Internal Linker — API Dashboard

Liveness probe. Returns status ok + current corpus size. The fastest way to confirm the service is up and the data file loaded. No auth.

Read-only inventory of the linkable corpus. Each entry shows id, url, title, primaryKeyword, status, publishedAt. Use this to spot-check what targets are available for linking. No auth.

HTML table flagging data-quality issues: multi-keyword PKs (comma blobs), posts with no aliases, non-slug ids. The quickest way to find ingested garbage that will produce bad links. No auth.

Concurrently HEAD-probes each corpus URL and classifies it. dead = 404/410/DNS-failure (confident broken); ambiguous = 403/5xx/timeout (could be a WAF false positive — human review). Use this monthly to catch drift after WP deletes. Admin-gated. No auto-prune — you decide what to DELETE.

Called by n8n after WordPress publish. id = stable slug, primaryKeyword = anchor phrase, aliases = extra anchor candidates, headings = H2/H3 text (relevance signal + alias source). Upsert semantics — same id replaces. Admin-gated.

Partial update. Only the fields you provide are touched; the rest are preserved. Use this to evergreen a year-bearing PK, enrich aliases, or fix a typo without a delete+reingest. Re-indexes and re-persists. Admin-gated.

Hard delete from corpus + inverted index + persisted file. Use after /corpus/verify flags a post as dead. Safe — the engine will no longer suggest links to it. Admin-gated.

Returns primaryKeywords[] (clean array — feed to the writing agent so it weaves them in) and phrases[] (every indexed phrase + its target, for debugging or richer prompt context). This is what the n8n Format Link Index node calls. No auth.

Regenerates embeddings for every post. Call after batch-adding posts without embeddings, or after changing the embedding model. Errors if no embedding provider is configured (the service runs pure deterministic). Admin-gated.

The core engine call. Given finished article markdown + sourceId, returns ranked suggestions with anchorText, targetUrl, position, score, scoreBreakdown, matchType. No mutation — safe to call as many times as you want. No auth (read-only).

Suggests + applies in one call. Returns rewritten markdown with links inserted (respecting all guardrails: skip first 100 words, no links in H1/code, max 5/article, one per target) and an audit entry per insert written to /data/audit.jsonl. This is what the n8n Apply Internal Links node calls after the Final Edit Agent. Admin-gated (it's a write).

Reverse direction of /suggest. Given the OLD post's markdown + sourceId + the NEW post's targetId, returns ranked positions where the old post should insert a link to the new one. Used after every publish to keep the link graph bidirectional. No auth.

Applies the /backlink-suggest result: rewrites the OLDER post's markdown with a link to the NEW post. Cap 1 per article. Admin-gated (it's a write). n8n calls this in the bidirectional-link loop after publish.

Stores a batch of suggestions for the approval UI. The review dashboard at GET /review then surfaces them for one-click accept/reject. If a target was previously rejected for this source, it's filtered out automatically. Admin-gated.

Lists pending review batches with accept/reject buttons. Open this in a new tab — it's a full page, not an API call. Admin-gated: requires the admin secret via a cookie/query-param handshake (also true of /review/stats, /review/:batchId, and /review/:batchId/result). The "Open" button above handles this automatically — it forwards the secret already saved in sessionStorage as ?secret= so you don't have to log in again.

Returns the live config: maxLinksPerArticle, introSkipWords, minAnchorChars, maxAnchorWords, blocklist. Use this to verify what rules /suggest and /apply are enforcing right now. No auth.

Partial update (deep merge). Override any of maxLinksPerArticle, introSkipWords, minAnchorChars, maxAnchorWords, blocklist, and weights (matchType.primary_keyword/alias/title, position, length, recency, headingCoOccurrence, semantic) — only the fields you send change, everything else is left as-is. Persists to disk (atomic write) and survives a restart/redeploy. Admin-gated.