Multi-session runtime

Koko supports multi-session browser execution — multiple isolated browser instances that can run concurrently without sharing process state.

Process model

Each koko serve instance is an independent browser process with its own memory, targets, and CDP endpoint. This differs from Chromium's multi-tab single-process model where tabs share a browser process tree.

Crawl worker pattern

# Worker 1
zig-out/bin/koko serve --port 9222 --browser-profile chrome-macos-catalina

# Worker 2
zig-out/bin/koko serve --port 9223 --browser-profile chrome-macos-catalina

# Worker N...

SDK crawl helper:

import { createCrawlWorker } from "koko-sdk";

const worker = await createCrawlWorker("http://127.0.0.1:9222");
const result = await worker.crawl({ url: "https://en.wikipedia.org/wiki/Earth" });
// result.ttfexMs, result.extractMs, result.title, ...

Benchmark architecture

The Wikipedia crawl benchmark compares:

When to use multi-session

• High-concurrency crawling with isolation
• Agent pools where one bad page must not crash others
• Memory-bounded density testing
• Per-tenant or per-profile session separation

Session persistence

import { captureSessionState, restoreSessionState } from "koko-sdk";

const state = await captureSessionState(page, "https://example.com");
await restoreSessionState(page, state);