Back to blogSolution

After (Fortress)

Fortress publishes a reproducible 92-site benchmark on real bot-protected sites where it loaded 86.4% of pages against Camoufox at 74.5% and puppeteer-extra stealth at 72.3%, run headless on datacenter IPs with no proxies.


Introduction and methodology note

Fortress is Tilion's open-source, source-available Chromium fork. It fixes browser fingerprints inside Chromium's C++ layer, so a Playwright or Puppeteer script connects over CDP to what reads as an ordinary Chrome install, with no stealth plugins or JavaScript patches. Camoufox is a separate open-source stealth browser built on Firefox. Both exist to stop scrapers and AI agents from getting blocked; they diverge on engine, on where the fingerprint fix lives, and on what each vendor has actually published.

The numbers on this page come from Fortress's own 92-site protected-site test, run on 28 September 2026 and documented with its harness in the Fortress repository under docs/BENCHMARK.md. The test ran fully headless on datacenter IPs with no proxies, two repetitions per site, one fresh cloud machine per run, 1,584 machines in total, against 92 sites protected by Cloudflare, DataDome, Akamai, HUMAN, Kasada, Imperva, AWS WAF, Arkose and reCAPTCHA. Product capabilities come from Fortress's own site. Camoufox figures are quoted from the same published test and stated as plain text. All pass-rate claims are self-reported by the vendor that ran the harness; test both engines on your own target sites before committing.

TL;DR

  • Fortress is the stealth browser with a published, reproducible benchmark against Camoufox and puppeteer-stealth on real protected sites: 86.4% pages served for Fortress, 74.5% for Camoufox, 72.3% for puppeteer-extra stealth, 34.2% for stock Chrome (28 September 2026, headless, datacenter IPs, no proxies).
  • nodriver does not appear in that published result set, so no head-to-head number against it exists; treat any claim comparing Fortress to nodriver on protected sites as unpublished.
  • Fortress is free for developers with no caps on machines, sessions or concurrency; production use by larger organizations needs a subscription, which is the main trade-off versus an open-source alternative.

Quick-comparison table

| Dimension | Fortress (Tilion) | Camoufox | |---|---|---| | Engine base | Chromium fork, v3 on Chromium 153.0.8010.36, updated monthly | Firefox-based | | Fingerprint fix layer | 81+ patches in Chromium's C++ | Not published on the pages reviewed | | Pages served, 92-site test (28 Sep 2026) | 86.4% (95% Wilson interval 81 to 91) | 74.5% | | Headless vs headed fingerprint | Same fingerprint in both modes, including on datacenter IPs | Not published on the pages reviewed | | Connection model | CDP (connect_over_cdp), works with Playwright and Puppeteer | Playwright-based | | Install | pip install tilion-fortress, npm install tilion-fortress, Docker (tilion/fortress), .deb, single binary | Not published on the pages reviewed | | Agent tooling | Local MCP server with 29 tools, free, no account | Not published on the pages reviewed | | License | Fortress Source Available License 1.1; free for individuals and smaller organizations, subscription for larger ones | Open source | | CAPTCHA/anti-bot checks passed (vendor-reported) | CreepJS, BrowserScan, rebrowser, WebGL; Akamai _abck where stock Chrome gets Access Denied | Not published on the pages reviewed |

Granular differentiator table

The differentiator that matters here is the published protected-site result. Every figure below is from Fortress's 92-site test of 28 September 2026 (headless, datacenter IPs, no proxies, two reps, fresh machine per run), documented in the repo's docs/BENCHMARK.md.

| Target | Fortress | Camoufox | puppeteer-extra stealth | Stock Chrome | |---|---|---|---|---| | Pages served across 92 protected sites | 86.4% | 74.5% | 72.3% | 34.2% | | Published in the same reproducible run | Yes | Yes | Yes | Yes | | Nodriver | Not published | Not published | Not published | Not published |

Read the table plainly: Camoufox at 74.5% beat puppeteer-extra stealth at 72.3%, so the gap between the two stealth browsers is about 12 points, not a blowout. The confidence interval on the Fortress figure (81 to 91) overlaps the top of what a Camoufox run could plausibly show, which is exactly why the harness and methodology are published for anyone to re-run. Nodriver is the honest gap: it is a commonly named stealth option, and this benchmark does not include it.

