Two peas, one pod
Exa vs SearXNG
By Sophie Filer · checked against both servers on 27 Oct 2026
Exa is the less work option: a hosted search engine made for models, with page text in every result and a URL you can add in seconds. SearXNG keeps queries in your own hands: it searches through an instance you run, with no API key or per-query cost. Choose SearXNG if privacy or zero vendor matters more than setup time.
SearXNG
Web search through your own SearXNG instance: no vendor, no API key.
Side by side
| Compared | Exa | SearXNG |
|---|---|---|
| Made by | Exa Labs | Ihor Sokoliuk and contributors |
| Who runs the search | Exa, as a paid API | You, on your own SearXNG instance |
| API key | Yes, or sign in on the hosted URL | None |
| Page text in results | Yes | No; a separate tool reads each page |
| Cost | Exa credit | Your hosting only |
| Works offline or air-gapped | No | Against an internal instance, yes |
| Tools | 2 (2 only read) | 4 (4 only read) |
| Context its tools take | At least ~300 tokens | At least ~700 tokens |
| Runs on | Your computer or hosted | Your computer |
| You need | An Exa API key for the local server | A running SearXNG instance with JSON output enabled; Node.js 18 or newer |
| Our install check | Started in our test | Started in our test |
| Licence | – | MIT |
| GitHub stars | 5.2K | 1.3K |
| Last commit | 19 Oct 2026 | 26 Oct 2026 |
Stars and commit dates refresh daily; tool counts come from our own run of each server.
Pick Exa if
- You want good results with sources and no infrastructure
- You use ChatGPT or another client that only takes a URL
- You want code examples as well as web pages
Pick SearXNG if
- You do not want queries going to a search vendor
- You already run SearXNG, or are happy to start one in Docker
- You want search without per-query costs
Can you run both?
There is little reason to run both in one client, since the assistant would have two search tools to choose between. Use public SearXNG instances with care: whoever runs one can see every query you send.