internal/

Package bench measures whether the distillation is actually worth doing.

Package canvas recovers meaning from content that exists only as pixels.

Package capture defines the wire format produced by the in-page extraction script and the merge logic that folds a sequence of scroll checkpoints into a single deduplicated set of nodes.

Package cli implements sieve's subcommands.

Package corroborate holds the confirm-only index used to decide whether a string recovered from pixels was really something the site shipped.

Package distill is the orchestrator: it takes a URL and returns an artifact, choosing how much work to do along the way.

Package emit renders a content graph to the output formats.

Package escalate decides how hard to work on a page.

Package fetch is tier 0: a plain HTTP GET, and the cheap checks that decide whether anything has changed since last time.

Package graph turns a raw scroll capture into a content graph: text runs reassembled, chrome separated from content, hidden content quarantined, heading levels inferred from typography rather than tag names, and reading order recomputed from geometry.

Package llm wraps a chat model for the two places sieve needs one: captioning canvas regions that yielded nothing to the scene-graph parse, and running the benchmark's question sets.

Package mcpserver exposes sieve to agents over the Model Context Protocol.

Package render drives a headless Chromium session: it loads a page, waits for it to stop moving, sweeps the viewport through the full scroll height, and hands back a deduplicated capture of everything that was ever on screen.

Package safety holds the guards that stand between an agent-supplied URL and the network: address filtering, robots.txt compliance, and per-domain rate limiting.

Package serve hosts distilled artifacts with content negotiation.

Package snapshot records a capture so a graph can be rebuilt without a browser, a network, or the site.

Package static extracts content from served HTML without a browser.

Package textnorm cleans recovered text once, at the graph boundary.

Package tokens estimates how many tokens a piece of text will cost.