Hyperbrowser launches CLI with under-50ms sandboxes and hx web commands

• Hyperbrowser shipped an hx CLI that turns sandbox lifecycle into terminal commands, with the launch post claiming startup in under 50ms via hyperbrowser's launch post and the Sandbox introduction repeating the same figure.
• The first surface area is basic VM control: according to hyperbrowser's connect and exec demo, you can attach to a running sandbox with hx vm connect or run commands directly with hx vm exec.
• Web data extraction also moved into the CLI, where hyperbrowser's web commands post adds hx web fetch, hx web search, and hx web crawl, mirroring the Web API overview.
• Stateful workflows are part of the pitch, because hyperbrowser's snapshot demo shows hx vm snapshot create and restore-from-snapshot flows, while the snapshot docs describe snapshots as reusable memory state for new sandboxes.

You can browse the CLI docs, skim the web API surface, and the same API stack is already showing up in higher-level demos, including a design-system extractor that Hyperbrowser says is built on Fetch hyperbrowser's design-md demo. The weirdly useful bit is that browser infra, crawl endpoints, and snapshotting now sit behind shell verbs instead of SDK calls.

The launch thread makes the CLI look like a thin shell over the sandbox API: create, connect, exec, expose ports, snapshot, and stop all show up in the terminal UI screenshot from hyperbrowser's launch post.

The docs fill in the rest. The CLI reference says hx covers profile setup, sandbox creation, volume management, process execution, file transfer, terminals, snapshots, image builds, and port exposure. The sandbox lifecycle docs describe the same handle-based flow for reconnecting to running sandboxes and exposing ports.

Hyperbrowser's three launch commands line up almost exactly with the product taxonomy in the docs:

• hx web fetch: fetch one URL, returning markdown, HTML, links, screenshots, or structured JSON, per the Web API overview
• hx web crawl: crawl a site and return structured data from multiple pages, per the same overview
• hx web search: search the web and return structured results, again from the overview

That makes the CLI more than a sandbox manager. It is also a shell entry point for the company's scraping and search APIs.

Snapshotting is the part most geared toward agent workflows. In the thread, Hyperbrowser frames it as state preservation and workflow forking, and the snapshot docs say a running sandbox can be captured as a memory snapshot and used to start fresh sandboxes from that saved state.

The practical effect is simple: environment setup can move from repeated bootstrap steps to reusable state. The CLI commands in hyperbrowser's snapshot demo map directly to that pattern with hx vm snapshot create and hx vm create --snapshot.

The docs add one last piece that the launch thread mostly skips: hx is not only a command runner. The CLI page also documents named profiles, hx auth login, and an interactive hx dash view for browsing sandboxes, images, snapshots, web tools, browser tools, and config.

That matches the screenshot in hyperbrowser's launch post, which shows tabs for sandboxes, images, snapshots, web, browser, and config inside the TUI. The shell-first pitch is real, but Hyperbrowser also shipped a small terminal console around it.