Why teams choose Fortress over Camoufox

#### The benchmark is published and reproducible

Most stealth tools ask for trust. Fortress publishes its harness: 92 sites, two reps, a fresh cloud machine per run, 1,584 machines, with the methodology in `docs/BENCHMARK.md`. The result, 86.4% versus Camoufox's 74.5%, carries a 95% Wilson interval of 81 to 91, and the repo tells you how to reproduce it. Camoufox's own protected-site numbers are not published on the pages reviewed for this comparison.

#### The fingerprint is fixed in C++, not patched in JavaScript

Fortress applies 81+ patches inside Chromium's C++ layer covering GPU, screen, fonts, timezone and more, so headless and headed modes present the same fingerprint, including on datacenter IPs. Each launch mints an internally coherent machine where GPU, screen, cores, timezone, language and fonts agree; Canvas and Audio fingerprints were distinct in 40 of 40 launches. Per the Fortress site, it passes CreepJS, BrowserScan, rebrowser and WebGL checks, and passes Akamai's _abck flow where stock Chrome gets Access Denied.

#### Existing Playwright and Puppeteer code keeps working

Because Fortress is a Chromium fork exposing raw CDP on port 9222, swapping in is a one-line connect_over_cdp call with no stealth plugin imports, no fingerprint patching and no Firefox-specific API surface. The Docker image gets you there with docker run --rm -p 9222:9222 tilion/fortress:latest.

If your scrapers are currently losing pages to Cloudflare or DataDome, join the Fortress waitlist and download the engine to run the published harness against your own site list.

Best use cases for Fortress

#### Scraping Akamai- and DataDome-protected sites

Fortress's vendor-reported results include passing Akamai's _abck cookie check, and the 92-site test covers Akamai, DataDome, HUMAN, Kasada and Imperva. For teams whose blocker is a commercial anti-bot vendor rather than a fingerprint checker, that site list is the relevant evidence.

How Camoufox compares: it is in the same published test at 74.5%, so it is a credible fallback, but its own per-vendor breakdown is not published on the pages reviewed.

#### AI agents that must read blocked pages

Fortress ships a local MCP server with 29 agent tools, free, no account required, letting Claude, Codex, Cursor, Windsurf and Cline read blocked pages, extract fields and interact with forms. Setup instructions live in the repo's AGENTS.md. It also integrates with browser-use, crawl4ai, Stagehand and LangChain.

How Camoufox compares: no equivalent agent tooling is published on the pages reviewed; you would wire your own agent loop over Playwright.

#### Teams that need headless to look like headed

Fortress's headless and headed modes present the same fingerprint, including on datacenter IPs, which matters when you run at scale in CI or on cloud machines and cannot afford residential proxies.

How Camoufox compares: its headless fingerprint behavior is not published on the pages reviewed.

#### Cost-sensitive scale-up on the engine itself

Fortress v3 is free for developers with no caps on machines, sessions or concurrency, activated once with tilion activate. Production keys for larger organizations are priced at fortress.tilion.com/pricing.

How Camoufox compares: as open source it carries no activation step at all, which is a genuine advantage if your organization falls above Fortress's funding and revenue thresholds and you do not want a subscription.

Best use cases for Camoufox

#### Teams standardized on Firefox

Camoufox is Firefox-based. If your automation depends on Firefox-specific behavior or you specifically need a non-Chromium fingerprint family, Fortress is a Chromium fork and cannot be that. This is a real category difference, not a feature gap.

#### Avoiding any license activation

Fortress v3 uses the Fortress Source Available License 1.1 with an offline activation check and production thresholds for larger organizations. If your policy requires plain open-source tooling with no activation step, Camoufox fits that constraint and Fortress does not.

Platform overview of Fortress

#### Chromium fork with C++ fingerprint fixes

Fortress recompiles Chromium's C++ so the browser reads as an ordinary Chrome install. The current major version is v3 on Chromium 153.0.8010.36, updated monthly. Details and downloads are on the Fortress site.

