Running 100 Browser Sessions in Parallel Without Losing Your Mind

One of the first questions teams ask when building agent pipelines is: how do we scale this?

A single browser session consumes ~150–300MB of RAM. Run 100 of them naively and you're looking at 30GB just for browser processes — before your actual application code.

The naive approach

Most teams start by spinning up one Playwright instance per task. It works fine at 5 concurrent tasks. At 50, you're OOMing. At 100, you've given up.

What Clawbrowser does differently

Instead of one-process-per-session, Clawbrowser uses a shared browser pool with isolated contexts. Each "session" from your agent's perspective is a lightweight browser context — separate cookies, storage, and network state — but sharing the underlying browser process.

The math changes dramatically:

Approach	100 sessions RAM
Naive (1 process per session)	~20–30 GB
Clawbrowser (shared pool)	~2–4 GB

Session lifecycle

Sessions in Clawbrowser are cheap to create and destroy. The API is designed around task-scoped sessions:

# Start a session
claw session start --profile work

# Run your task
claw navigate https://example.com --session work
claw extract "main article" --session work --json

# Done — session is cleaned up
claw session end --session work

What this means for your pipeline

You can run your agent tasks in parallel without worrying about resource exhaustion. Clawbrowser handles the pool management — you just describe what you need.