

Where a DNS-shaped LLM actually shines.

How it works

From dig to answer in one round trip.

Your question travels as a TXT query. The server forwards it to an LLM gateway, then streams the answer back as DNS records — chunked to fit the 255-character limit.

On the way back: long answers are split into 255-char TXT records, with automatic model fallback if the primary model fails.
Documentation

Read deeper.

Full guides live in the repo's docs/ directory.