Skip to content

Hound 10.0.0 — Internet Archive recovery + research-grade envelope

Choose a tag to compare

View all tags
@dondai1234 dondai1234 released this 18 Jul 07:06
v10.0.0
0cbde18

The web research tool that never gives up — and tells your agent what to do next.

🗄️ Automatic Internet Archive recovery (new)

When a live site hard-blocks your agent (404 / 451 / 500 / bot wall / auth wall), Hound silently pulls the page from the Wayback Machine's closest snapshot and returns it — dated and honestly marked (source=archive.org, archived_at). Fetched via the Wayback id_ marker so the content is clean (no toolbar, original links intact). The only free, keyless, zero-setup fetch tool that does this. Opt out with archive_fallback=false or HOUND_ARCHIVE_FALLBACK=0.

🧠 Research-grade response envelope (major upgrade)

Every smart_fetch response now carries:

  • page_type — article / docs / list / forum / qa / pdf / js_shell / auth_wall / paywall / redirect / …
  • content_age_days + is_stale — from the page's own published/modified dates
  • source_type + is_official — gov / edu / github / docs / forum / news / … (conservative official flag)
  • A smart next_action: a list page points to its top links, a stale article suggests a dated search, an auth wall suggests the archive, a redirect points to the canonical URL. ~26µs per response.

🔧 Professional internals

  • Truncation can no longer silently drop fields (model_copy refactor — kills the vanish-on-truncation bug class).
  • Cache hits now return the full response (metadata / links / quality / envelope) — fixes a silent field-loss bug.
  • pytest pythonpath=[src] kills the stale-built-wheel test trap that shipped untested code 4+ times.
  • New modules envelope.py + archive.py (stdlib-only, fully annotated, httpx lazy-imported).

Compatibility

MAJOR bump: archive fallback is default-on (behavior change on the failure path). All new ResponseModel fields are additive with defaults. The 6 tools, their schemas, and all existing fields are unchanged. Opt-out exists for the one behavior change.

687 tests (+52 v10). Live check recovers a real page from archive.org end-to-end. Profiled: envelope adds ~26µs/response.

Full notes: https://github.com/dondai1234/master-fetch/blob/master/CHANGELOG.md

Assets 4
Loading