Engineering disclosure
Architecture, security model and known limits
Owl Browser is a Chromium 156 browser engine for AI agents and large-scale automation, hosted for you by Olib AI or self-hosted on Enterprise. Its fingerprint values are produced inside the browser's C++ code. This page explains what that does and does not achieve, how we test it, the conditions that still lead to detection or data exposure, and the security controls in the product. It is not a certification or a promise that every site will accept every session.
Written and maintained by Akram Hasan Sharkar, Founder and CTO. Last updated November 17, 2026.
1. Engine at a glance
Figures verified against the product source for release 1.4.2.
| Browser engine | Chromium 156.0.8078.48 (CEF), built from source |
|---|---|
| Fingerprint code | 27 C++ override modules and 66 Chromium/CEF patches |
| Fingerprint profiles | 1,400: 100 hardware and OS bases x 14 Chrome majors (143 to 156) |
| Isolated contexts | Up to 256 per instance (documented limit) |
| API server | C99, poll-based HTTP server; 187 REST tools, WebSocket and MCP on one port |
| Server to browser | 64-connection IPC socket pool, so commands for different contexts run in parallel |
| Page rendering for agents | Per context: pixel (default), agent (text OwlMark view) or both |
| Current release | 1.4.2 |
2. System architecture and trust boundary
Owl is a Chromium and CEF browser engine. Browser contexts, profiles, page content, cookies, screenshots, proxy credentials and automation output remain on the instance that runs it. The REST, WebSocket and MCP interfaces expose the same engine on that instance; there is no shared public API at owlbrowser.net.
Where that instance runs depends on the plan. On Developer Trial and Developer it is shared hosting operated by Olib AI. On Starter and Business it is a dedicated AWS instance that Olib AI operates in the AWS region you choose. On Enterprise, the only self-hosted plan, it is your own infrastructure: the main security boundary moves to your host, container runtime, network, and chosen model or proxy providers, and Olib AI does not receive automation traffic simply because Owl is running. In every case, license activation and validation are separate requests to Olib AI's license server and include the identifiers described in the privacy policy and the license agreement.
3. How source-level fingerprint controls work
Stealth plugins usually replace JavaScript properties after a page starts. That can leave observable property descriptors, prototype changes, injected functions, or values that differ between frames and workers. Owl changes selected fingerprint values in the Chromium and Blink implementation, before those values reach page JavaScript, and keeps each context on one consistent profile (operating system, Chrome version, GPU, screen, fonts, Client Hints).
This removes one class of artifact. It does not prove that a fingerprint came from physical hardware. A detector can still correlate TLS behaviour, IP reputation, request timing, interaction patterns, account history, graphics output, fonts, codecs, or a new browser inconsistency. "Source-level" describes where a value is produced, not a guarantee of acceptance.
4. Validation and reproducibility
We test with a public harness against CreepJS and compare with default and stealth-plugin Playwright and Puppeteer. For agent efficiency we use six fixed tasks, a pinned local model, versioned agent clients, repeated runs, explicit expected answers and machine-readable grading, and we record correct, wrong and abandoned runs separately. Each published result states the date and Owl version it measured; the latest runs predate 1.4.2.
These are Olib AI's own tests, not an independent laboratory audit. The raw result files are published with the benchmark so the numbers can be checked; publishing them is not third-party validation. Results, dates and raw files: benchmark.
5. Known failure modes
No browser is undetectable across all sites and versions, Owl included.
- Poor proxy reputation, or a location, timezone, language and IP mismatch, can trigger review.
- Unrealistic click timing, navigation patterns or high request volume can reveal automation even when browser properties are consistent.
- Authenticated sites can use account history and server-side signals that the browser cannot change.
- New Chromium releases and detector updates can introduce signals before Owl adds or validates a matching control.
- Fonts, GPU behaviour, codecs, media devices and graphics hashes can differ from the selected profile on some hosts.
- CreepJS still scores Owl's profiles 31 to 38% "like headless" in our captures; a clean result on one scanner does not predict another site.
6. CAPTCHA behaviour
CAPTCHA tools send the challenge to a vision-capable model at the OpenAI-compatible endpoint you configure for the context (the Docker image does not bundle a model), and are not instant or guaranteed. Each solve makes up to a configurable number of attempts (three by default); how long it takes depends on the challenge, the hardware and whether the provider presents another challenge. Some challenges will fail or need a human. We do not publish success rates. Signed-agent identification (Web Bot Auth) avoids a challenge only on sites that recognise and allow that identity.
7. Security controls and SOC 2 status
SOC 2 aligned controls, not certified. Olib AI maps its controls to the SOC 2 Trust Services Criteria. Olib AI and Owl Browser are not SOC 2 certified, and no SOC 2 attestation report is available.
Controls built into the Owl Browser software and its container image include encrypted profile and credential storage (AES-256-GCM), TLS on network interfaces, rate limiting, audit logging of security events, non-root container execution, capability dropping and read-only container deployment options. They describe the software, not an audited description of how hosted instances are operated. They can support a customer's compliance program; deploying Owl on a compliant cloud does not make the customer's workload or Olib AI SOC 2 compliant.
Enterprise customers can request the control mapping and available evidence for vendor review at enterprise@olib.ai. Report vulnerabilities through the security program.
8. Privacy, telemetry and retention
A self-hosted Enterprise instance does not send browsing history, page content, cookies, screenshots, automation output or feature analytics to Olib AI by default; on the hosted plans that data is on the instance Olib AI operates for you. Neither case means the license service, a proxy you configure or an external model provider collects nothing. License identifiers, activation timestamps, retention periods and deletion rights are described in the privacy policy.
Operators remain responsible for retention, access controls, lawful processing, data-subject requests and the terms of the sites they automate. Self-hosting on Enterprise gives the operator control of the data plane; it does not transfer or remove those obligations.