Choosing a password manager for development and production involves more than checking an “autofill” box. If your preview application and production admin panel use different credentials, decide where each login should be suggested. Then compare that requirement with the product’s documented matching behavior and the client you actually use.
“Exact host” and “exact URL” are different requirements. We ran an offline URL-parsing experiment to make that difference visible. We did not test either password manager, fill a form, or establish that a paid plan is better. The result is a small acceptance checklist to take into a product evaluation—not a product security ranking.
Start with the login you do not want suggested
Imagine a developer who maintains these fictional addresses:
- Production:
https://admin.example.test/login - Preview:
https://preview.example.test/login - Another service:
https://admin.example.test:8443/login
The requirement might be: “Suggest the production login on the first address, but not the other two.” Another team might intentionally share an identity provider across several hosts. Neither team should inherit the other’s requirement without checking.
Write down the intended destinations before changing a setting. Use invented hostnames in a shared worksheet. Do not export a vault or paste passwords, session URLs, recovery codes, or customer domains into a tool to do this exercise.
What the two products document
1Password’s autofill behavior guide describes a broad default, an exact-host option that includes the port, and a never-fill option. Multiple website fields combine. Its broad matching uses the Public Suffix List and additional rules; it is not equivalent to comparing the last two labels of a hostname. The guide also identifies a platform caveat: built-in login suggestions on Mac, iPhone, and iPad do not reflect specific-host settings; the documented 1Password selection flow is needed to see suggestions with those behaviors applied.
Bitwarden’s URI documentation distinguishes base-domain, host, exact, prefix, regular-expression, and never matching. Host matching includes a specified port. Exact matching is narrower than host matching, including differences such as the path or query string. The documentation warns that prefix and regular-expression choices can be dangerous when configured incorrectly.
These are documentation findings, not results from installed clients. Do not translate either vendor’s option label into the other vendor’s behavior without checking. We have not verified the availability of every option in every plan, platform, or organizational policy.
Our offline experiment: separate host, origin, and URL
We used Node.js v25.8.0’s WHATWG URL implementation on November 10, 2026. Node documents host, origin, and href as distinct properties. The experiment compares those parsed properties with one reference address. It does not implement either vendor’s matcher, the Public Suffix List, related-site exceptions, redirects, or a phishing detector.
Reference: https://admin.example.test/login.
| Candidate address | Same parsed host | Same origin | Same normalized URL |
|---|---|---|---|
| Reference address | Yes | Yes | Yes |
https://preview.example.test/login | No | No | No |
https://admin.example.test:8443/login | No | No | No |
http://admin.example.test/login | Yes | No | No |
https://admin.example.test/billing | Yes | Yes | No |
https://admin.example.test/login?next=%2Fbilling | Yes | Yes | No |
https://admin.example.test.other.test/login | No | No | No |
https://[email protected]/login | No | No | No |
https://admin.example.test:443/login | Yes | Yes | Yes |
The HTTP row is a useful requirements check: equal hosts alone do not express “HTTPS only.” It does not show that either product silently fills an HTTP page; product warnings and filling rules were not exercised.
The path and query rows expose the opposite tradeoff. A full-address requirement can separate pages on one host, but you must decide what should happen when a legitimate login adds a return-path query. “Stricter” is not enough of a specification.
The last row shows normalization: this Node parser removes the explicit default HTTPS port. A raw string comparison would differ. Again, do not assume that the vendor’s exact mode normalizes inputs in precisely the same way.
A familiar-looking prefix is not a host check
The suffix and user-information examples both start with the literal string https://admin.example.test. Yet their parsed hosts are admin.example.test.other.test and other.test. A script using only startsWith() would accept those prefixes.
This is an observation about our deliberately naive string check, not a discovered vulnerability in 1Password or Bitwarden. It explains why a home-grown URL check or a loosely specified prefix requirement is a poor substitute for understanding the destination.
Reproduce the result without opening any address
Download the standalone Node worksheet, inspect it, then run it locally with an existing Node installation:
node autofill-url-boundaries-20261110.mjs
The script contains nine fixed synthetic addresses, uses built-in assertions, and prints JSON. It makes no network requests, creates no sockets, reads no vaults, and does not need a browser extension or service account. A failed assertion stops execution; passed: true means the listed parser expectations held, not that a password manager passed a security test.
A minimal part of the experiment is:
const saved = new URL('https://admin.example.test/login');
const candidate = new URL('http://admin.example.test/login');
console.log(candidate.host === saved.host); // true
console.log(candidate.origin === saved.origin); // false
Do not use the worksheet as an authorization filter for production requests. It deliberately omits many concerns that such a filter would require.
Turn the observations into a buying decision
Before paying or switching, prepare an acceptance row for each destination you actually need. Keep the expected result separate from the observed result:
| Requirement | Evidence to collect in an authorized evaluation |
|---|---|
| Production login absent from preview suggestions | Exact client/version, item settings, and observed suggestions on each host |
| Separate services on one hostname | Required port or path boundary and the product’s behavior there |
| Legitimate return URLs still work | Expected handling of changing query strings |
| Desktop and mobile workflows agree | Results for the actual extension or operating-system suggestion flow, not just one desktop screenshot |
| Shared login has more than one permitted destination | All website fields and their combined effect |
Start with dummy credentials and systems you are authorized to test. Record unknown results as unknown. A documentation match is a reason to test an option, not evidence that your configured workflow works.
If your existing tool meets the required destinations and platforms, this exercise supplies no reason to replace it. If a gap remains, compare the candidate’s exact client behavior, plan coverage, and administrative requirements before purchasing. Do not infer a free product’s limitations—or a paid product’s superiority—from these parser results.
For a team purchase, autofill is only one requirement. Our secrets-management offboarding checklist covers access removal, copied credentials, and operational responsibilities, with links to product and pricing information. Keep that evaluation separate from where a browser suggests a login. Neither browser matching nor a successful local worksheet proves that a former teammate’s copied credential has been revoked.
For projects using our AI coding rules gallery, a useful team instruction is to keep dummy login fixtures separate from real credentials and record the exact environment under test. A rule file itself does not enforce a password manager’s matching policy.