Why self-host your search?¶
February 17, 2026
Search is the one feature nobody notices until it is missing. Type, get results, move on. It is worth asking, occasionally, where that actually happens.
For most websites, it happens elsewhere. The content is sent to a third party, indexed on their infrastructure, and queried through their software. That arrangement is perfectly reasonable, and it has consequences worth naming.
What changes when you host it yourself¶
The content stays put¶
Self-hosted search means the index never leaves the infrastructure it came from. Nothing is copied to a third party, and no external service can be breached, discontinued or repriced out from under the site. For medical content, legal documents or internal wikis, that is often a requirement rather than a preference.
Digital sovereignty is the usual term: genuine control over the tools and data an organisation depends on. Search is a small part of that, but an unusually legible one — it is easy to see exactly what would be sent away and to whom.
The costs stop moving¶
Hosted search is typically priced per record, per query, or both. At small volumes that is a few euros a month. The number grows with the site, and it grows during exactly the traffic spikes that are otherwise good news. A self-hosted index costs whatever the server already costs.
Leaving stays possible¶
Vendor lock-in is what happens when migrating away becomes expensive enough that nobody proposes it. Open source self-hosted software inverts the default: the index is in a standard MySQL database that can be queried, exported or built upon directly, and the code can be modified or forked without asking anyone.
It is less work than it sounds¶
Self-hosting has a reputation for complexity, earned honestly by plenty of software. This particular case is a single Python application and a single database. No cluster, no orchestration, no distributed state.
When to choose something else¶
Millions of queries a day need infrastructure built to scale horizontally. Dead Simple Search runs on one server and is not that.
Typo tolerance, faceted search or semantic ranking are genuine features, and Meilisearch or Typesense implement them properly. MySQL full-text search does not.
No appetite for running a server is a legitimate position, and a managed service is the honest answer to it.
Picking the tool that fits is not a compromise. Dead Simple Search is built for modest requirements, and it is worth being clear about where those end.
The middle ground¶
The choice is not between a large cloud service and no search at all. There is room for something small and focused that covers the ordinary case without the overhead — powerful enough for most small and medium sites, simple enough to maintain, and open enough to change when requirements do.
The code is on Codeberg, and the getting started guide covers installation through to a first search.