Open WebUI: 92,252 fingerprints, 92,051 titles, and 7,087 hosts that kept the default port

Most products in this batch show a wide gap between their application fingerprint count and their page title count. Open WebUI does not. The two numbers converge within a few hundred of each other, and the convergence is worth more attention than the size of the figure.

The measurements

On 23 October 2026, ZoomEye returned 92,252 matches for app="Open WebUI", 92,051 matches for title="Open WebUI", and 7,087 matches for title="Open WebUI" && port="8080".
A convergence of that order says the fingerprint is derived from something in the page that also supplies the title. In practice that means the scanner recognises the product from its interface response rather than from a header or a banner that other software might reuse. The agreement is a sign of a clean fingerprint, not of a large vulnerable population.
The port number tells the opposite story. Only 7,087 of the 92,051 title matches answer on 8080, which is the default in the product's quick-start documentation. Roughly ninety-two percent of the population is elsewhere: behind a reverse proxy on 443, on a custom port, or on a host where the port mapping was changed during deployment.

What sits behind the login

Open WebUI is a chat interface in front of a model server. Its configuration holds upstream endpoints, and those endpoints frequently require an API key. It also stores conversation history, uploaded documents used for retrieval, user accounts, and often a database with all of the above inside it.
Two design details make that store more consequential than a normal web application's data. The upstream API keys are credentials with a billing relationship attached, so a leak has a financial dimension as well as a data one. And the retrieval feature means the application holds documents that users uploaded specifically because they wanted them searched, which is a category that tends to include internal material.

Why the default-port minority matters

A deployment that kept port 8080 is likely to have been installed from the quick-start rather than composed for production. Quick-start installations are the population where default administrator credentials, open registration, and a direct connection to a local model server are most common.
That is an inference about deployment maturity, not a claim about security. It is useful because it narrows the review to 7,087 hosts out of 92,051 without requiring any additional data.
The larger group behind proxies is not safer by definition. A proxy adds TLS and a hostname. It does not add authentication if the application's own login is the only gate, and it does not restrict registration if the application allows it.

Reading the evidence without overreach

State the facts: 92,252 fingerprint matches, 92,051 title matches, 7,087 title matches on port 8080, sub-type all, collected on 23 October 2026. None of those counts describes a version. Open WebUI ships frequently and its changelog is the authoritative record of what changed.
What the numbers support is a statement about adoption. This is a widely deployed interface for language models, and most of its deployments are fronted by a proxy on a standard port. An organisation that wants to know its own exposure does not need the global number. It needs to know whether its own instance allows self-registration, which model endpoints it can reach, and whether the accounts that hold those endpoint keys belong to the application or to a person.

Concrete hardening

Disable open registration unless the instance is intentionally public. An interface that lets a new account in also lets that account use the model endpoints the administrator configured, which is a budget problem and a data problem at once.
Scope the upstream keys. A key that can call every model on a provider account should not be the key an application uses. Provider-side limits and separate keys per deployment contain the cost of a leak.
Then look at what the retrieval store contains. An uploaded document collection is a data set, and the question worth asking is whether it holds material that the application's user base is not cleared to read. That is a governance question the product cannot answer for you.

The closing measurement point

Converging fingerprints and titles are a gift to anyone building an inventory: the two independent methods agree, so the count is trustworthy. Diverging port counts are a gift of a different kind, because they separate the deployments that followed the documentation from the ones that went further. Both observations come from the same three queries, and neither of them required touching a single host.

References

• Open WebUI quick-start documentation, https://docs.openwebui.com/getting-started/quick-start/
• Open WebUI project repository, https://github.com/open-webui/open-webui
• ZoomEye AI query results collected 23 October 2026 for app="Open WebUI", title="Open WebUI", and title="Open WebUI" && port="8080", sub-type all