Moli Browser — Structure first. Pixels on demand. Open source browser for AI agents.

Moli

English | 简体中文 | 日本語 | Deutsch | Français | Español

Moli is a production-ready headless browser for AI agents. Its on-demand layout and rendering design combines a complete browser runtime with a lightweight resource footprint. Moli helps your AI agent fetch and extract web pages, search the web, and automate browser tasks. Use it through the CLI, CDP, WebDriver Classic, or WebDriver BiDi. Moli supports Linux, macOS, and Windows. ## Quick start Give this prompt to your AI agent: ```text Install the skills under https://github.com/lexmount/moli/tree/main/skills, follow their instructions to download and install the latest prebuilt Moli binary, then use moli-webfetch to fetch https://example.com and show me the result. ``` ### Direct installation On Linux or macOS: ```sh curl --proto '=https' --tlsv1.2 -fsSL \ https://github.com/lexmount/moli/releases/latest/download/moli-installer.sh | sh ``` On Windows, run in PowerShell: ```powershell irm https://github.com/lexmount/moli/releases/latest/download/moli-installer.ps1 | iex ``` ## Showcase

An HTML5 game rendered by Moli and inspected through Chrome DevTools

An HTML5 game rendered by Moli and inspected live through Chrome DevTools.

rust-lang.org rendered by Moli and inspected through Chrome DevTools

rust-lang.org rendered by Moli, with its live DOM, CSS, and geometry available in Chrome DevTools.

## CLI usage ### Extract a page Render the page as Markdown with Moli's default completion strategy: ```bash moli fetch \ --dump markdown \ --wait-until done \ https://example.com ``` Or directly return a compact, model-friendly semantic tree: ```bash moli fetch \ --dump semantic_tree_text \ --wait-selector body \ https://example.com ``` For visual output, enable on-demand layout and write a viewport PNG, a full-document PNG, or a paginated PDF: ```bash moli fetch --layout --dump screenshot https://example.com > page.png moli fetch --layout --dump screenshot_full https://example.com > full-page.png moli fetch --layout --dump pdf https://example.com > page.pdf ``` Run `fetch --help` for the complete option list, including output formats, page-load/response waits, profiles, proxy settings, resource policies, and tracing options. ### Start the automation server ```bash # Basic automation server for DOM-first workloads moli serve # Enable real geometry, coordinate input, and screenshot/screencast surfaces moli serve --layout # Also fetch optional image, font, audio, video, media, and text-track resources moli serve --layout --resource ``` The same endpoint serves all three protocols: CDP, WebDriver Classic, and WebDriver BiDi. Playwright can connect directly over CDP: ```js import { chromium } from "playwright"; const browser = await chromium.connectOverCDP("http://127.0.0.1:9222"); const context = browser.contexts()[0]; const page = context.pages()[0] ?? await context.newPage(); await page.goto("https://example.com"); console.log(await page.locator("body").innerText()); await browser.close(); ``` ## Why Moli Three qualities matter most for agent workloads, and Moli brings them together: - **Full-featured** — real JavaScript, DOM, CSS, networking, storage, layout, screenshots, and standard automation protocols, all integrated into one headless browser. - **Fast** — most automation requests never need visual rendering, so structure-first operations skip layout and paint entirely. - **Resource-efficient** — layout and pixels are generated only when needed, so Moli does not have to continuously maintain and update a fully rendered visual state. What most browser automation tasks actually need is page structure, not a continuously rendered visual world. Moli treats the native DOM and style state as the single source of truth, triggering layout or software paint only for operations that genuinely require them. | Agent request | What Moli does | | --- | --- | | Extract HTML/Markdown, query the DOM, run JS, inspect network/storage | Reads browser runtime state directly — does not trigger layout or paint | | Read an element's box, hit-test coordinates, send coordinate input | Runs one layout calculation and retains only the latest frozen layout tree | | Capture a screenshot or refresh a screencast | Rebuilds from the current DOM/style, replaces the frozen tree, renders a fresh frame, and discards the frame after use |

How Moli handles a request: DOM-first by default, with layout and paint built fresh only on demand

