[Feature] Enable Ask AI with Kapa: Source Groups, ASF CSP, and Privacy Compliance #495

Scope and current result

Prepare consent-gated Ask AI for HugeGraph's official site through #496. The PR contains enabled=true and the Kapa/hCaptcha CSP validated on OINK staging. It remains unmerged; no production website deployment has been performed.

The production origin https://hugegraph.apache.org is already allowed in the enabled Kapa widget 0b277570-4740-451e-96fa-1e4ac1ac5e88.

Production documentation sources

No frontend path mapping is required. Production's current latest is a5a9861ea15f2183347149d0943a758493f1f5b1, sharing content tree aa69a087ecaa52d4c8f7eafb8b2335d531875abd with this PR. An independent check verified all 192 production document URLs return HTTP 200 with one nonempty .td-content; 16 EN/CN samples have identical normalized body text to staging.

Both sources use anchored language-specific URL whitelists, .td-content extraction and manual refresh. Global and the common parent have no direct sources; old uploads and mixed GitHub sources remain in sibling Legacy.

Latest real CN test returned RocksDB configuration, bin/init-store.sh and bin/start-hugegraph.sh, citing https://hugegraph.apache.org/cn/docs/.... The earlier CN hCaptcha acceptance gap is resolved. A real EN test also answered successfully, but still cited staging at the last observation; EN production citation migration is therefore not yet proven complete.

Browser access has recovered. A fresh EN conversation returned a valid answer but still cited staging. The saved Start URL, anchored Include rule and .td-content selector were rechecked and are correct; the same background crawl still reports in progress with0 updates and the old96 deployed items retained (last check37 minutes earlier). It has not been stopped, restarted or duplicated.

Validation

• Current CI at dc982f312: 193 Python, 40 UI unit, 7 workflow, 76 Chromium and 8 visual checks pass. Publication was skipped for the PR.
• Three independent code reviews and fix re-reviews completed. All 13 feat: upgrade OINK to 1.1 with maintainable site extensions #496 and four fix(build): resolve OINK from candidate lock files #497 review threads were resolved; GitHub's required formal review remains a separate merge gate.
• Actual desktop/mobile EN/CN answers, language-group filtering, consent-before-loading, cancellation/focus, timeout/retry, native-search fallback, historical latest-document notices and HugeGraph icon behavior have staging evidence.
• AI-off rollback passed: a fresh browser page made zero AI requests and returned 10 native search results. AI-on restore passed, publication df426022dac03ac6db3bd9f8b4e9ac08221897fa, with all five live version SHAs matching the manifest.

Remaining work

1. Inspect the existing EN reindex job to its terminal state, then verify a fresh EN answer cites official documentation. Browser access is now available.
2. Obtain the required GitHub review before any separately authorized merge. After an authorized production publication, verify the actual production CSP and EN/CN question flow.

CSP evidence boundary

The same integration already runs on older HugeGraph staging: source, deployment. The current CSP uses only necessary Kapa/hCaptcha hosts through CSP_PROJECT_DOMAINS, retaining ASF defaults. Optional Google reference favicons remain blocked and fall back to icons without preventing answers.

ASF's FAQ permits Kapa with consent before loading. A separate VP Privacy approval record for the dependency hosts has not been located; existing deployment is not represented as verified formal approval. No automatic-refresh or OSS-plan approval claim is made.

Rollout runbook