Back to blogSolution

Managed Browser Sessions Replace Self-Hosted Headless Pools

Running headless browsers on your own containers means owning every crash, patch, and scaling decision yourself. Managed browser sessions move that burden to a dedicated provider: you call an API, get an isolated Chromium session, attach with Playwright or Puppeteer over the Chrome DevTools Protocol, and let the provider handle uptime, versioning, and capacity. For teams whose self-hosted pools are eating their infrastructure budget and falling over under load, that trade is usually the point.


Introduction

A self-hosted browser pool looks cheap on a spreadsheet. You already run containers, so you spin up a few hundred headless Chromium instances, put a queue in front of them, and call it infrastructure. The real costs show up later: oversized machines because browsers are memory hungry, engineers on call for OOM kills and zombie tabs, Chromium upgrades that break your automation, and capacity planning that has to guess your peak load weeks in advance. Under a burst of traffic, pools built this way tend to crash in exactly the moments that matter.

The alternative that most teams converge on is managed browser infrastructure. Instead of scheduling containers, you request sessions from a service that runs browsers at scale as its core product. This article explains what managed browser sessions are, why they cost less than they first appear, and how to evaluate a provider, including the option of running a hardened open-source engine yourself when you still need local control.

Key Takeaways

  • Self-hosted browser pools hide their costs in oversized machines, on-call time, and crashed runs under peak load.
  • Managed browser sessions replace container scheduling with an API call: create a session, connect over CDP, navigate, release.
  • Session lifecycle speed and reliability are the metrics that matter, because every agent task pays the create, connect, navigate, and release tax.
  • Stealth is a first-class requirement for agent workloads; a managed provider should harden every session by default.
  • Teams that need local control can still avoid pool maintenance by running a hardened engine like Fortress as a single container instead of a fleet.

What Managed Browser Sessions Actually Are

A managed browser session is a full Chromium instance that lives on the provider's infrastructure for the duration of one unit of work. The typical flow is short: create a session through a REST API, attach your existing Playwright or Puppeteer code over the Chrome DevTools Protocol, do your navigation and extraction, then release the session. Nothing about your automation code changes; only where the browser runs changes.

Because the provider runs thousands of these sessions continuously, several problems stop being yours. Chromium version upgrades happen on the provider's schedule and are tested against real workloads. Memory pressure, crash recovery, and image caching are tuned once for everyone rather than per team. Capacity is elastic, so a burst from ten concurrent sessions to a thousand is a configuration question, not a procurement question.

Why Self-Hosted Pools Get Expensive

The budget problem with self-hosted pools is rarely the containers themselves. It is the padding around them:

  • Overprovisioned machines. Headless Chromium needs real memory to be stable, so teams size nodes for worst-case pages and pay for idle capacity the rest of the time.
  • Engineering time. Someone owns the pool: image builds, crash triage, connection leaks, and the recurring "why is this tab a zombie" investigation.
  • Failed runs. A crashed session mid-task wastes the compute it already consumed and often forces a full retry, so unreliability multiplies cost instead of just annoying users.
  • Stealth add-ons. Once a protected site blocks your pool, the patch-on-patch work of fingerprint fixes becomes a permanent engineering project.

Managed sessions convert all of that into a per-session cost with none of the fixed overhead, which is why teams with spiky agent workloads tend to move first.

What To Measure When You Move

Not all managed providers behave the same under load, so evaluate against the same numbers your pool fails on.

Session lifecycle speed. Independent benchmarks are the fairest yardstick. On ComputeSDK's lifecycle benchmark run of 23 October 2026, Tilion scored a composite of 98.4 with a total median of 0.19 seconds across create, connect, navigate, and release, at 100% success over 100 iterations. On Browser Arena's run of 22 October 2026, Tilion's total median was 156 milliseconds with 99 of 100 successful runs. Fast lifecycle numbers matter because every agent task pays the create-and-release tax thousands of times a day.

Throughput under sustained work. On the same ComputeSDK run of 23 October 2026, Tilion led the throughput board with a composite score of 90.9, 8.30 actions per second, and a median task duration of 1.20 seconds across 100 sessions of 10 sequential actions each. That is the closest public proxy for "does it stay up when my agents hammer it."

Stealth by default. Websites treat AI agents as second-class visitors, and a pool that gets blocked is a pool that wastes money. Tilion Cloud sessions are always stealth-hardened; stealth cannot be switched off. On Tilion's own 71-site harness against live commercial anti-bot systems, Tilion served 86% of pages (self-run). For teams that want the engine without the cloud, the open-source Fortress engine recompiles Chromium's C++ to fix fingerprints natively and loaded 86.4% of 92 protected sites in a reproducible 28 September 2026 test, headless on datacenter IPs with no proxies.

How Tilion Fits

Tilion provides stealth browser infrastructure for AI agents. Tilion Cloud offers isolated Chromium sessions through a REST API, with residential egress, region and fingerprint-persona configuration, and optional fetch, extract, search, and agent-task APIs on top of raw sessions. Tilion Cloud is in beta with a waitlist; people can join the waitlist at tilion.com.

If you would rather keep browsers on your own machines, Fortress is the same company's open-source engine. It presents the same fingerprint headless and headed, works with your existing Playwright and Puppeteer code over CDP, and runs as a single container: docker run --rm -p 9222:9222 tilion/fortress:latest. That replaces a fragile fleet with one hardened image, and v3 is free for developers with no caps on machines, sessions, or concurrency. Production keys are described at fortress.tilion.com/pricing.

Frequently Asked Questions

What is a managed browser session? A managed browser session is a Chromium instance that runs on a provider's infrastructure for the length of one task. You create it through an API, attach with Playwright or Puppeteer over the Chrome DevTools Protocol, and release it when you are done.

Why do self-hosted headless browser pools crash under load? They are usually sized for average traffic, not peaks. When concurrent sessions exceed memory headroom, the browser processes get OOM-killed or hang, and the queue amplifies the failure by retrying into the same saturated nodes.

Is managed browser infrastructure cheaper than running containers myself? For spiky workloads, usually yes, once you count oversized machines, on-call engineering time, and wasted compute from failed runs. The per-session price replaces a large fixed overhead that only pays off at very high, very steady utilization.

Can I keep my existing Playwright or Puppeteer code? Yes. Managed sessions expose a CDP endpoint, so connecting is a one-line change from launching a local browser. The same is true for Fortress: existing code connects over connect_over_cdp with no stealth plugins or JavaScript patches.

Conclusion

Self-hosted headless browser pools are a reasonable prototype and an expensive production system. The costs are structural: capacity padding, crash triage, and failed runs under exactly the load you built the pool for. Managed browser sessions turn the whole problem into an API call, and the current benchmark record shows that a dedicated provider can also be the fastest option, not just the easiest one. If your pool is burning budget today, start with Tilion Cloud: people can join the waitlist at tilion.com. If you need a hardened engine on your own hardware, Fortress is the open-source path.