Playwright persistent profile: what it fixes and breaks

A persistent profile solves a real problem: logging in once instead of every run. It also carries three traps, one of which silently switches off protection you believe you have, and none of them appear in the guides about keeping sessions alive.

This page covers what a profile actually stores, how it interacts with a seeded fingerprint, the permission that undoes your proxy work, why the age of a profile has to hold together, the upstream bugs that are not your fault, and what to do about all of it.

storageState or a persistent profile: what each one carries

Two mechanisms get conflated constantly and they are not interchangeable.

storageState captures cookies and local storage and replays them into a fresh context. It is portable, it is a file you can open and read, and it carries nothing else: no cache, no permission decisions, no extensions, no profile-level settings.

A persistent profile is the browser’s own user data directory. It carries all of the above and keeps accumulating: the cache, the permissions, the extensions, the site-specific settings, everything a real browser writes down about itself over time.

For “log in once and stay logged in”, storageState is usually enough and much easier to reason about. For “this identity should look like it has been used”, the profile is the one that does it, because it accumulates the things you would not think to fake.

When each is the right choice

A test suite that needs an authenticated session: storageState.
One long-lived identity you return to over weeks: a profile.
Many parallel workers: storageState per worker, or one profile per worker and never a shared directory.

The pairing rule: the profile stores history, the seed stores the machine

This is the trap specific to any tool that derives a fingerprint from a seed, and it is easy to hit because the two settings live in different places.

The profile says: this browser has been here before, it has cookies from weeks ago, it has a cache.

The seed says: this browser runs on a particular machine, with a particular GPU, screen, font set and audio device.

Pair a stable profile with a changing seed and you have described a session whose history spans weeks and whose hardware changed overnight. Nobody replaces a graphics card between two visits to the same site while keeping their cookies.

with InvisiblePlaywright(seed=42, profile_dir="/profiles/identity-42") as browser:
    ...

One directory, one seed, permanently. Write the pairing down, because the directory outlives your memory of which number produced it.

The inverse mistake is quieter: a stable seed with a fresh profile every run produces a machine that has been on the internet many times and never remembers anything, which is the same thing that makes a fresh browser score badly on reCAPTCHA v3.

The stored permission that disables WebRTC address protection

This one deserves its own section because it is invisible, permanent, and it undoes work you did somewhere else entirely.

Firefox conditions two separate WebRTC privacy behaviours on whether the page has camera or microphone access:

restricting ICE gathering to the default route rather than enumerating every local interface, and
masking the host candidate behind an mDNS .local name instead of the real LAN address.

Both are switched off together when a permission is present.

The critical detail is that the check counts a persisted permission, not only an active capture. Granting camera access once on an origin, in a directory you then reuse, turns those protections off for that origin in every future session on that profile, until the permission is removed.

So a liveness or verification flow that asks for a camera, or a grant_permissions call you made months ago while debugging, quietly undoes the WebRTC work described here. Nothing warns you; the session simply starts reporting more than it used to.

What to do about it

Audit the stored permissions on any profile used with a proxy. It takes a minute, and the failure mode is exactly the one the proxy was there to prevent.

Profile age has to be internally consistent

A directory created this morning and filled with cookies dated last year is telling two stories.

Things that carry an age and have to agree:

Cookie expiry dates, and the issue dates they imply.
Analytics identifiers that encode a first-seen timestamp.
The cache, which is either present and plausible or absent.
Local storage entries recording a first-visit date.
Whether consent decisions exist for sites the profile claims to have visited.

This matters most if you seed a profile rather than grow one. A grown profile is consistent for free. A written one is consistent only if you were careful, and a stale artefact is worse than a missing one, because absence has innocent explanations and an anachronism does not.

Upstream issues that are not your fault

Two open Playwright reports affect anyone doing this:

Session cookies not persisting through launch_persistent_context despite the documentation: microsoft/playwright#36139.
Persistent context failing to read cookies written by a previous session on the same directory, plus profile corruption: microsoft/playwright#35466.

If cookies vanish between runs, read those before rewriting your own code.

Add one more that is not a bug report: never run two browsers against the same profile directory at once. A user data directory has no concurrency story, and the failure is corruption rather than an error.

Conclusion

A persistent profile fixes history and nothing else. The machine underneath still answers for itself, so the GPU, the fonts, the screen and the audio device are unchanged by anything in that directory.

Used well it is one directory paired with one seed, with its permissions audited, never shared between concurrent runs, and grown rather than written. Used carelessly it is a way to accumulate contradictions that a fresh profile would not have had.

Short answers to the questions that lead here

Does a persistent profile stop me being detected? It fixes having no history. It does nothing about the machine: the GPU, fonts, screen and audio device come from where you run, not from the directory.

Should I use storageState or a persistent profile? storageState for logins, a profile when the identity should accumulate everything else.

Why do my cookies disappear between runs? Possibly one of the upstream issues above, possibly two processes sharing a directory, possibly a session cookie behaving as one.

Can I reuse one profile for many identities? No. The profile is the identity, and shared cookies link the identities that share them.

Do I need a different seed per profile? Yes, and the same seed every time for the same profile.

Why would a granted camera permission matter? Because Firefox conditions its WebRTC address protections on the absence of one, and a persisted grant counts.

See also: WebRTC leak with a proxy, which is what the permission trap undoes, reCAPTCHA v3 score, for why a profile with a past is worth keeping, and browser extensions are a fingerprint surface, since a profile is how you install one.

Sources

Playwright persistent context documentation and the two open issues linked above.
Firefox’s own conditioning of default_address_only and obfuscate_host_addresses on active-or-permitted capture, which is the mechanism behind the permission trap.
This project’s proxy and profile handling, described in the pages linked above.

From the notes of invisible_playwright, a Firefox patched at the C++ level. The permission trap is recorded in our own notes as a gap with no patch, which is why this page tells you to audit for it rather than promising it is handled.