Lightpanda
A headless browser written from scratch in Zig for AI agents and automation — not a Chromium fork — claiming ~9x faster execution and ~16x less memory than headless Chrome. It is the default engine behind Kevin's Browser Testing Skills stack.
What it is
Lightpanda (lightpanda-io/browser) is a new headless browser built specifically for the AI-agent and automation workload, where thousands of short-lived page loads dominate and a full desktop browser is wasteful. It is not a Chromium fork or a WebKit patch — it is a ground-up implementation written in Zig, exposing a Chrome DevTools Protocol (CDP) server so existing CDP clients (Playwright, Puppeteer, Browser Testing Skills) can drive it without rewrites. Source: lightpanda-io/browser README, https://github.com/lightpanda-io/browser, 2026-06-12
The thesis: every agent doing web automation today spins up headless Chrome under the hood, which means a ~2 GB, multi-process desktop browser per session just to read a few DOM nodes. Lightpanda strips that to a single lightweight binary. Source: X/@ihtesham2005, 2026-03-15
Current Source Snapshot
As of the 2026-07-02 source check, lightpanda-io/browser is AGPL-3.0, has 31.5K stars and 1.4K forks, latest GitHub release 0.3.4 published 2026-07-01, and default-branch HEAD 60aacfa2b064877c3eecddf98a5dd61048c1bdd8 from 2026-07-02. Source: GitHub API, 2026-07-02
The reviewed bookmark image showed the README at 18K stars and the older launch framing of "11x faster / 9x less memory"; keep that as historical launch evidence, not current source authority. Source: local X artifact wiki/assets/x-bookmarks/2033166398214783443/image-01.jpg, reviewed 2026-07-02
Benchmarks
Requesting 933 real web pages over the network on an AWS EC2 m5.large instance: Source: lightpanda-io/browser README, https://github.com/lightpanda-io/browser, 2026-06-12
| Metric (100 pages) | Lightpanda | Headless Chrome | Difference |
|---|---|---|---|
| Memory (peak) | 123 MB | 2 GB | ~16x less |
| Execution time | 5 s | 46 s | ~9x faster |
The viral launch tweet framed the gap as "11x faster, 9x less memory"; the repo's current published benchmarks show ~9x faster / ~16x less memory, so treat the exact multiplier as workload-dependent and the order-of-magnitude win as the durable claim. Source: X/@ihtesham2005, 2026-03-15
Install
# Homebrew (nightly)
brew install lightpanda-io/browser/lightpanda
# Direct binary (Linux x86_64)
curl -L -o lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-x86_64-linux && chmod a+x ./lightpanda
./lightpanda version
Nightly binaries ship for Linux and macOS (x86_64 + aarch64). There is no native Windows binary — run it inside WSL2. Linux releases are linked against glibc, so musl distros (Alpine) need a glibc base image or a source build. Source: lightpanda-io/browser README, https://github.com/lightpanda-io/browser, 2026-06-12
Where it fits Kevin's stack
Per AGENTS.md, agent-browser defaults to the Lightpanda engine for all ad-hoc browsing, scraping, form filling, and screenshot capture, and only falls back to --engine chrome for auth-heavy flows. Lightpanda does not support --extension, --profile, --state, or --allow-file-access; those require the Chrome engine. Source: AGENTS.md Layer 7, kevin-wiki, 2026-06-12
| Browser need | Tool / engine |
|---|---|
| Ad-hoc scraping, screenshots, agent loops | Browser Testing Skills on Lightpanda (default) |
| Auth/profile/extension/file-access flows | agent-browser --engine chrome |
| Site-specific reusable playbooks | Browse.sh (Browserbase Skills) / Autobrowse (Browserbase Skills) |
| Visual regression against real Chrome | Browser Harness - Self-Healing Browser Automation |
| Deterministic CI gates | Playwright |
This is the speed-and-memory floor for the Complementary Browser Testing strategy: cheap Lightpanda passes for the agent loop, heavier engines reserved for what genuinely needs them.
Do not confuse this with stealth. Lightpanda is optimized for agent automation throughput, not anti-bot realism. For protected scraping targets where TLS, HTTP/2, JS, automation-protocol, or IP fingerprints are the gate, route through Web Scraping Stealth (TLS Fingerprinting & Anti-Bot Evasion) and consider managed crawl APIs or Camofox Browser instead. Source: X/@leftcurvedev_, 2026-04-09; Source: GitHub jo-inc/camofox-browser, 2026-06-30
Timeline
- 2026-03-15 | @ihtesham2005's "RIP Chrome for AI agents" thread surfaced Lightpanda to a wide audience — headless browser from scratch, "11x faster, 9x less memory" than Chrome. 1,350 likes, 2,010 bookmarks. Source: X/@ihtesham2005, 2026-03-15
- 2026-07-02 | Deep-reviewed the local Lightpanda screenshot artifact and refreshed source authority: GitHub release
0.3.4, HEAD60aacfa2b064877c3eecddf98a5dd61048c1bdd8, AGPL-3.0, 31.5K stars, 1.4K forks. The durable route stays speed/token-efficient agent browsing, not stealth scraping. Source: GitHub API and local X artifact, 2026-07-02 - 2026-06-30 | Added the stealth caveat from the TLS-fingerprinting bookmark review: Lightpanda is a speed/token-efficiency tool, while protected scraping may need Camofox Browser, Camoufox, or managed crawl APIs. Source: X/@leftcurvedev_, 2026-04-09
- 2026-06-12 | Page created from X-bookmark absorption (
2033166398214783443); benchmarks, install, and engine-fallback constraints sourced from the GitHub README andAGENTS.md. Source: x-bookmark-absorb, 2026-06-12