velgent.

#Changelog

Notable changes to the Velgent AI Engine API. Backwards-incompatible changes are called out explicitly.

#2026-07-14

  • Added a security block to POST /api/v1/summarise responses: PII detection and redaction counts with a per-entity-type breakdown, plus prompt-injection screening status — on every call.
  • Added confidence (0–1) to the intent block.

#2026-05-27

  • Added the Data Extractor: POST /api/v1/extract turns text, images, or PDFs into structured output. output_format selects json, html_fields, or html_document. Every field carries a 0–1 confidence score and a source anchor, verified server-side.
  • Versioned, pinnable extraction templates with admin-time schema inference from sample documents.
  • PII detection runs on every extract call and emits pii_detected security events.

#2026-05-17

  • Added the Policy Engine: POST /api/v1/policies/evaluate evaluates plain-English policies in five modes — validate, generate, decide, score, classify.
  • Added POST /api/v1/policies/chain for sequential composition and POST /api/v1/policies/graph for DAG execution with conditional edge routing.
  • Weighted score aggregation buckets multi-policy results into pass / review / block bands.

#2026-05-07

  • Added intent block to POST /api/v1/summarise responses. Includes action_needed, due_date, urgency_score, and sensitivity fields. Existing fields are unchanged.
  • Per-operation summarisation routing is now live: summarise-text, summarise-ticket, summarise-approval, and summarise-knowledge each use a tuned prompt for their input shape.

#2026-04-12

  • Added Idempotency-Key support to all POST endpoints. Repeat calls within 24h return the original response.

#2026-03-01

  • Initial public release of the Velgent AI Engine API.
  • Operations available at launch: summarise-text, summarise-ticket, summarise-approval, summarise-knowledge.