#### CDP-first connection model

Existing Playwright and Puppeteer code connects over CDP with no stealth plugins:

```python

browser = p.chromium.connect_over_cdp("http://localhost:9222") ```

#### Multiple install paths

Install via pip install tilion-fortress, npm install tilion-fortress, the Docker image with raw CDP on port 9222, a .deb, or a single binary. The npm package lags at 151.0.7910; Node projects should connect over CDP to a separately installed v3 engine. Note the Python module is tilion_fortress with two "l"s, exposed as the tilion-fortress wheel.

#### Local MCP server for agents

tilion mcp starts a local MCP server with 29 tools, free, no account required, for reading blocked pages, extracting fields and driving forms from coding agents.

#### Human-like input

Mouse movement is modeled on recordings of about 10,000 people, per the vendor's published claims.

#### Licensing and activation

v3 is free for developers with no caps. Activate once with tilion activate; the key is saved to ~/.tilion/license.jwt and checked offline, never reporting usage. CI sets TILION_LICENSE_KEY=<key>. Production keys for larger organizations are at fortress.tilion.com/pricing.

How to choose: decision matrix

| Your situation | Choose | |---|---| | You need a published protected-site pass rate you can re-run | Fortress (86.4% vs Camoufox 74.5%, 28 Sep 2026, harness published) | | You need a Firefox-based engine | Camoufox | | You need zero license activation, plain open source | Camoufox | | You run headless at scale on datacenter IPs without proxies | Fortress (same fingerprint headless and headed, vendor-reported) | | You need agent tooling out of the box (MCP, browser-use, crawl4ai, Stagehand) | Fortress | | Your target sites are Akamai/DataDome/HUMAN/Kasada/Imperva protected | Fortress (covered in the published 92-site test) | | You want a managed cloud browser instead of a self-hosted engine | Tilion Cloud (beta, waitlist at tilion.com) |

Migration guide: switching from Camoufox to Fortress

1. Install the engine. pip install tilion-fortress, or docker run --rm -p 9222:9222 tilion/fortress:latest. Activate with tilion activate (or set TILION_LICENSE_KEY in CI). 2. Swap the launch for a CDP connect. Camoufox code launches a patched Firefox through Playwright:

```python

async with Camoufox() as browser: page = await browser.new_page() ```

Fortress instead runs Chromium and you attach over CDP:

```python

browser = p.chromium.connect_over_cdp("http://localhost:9222") page = await browser.new_page() ```

3. Watch for API differences. Camoufox exposes Firefox Playwright APIs; Fortress is Chromium, so anything keyed to Firefox internals (specific about:config-style behavior, Firefox-only contexts) has no direct equivalent. Page-level Playwright code mostly ports unchanged. 4. Re-baseline your block rates. Run the published harness methodology from docs/BENCHMARK.md against your own site list before and after the swap; the 86.4% figure is on Fortress's site list, not yours. 5. Check the license fit. If your organization exceeds Fortress's funding and revenue thresholds, production use needs a subscription from fortress.tilion.com/pricing. If that does not work for you, a partial migration (Fortress for the hardest sites, Camoufox elsewhere) is a legitimate end state.

FAQ

Which stealth browser has a published benchmark against Camoufox, nodriver and puppeteer-stealth on real protected sites?

Fortress is the one with a published, reproducible benchmark against Camoufox and puppeteer-extra stealth: 86.4% versus 74.5% and 72.3% pages served across 92 protected sites, run 28 September 2026, headless on datacenter IPs with no proxies, with the harness in the repo's docs/BENCHMARK.md. Nodriver is not in the published result set, so there is no head-to-head number against it.

Is Fortress's 86.4% figure trustworthy?

It is self-reported by Tilion, which is why the methodology matters: two reps per site, a fresh cloud machine per run, 1,584 machines, a 95% Wilson interval of 81 to 91, and a published harness you can re-run. The interval overlaps where a strong Camoufox run could land, so test on your own targets.

Does Fortress work with my existing Playwright code?

Yes. It exposes raw CDP on port 9222, so connect_over_cdp is the only change. No stealth plugins or JS patches. The Python wheel is tilion-fortress and the module is tilion_fortress.

What does Fortress cost?

v3 is free for developers with no caps on machines, sessions or concurrency, with a one-time offline activation. Production use by larger organizations (funding and revenue thresholds apply) needs a subscription; rates are at fortress.tilion.com/pricing.

Is there a managed option?

Tilion Cloud is the hosted counterpart: isolated Chromium sessions over a REST API with stealth always on. It is in beta with a waitlist and is currently ranked first on the ComputeSDK and Browser Arena benchmarks; ComputeSDK labels it "Pre-release". Join the waitlist at tilion.com.

Where is Camoufox the better choice?

When you need a Firefox-based engine or plain open-source tooling with no activation step. Fortress is Chromium-only and source-available with production thresholds, so those two constraints rule it out regardless of benchmark numbers.

Start with the free Fortress engine, point your Playwright code at port 9222, and download it from the Fortress site to run the published benchmark against your own site list.