Quick Answer: PiBrowse is a free, open source fingerprint browser for keeping browser profiles separate. You can run it locally with no account or server, or self-host it for a team. The open-source edition stays free to download and use. A paid hosted service, PiBrowse Cloud, is still in development.

Most fingerprint browsers charge a monthly fee and keep their code closed, so you're asked to trust a spoofing layer you can't inspect. PiBrowse, an open source fingerprint browser from developer Aron, takes the opposite bet. Its browser core is a Chromium 157 fork you can build and read, and it runs solo on your own machine with no account. If you manage several accounts or client profiles, that difference matters more than the feature list.

What a fingerprint browser actually changes

A fingerprint browser presents a consistent set of browser signals so each profile looks like a different machine. Those signals include the user agent, navigator.platform, the WebGL renderer, screen dimensions, timezone and Client Hints. The hard part isn't changing any one value. It's making all of them agree.

Here's what a profile using a Windows 11 / RTX 4060 persona reports when launched on an Apple Silicon MacBook, according to the PiBrowse README: navigator.platform reads Win32, the user agent is Windows NT 10.0 on Chrome 157, the WebGL renderer names an NVIDIA GeForce RTX 4060, the screen is 1920×1080 with 1032 px available (the taskbar is subtracted), Client Hints say Windows, the timezone is Europe/Berlin, and navigator.webdriver is false. The README says these values hold in the main frame, in a Worker and in three kinds of iframe.

That cross-context consistency is where most spoofing fails. A mismatch between execution contexts takes only a few lines of JavaScript to detect. PiBrowse makes the spoof in C++ inside the Chromium fork rather than injecting JavaScript, because injected scripts leave traces that page code can probe.

The counter-intuitive part is noise. Many people assume a fingerprint should change on every read, since that sounds more private. PiBrowse does the reverse. Its noise is deterministic, so the same profile produces the same canvas hash every time. The README's reasoning is that a fingerprint which changes between calls describes hardware that moves while you watch, which is a stronger tell than an honest machine.

That said, there's a real catch. Speech synthesis voices are a known gap. On a Mac, a Windows persona still reports macOS voices, because the system has no Windows voices and they can't be invented. The practical rule is to use a persona that matches the OS of the machine you run it on.

Why the open source part is the real feature

Anti-detect browsers often cost $30–150 a month per team, according to the Fury project's own competitor comparison, and most are closed source. That means nobody outside the vendor can check what they spoof. Fury's team reports measuring one popular product where the canvas readback was byte-identical to a clean machine, even though the interface listed canvas as protected. Treat that as a claim from a competitor-focused project, not an independent audit, and run the measurement yourself.

Open source gives you three things you can verify directly:

  • Patches in the repo. The core is a Chromium 157 fork with 41 patches under core/patches/.
  • Documented vectors. Each spoofed signal is tied to the Chromium file it lives in.
  • A bundled bench. You can measure the output on your own hardware instead of trusting a screenshot.

Here's the trade-off, though. Open source doesn't mean secure by default. Reading the patches tells you what should happen, and only measurement tells you what does. Fury's own README also lists the things it hasn't finished, so check the current status before you rely on a feature.

Three ways to run it

The Fury documentation, which PiBrowse builds on, describes three operating modes. They differ mainly in who runs the server.

ModeWhat you hostBest forMain trade-off
SoloNothing: no account, no databaseOne person managing their own profilesNo sharing, roles or distributed locks
Team, self-hostedOne binary plus PostgreSQL on a VPSAn agency or team needing projects and access grantsYou patch, back up and secure the server
Team, hostedNothingTeams that don't want to run serversThe server is someone else's machine, though bundles are encrypted on the operator's side

Solo is the default, and the team layer adds nothing to the fingerprint itself. Self-hosting is what you turn on once a team is involved. The hosted option is Fury's service, with a prefilled address that a team can edit to point at its own server. PiBrowse Cloud, the paid service from this project, is still in development. Don't assume it's available yet.

