lightpanda-browser

What this skill does

# Lightpanda Browser

## Overview

Lightpanda is an open-source headless browser built from scratch in Zig for machine/AI usage. It executes JavaScript and renders the DOM like Chrome, but is 11x faster, uses 9x less memory, and has instant startup. Use it via CLI shell commands instead of Playwright or WebFetch.

**Status:** Beta. Works for most sites. Complex JS-heavy SPAs may fail -- fall back to Playwright for those.

## Binary Path

The binary is at `~/.local/bin/lightpanda`. Bare `lightpanda` works if `~/.local/bin` is in PATH.

**All flags use UNDERSCORES, not hyphens** (e.g. `--strip_mode`, `--log_level`, NOT `--strip-mode`, `--log-level`).

## Ideal Command for AI Content Extraction

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal <url>
```

- `--dump` outputs rendered HTML to stdout
- `--strip_mode full` removes scripts, stylesheets, images, video, SVGs -- leaves just content
- `--log_level fatal` silences all log output on stderr

**NOTE: Output is always HTML, even with `--strip_mode full`.** This is expected and correct — `strip_mode` removes non-content elements but does NOT convert to markdown or plain text. The HTML output is clean and readable. Do NOT abandon lightpanda just because the output is HTML.

## CLI Reference

### Commands

| Command   | Purpose                                       |
| --------- | --------------------------------------------- |
| `fetch`   | Fetch a URL, execute JS, dump rendered HTML   |
| `serve`   | Start a CDP server (for Playwright/Puppeteer) |
| `help`    | Show usage                                    |
| `version` | Show version                                  |

### Fetch Options

| Flag                   | Description                                |
| ---------------------- | ------------------------------------------ |
| `--dump`               | Output rendered HTML to stdout             |
| `--strip_mode <modes>` | Comma-separated: `js`, `css`, `ui`, `full` |
| `--with_base`          | Add `<base>` tag to output                 |

Strip modes:

- `js` -- remove script tags and preload links
- `css` -- remove style tags and stylesheet links
- `ui` -- remove img, picture, video, css, svg
- `full` -- all of the above (recommended for AI)

### Common Options (fetch and serve)

| Flag                               | Default   | Description                               |
| ---------------------------------- | --------- | ----------------------------------------- |
| `--obey_robots`                    | false     | Respect robots.txt                        |
| `--http_proxy <url>`               | none      | HTTP proxy (supports user:pass@host:port) |
| `--proxy_bearer_token <token>`     | none      | Bearer auth for proxy                     |
| `--http_timeout <ms>`              | 5000      | Transfer timeout (0 = no timeout)         |
| `--http_connect_timeout <ms>`      | 0         | Connection timeout (0 = no timeout)       |
| `--http_max_concurrent <n>`        | 10        | Max concurrent HTTP requests              |
| `--http_max_response_size <bytes>` | unlimited | Limit response size                       |
| `--log_level <level>`              | warn      | debug/info/warn/error/fatal               |
| `--log_format <fmt>`               | logfmt    | pretty/logfmt                             |
| `--user_agent_suffix <str>`        | none      | Appended to "Lightpanda/1.0"              |

### Environment Variables

| Variable                            | Description             |
| ----------------------------------- | ----------------------- |
| `LIGHTPANDA_DISABLE_TELEMETRY=true` | Disable usage telemetry |

## Common Patterns

### Fetch a page (clean content)

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal https://example.com
```

### Search the web via DuckDuckGo

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal "https://duckduckgo.com/html/?q=search+terms+here"
```

Use the `/html/` endpoint for simpler, lighter HTML output from DuckDuckGo.

### Fetch with extended timeout (slow sites)

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal --http_timeout 15000 https://slow-site.com
```

### Fetch respecting robots.txt

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal --obey_robots https://example.com
```

### Fetch with base tag (for resolving relative URLs)

```bash
~/.local/bin/lightpanda fetch --dump --strip_mode full --log_level fatal --with_base https://example.com
```

## Output Behavior

- **stdout**: Rendered HTML (after JS execution)
- **stderr**: Log messages
- `--log_level fatal` or `2>/dev/null` gives clean stdout-only output
- Output is HTML, not markdown. Parse it directly or use shell tools to extract text.

## When to Fall Back to Playwright

Use Playwright instead of lightpanda when you need:

- Interactive page manipulation (clicking buttons, filling forms)
- Multi-step navigation with session/cookie state
- Screenshots or visual testing
- Complex JS-heavy SPAs that lightpanda fails to render
- Waiting for specific elements or network conditions

## Installation

macOS (Apple Silicon):

```bash
curl -L -o /usr/local/bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-aarch64-macos && chmod a+x /usr/local/bin/lightpanda
```

macOS (Intel):

```bash
curl -L -o /usr/local/bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-x86_64-macos && chmod a+x /usr/local/bin/lightpanda
```

Linux (x86_64):

```bash
curl -L -o /usr/local/bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-x86_64-linux && chmod a+x /usr/local/bin/lightpanda
```

Linux (aarch64):

```bash
curl -L -o /usr/local/bin/lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-aarch64-linux && chmod a+x /usr/local/bin/lightpanda
```