Browsers and ownership

Offprint controls Chromium-based browsers through the Chrome DevTools Protocol, the browser's remote inspection and automation protocol. Browser selection, browser source, discovery policy, and installation policy are separate concepts.

Browser selection

A capture request chooses one BrowserSpec:

Selection	Meaning
auto	Use the service's configured discovery path
executable	Launch the named local executable
remote	Attach to the named HTTP, HTTPS, WebSocket, or secure WebSocket endpoint

For automatic local discovery, Offprint considers an explicit service path, installed managed browsers, then compatible system browsers. The public BrowserSourcePolicy limits discovery to auto, managed, or system.

BrowserSource records how the selected browser was obtained: managed, system, explicit, or remote.

Managed and system browsers

A managed browser is a pinned Chrome for Testing revision whose archive digest appears in Offprint's browser catalog. Installation locks the cache, downloads from the catalog URL, verifies the archive, validates extraction, probes the executable version, and commits the staged installation.

BrowserInstallationPolicy controls automatic provisioning:

• install-managed allows first use to install the pinned managed browser.
• existing-only requires a compatible browser to be available already.

System discovery supports Google Chrome, Chromium, and Microsoft Edge based on Chromium major version 120 or newer. Managed release packages target Linux x86-64 with glibc, macOS arm64 and x86-64, and Windows x86-64.

Process and context ownership

One Offprint service reuses one local browser process. Each capture receives a fresh isolated capture context. Offline verification opens a separate verification context in a compatible service-owned process. Offprint disables browser downloads and closes each page and context on success, failure, or cancellation.

A running local process must match the next capture's executable selection and headed state. Switching those choices while the process is active returns offprint.browser.selection_conflict. Close the idle process or use separate services for incompatible browser choices.

Remote browser

Remote capture requires:

• An HTTP or HTTPS page URL
• The unrestricted network policy
• Static verification
• A browser endpoint that permits Browser, Target, Page, Runtime, DOM, Log, Security, Network, Emulation, and Fetch operations used by capture

unrestricted permits every address class. Offprint blocks page-created WebSocket, EventSource, and WebRTC transports in the managed page and frame targets. It also rejects coordinator-local file: capture. Batch requests can use a remote browser when each affected request satisfies the remote rules. Crawl rejects a remote CDP selection and can use the local default or an injected Rust backend.

Offprint owns the contexts it creates on a remote browser. The caller owns the endpoint, browser process, browser-side state, credentials, and network controls. Closing the service closes its protocol connection and leaves the remote process running.

Doctor report

offprint doctor discovers candidates, probes the selected browser and collector, checks the managed cache and output directory, and returns recovery actions. Its ready field describes the canonical local offline capture path. A deliberately configured remote static capture can work even when that field is false.

The probe can launch a short-lived local browser or create and dispose a remote context. Review endpoint side effects before running doctor against a shared remote service.

Continue with managed browser workflows or the remote browser guide.