Getting started on macOS (preview caveats)

The current PiBrowse release is version 0.5.5, a preview for macOS on Apple Silicon. Here's the install path, in order:

  1. Download the desktop app (.dmg) from the v0.5.5 release page.
  2. Download the Chromium 157.0.8103.39 core (.tar.xz). The app and the core are separate downloads, and the in-app Download the browser button fetches the core from the repository.
  3. Verify both files against SHA256SUMS.txt before you open anything. On macOS, shasum -a 256 -c SHA256SUMS.txt does this. A matching checksum proves the file wasn't corrupted or swapped in transit. It doesn't prove who built it, so decide how much you trust the release page itself.
  4. Open the app. This preview is ad-hoc signed with no Apple Developer ID and no notarization, so Gatekeeper may stop you. The install notes in docs/15-install.md cover the workaround.
  5. Expect manual updates. Nothing updates itself in this preview.

Windows and Intel Mac binaries aren't part of this release. The upstream Fury project says its Windows build works, but that doesn't make a PiBrowse Windows binary available.

Building from source (and how long it really takes)

If you'd rather build than download, budget time. The Fury build guide reports that the core took 2 h 42 min on an Apple M5 with 10 cores and 16 GB of RAM, using the macos-arm64-lowmem configuration. Its checkout and one build directory came to about 39 GB, measured. The fetch.sh script asks for 150 GB free, which is a cautious estimate rather than a measurement.

Incremental rebuilds usually take 5 to 30 minutes. Don't reach for ccache or sccache to speed this up. The build uses -fmodules, and according to the Fury docs those caches miss on nearly every file.

The non-core parts are quicker. Build the agent with cargo build --release, then the desktop shell with npm install && npm run app:build inside desktop/. The agent is still named fury-agent in its commands, which is a leftover from the upstream project and not a sign you have the wrong code.

Automation, the MCP server and the limits to know

The local API has six endpoints and uses one bearer token. It stays off until you set FURY_API_PORT. The examples/ folder has four working scripts: curl, Playwright, Puppeteer, and one that runs a job across every profile one at a time and closes browsers after a failure. That last pattern matters because a crashed run can otherwise leave dozens of browser windows open and keep their sessions alive.

PiBrowse also includes an MCP server. Open AI assistants in the sidebar and press Connect, and clients such as Claude Desktop, Cursor or Claude Code can create profiles and proxies, open them, navigate, read pages and click. Treat the bearer token like a password, because it controls every browser it can reach. Keep the API port off public networks.

Here's the short list of limits to plan around:

  • Platform: the 0.5.5 preview covers macOS Apple Silicon only.
  • Signing: ad-hoc signed, with no notarization and manual updates.
  • Voices: speech synthesis voices reflect the host OS, so match persona to machine.
  • Build cost: the core needs roughly three hours on a strong laptop and tens of gigabytes of disk.
  • Detection: no fingerprint is guaranteed to pass every site. Verify behavior with the bench before you rely on it.

Frequently Asked Questions

What is a fingerprint browser?

A fingerprint browser is a browser that presents a consistent, configurable set of signals, such as user agent, WebGL renderer, screen size, timezone and Client Hints, so each profile looks like a separate machine. It keeps accounts and sessions apart on one computer. PiBrowse does this inside a Chromium 157 fork rather than by injecting JavaScript.

Is an open source fingerprint browser free to use?

The open-source PiBrowse edition is free to download, use locally and self-host, according to the project README. PiBrowse Cloud, a separate paid hosting service for teams, is still in development. Check the current release notes before you assume a feature is available.

How do I install PiBrowse on a Mac?

Download the 0.5.5 .dmg and the Chromium 157 core .tar.xz from the release page. Verify both against SHA256SUMS.txt with shasum -a 256 -c, then open the app. Because this preview isn't notarized, follow the install notes in docs/15-install.md if macOS blocks it.

Why does a fingerprint need to stay the same between page loads?

Sites often read the same signal more than once and compare the results. If your canvas hash changes between calls, the page sees hardware that appears to change, which is a red flag. PiBrowse uses deterministic noise so each profile returns the same canvas hash every time.

Final takeaway

If you need an open source fingerprint browser, start with solo mode, install the 0.5.5 preview, and measure what it reports on your own machine instead of trusting the marketing. Match the persona to your OS, keep the automation token private, and move to a self-hosted server only when a team actually needs shared access. Try the bench this week and note what it reports on your own machine. Pass this to someone who's stuck paying a subscription for a browser they can't inspect.