Moli still includes the complete set of capabilities: V8, CSS, layout, text shaping, hit-testing, software paint, and more. The only difference is *when* visual work runs and *how long* its results are retained. This cost model is especially well suited to crawling, browser-use agents, retrieval pipelines, evaluation environments, and reinforcement-learning workloads. ## Current capabilities - **Complete web runtime** — streaming HTML parsing, native DOM, V8 JavaScript, modules/timers/microtasks/events, iframes and workers, CSS cascade, Fetch/XHR/WebSocket, cookies, WebCrypto, and profile-scoped storage (localStorage, IndexedDB, OPFS). - **Extraction-optimized outputs** — the CLI directly produces HTML, Markdown, JSON, semantic text trees, and frame-aware serialization, with selector/script/response waits and network tracing. - **Unified automation binary** — CDP, WebDriver Classic, and WebDriver BiDi share the same kernel and scheduler. No separate ChromeDriver, geckodriver, or browser installation is required. - **Real visual capabilities on demand** — add `--layout` to enable complete box construction, Taffy layout, Parley text layout, layout-backed hit-testing/input, viewport screenshots, and low-frequency CPU-rendered DevTools screencast frames. - **Controllable operational options** — profiles, cookies, HTTP cache, proxies, resource families, connection limits, timeouts, private-network policy, user-agent overrides, structured logging, and network diagnostics are all available. ## Moli's relationship with Lexmount Moli is Lexmount's open-source headless browser; Lexmount Browser is the managed cloud runtime and control plane built around it. **The open-source headless browser is fully usable without Lexmount Browser.** ## Cost controls Expensive browser operations in Moli require an explicit opt-in and are never enabled by default: | Mode or option | Behavior | | --- | --- | | Default | `LayoutPolicy::Mock` — deterministic geometry in a compatible format, with no real layout or paint | | `--layout` | `LayoutPolicy::OnDemand` — real layout, geometry, hit-testing, coordinate input, screenshots, screencast | | `--resource` | Fetch all optional visual/media resource families | | `--image`, `--font`, `--audio`, `--video`, `--media`, `--text-track` | Enable one specific optional resource family | | `--profile-dir`, `--http-cache-dir`, `--cookie-file` | Selectively enable the persistence required by the workload | Layout is an on-demand snapshot rather than continuously maintained state. The first geometry request (a cold start) builds a working layout tree from the current DOM/style, freezes its canonical geometry into an immutable, DOM-independent `FrozenLayoutTree`, and retains only that latest tree. Ordinary geometry reads may reuse it even if the page has changed; screenshots and screencasts always rebuild, replace the frozen tree, and never reuse stale paint results. ## Architecture Moli is a standalone browser kernel, not a Chromium wrapper. It is built in Rust, has its own ownership and lifecycle rules, and relies on: - `libcurl` — network transport and multi-request runtime - `html5ever` — HTML parsing - `rusty_v8` / V8 — JavaScript execution - Servo/Stylo — selectors, cascade, computed style - Taffy + Parley — box and text layout - AnyRender/Vello CPU, `usvg`, and the Rust image ecosystem — software rendering Document and style have a single source of truth: the native DOM and its Stylo integration. Every real refresh builds a temporary working tree from that source, optionally produces and consumes one fresh paint snapshot, freezes the canonical box/fragment geometry into a compact `FrozenLayoutTree`, and discards the working tree, style borrows, layout caches, diagnostics, and paint state. Source lookup and hit-test candidates are derived from the frozen tree when queried. The system has no incrementally maintained layout tree, damage graph, retained display list, GPU compositor, or persistent window. ## Test data The following two measured data sets show Moli's current capability envelope. The tests cover real websites, real automation clients, focused Chromium/WPT behavior checks, and a large nextest regression suite. ### Mixed public-web crawl test The test covers 192 public URLs from major Chinese and international sites. A page only counts as successful if it produces meaningful content after JavaScript runs — an HTTP 200, challenge page, login wall, empty response, or shell-only application does not count. | Browser | Useful pages | Success rate | Median time | Median RSS | | --- | ---: | ---: | ---: | ---: | | **Moli** | **103** | **53.6%** | **1.43 s** | **73 MiB** | | Chrome Headless | 101 | 52.6% | 1.43 s | 773 MiB | | Lightpanda | 85 | 44.3% | 0.97 s | 40 MiB | | Obscura | 57 | 29.7% | 1.30 s | 39 MiB | ### Sample agent workload | Metric | Moli | Chromium | | --- | ---: | ---: | | CDP ready | 34.85 ms | 169.37 ms | | Episode active p50 | 33.40 ms | 57.13 ms | | Peak PSS | 102.46 MiB | 348.82 MiB | | Peak processes / threads | 1 / 24 | 11 / 123 | In the current WPT selection used to validate Moli's agent-browser scope, one complete run recorded **1.612 million passing tests**. ## Project scope Within the agent-browser scenarios defined in its documentation, Moli is ready for production use and remains under active development. Its current intentional boundaries include: - No GUI browser, persistent window, GPU compositor, or retained multi-frame paint architecture. - It does not pursue pixel-for-pixel parity with Chrome or provide high-fidelity Canvas/WebGL/media playback. - `--layout` supports software screenshots and raster-backed CDP PDF generation, but not every Chrome screenshot or print mode is implemented. Unsupported protocol paths return explicit errors — Moli never pretends that a browser action, event, network observation, or visual result occurred. ## License Unless a file or directory states otherwise, you may use Moli under either the [Apache License 2.0](LICENSE-APACHE) or the [MIT License](LICENSE-MIT), at your option. Separately licensed third-party components and fixtures remain subject to their own licenses and notices.