New

TrustFlow 2.0 is live — ML pipeline management with zero-downtime model swap. Learn more →

Platform Changelog

What's new in Trustinera AI.

Every release — features, improvements, fixes, and deprecation notices. We maintain a 12-month deprecation window for all breaking changes.

New Improved Fixed Performance Deprecated
v4.3.1 Patch

Hotfix: Merchant alias edge case in UK direct debits

  • ✓ Resolved edge case where UK Faster Payments direct debit descriptions with trailing spaces were not matched by the alias disambiguation layer introduced in v4.3.0
  • ✓ Corrected a race condition in the Corrections API webhook delivery that caused duplicate webhook events on high-throughput accounts (>5,000 transactions/min)
  • ⚡ Reduced median p99 latency for POST /categorise/batch from 340ms to 280ms through pre-warmed model shard caching
v4.3.0 Minor Release

Categorise v4.3: Merchant Alias Disambiguation + Real-Time Feedback Loop

+2.1pp accuracy on corporate card transactions
  • ★ Merchant alias disambiguation engine: maps 4,200+ common UK merchant aliases ("AMAZON EU SARL", "AMZN PRIME", "AMAZON DIGITAL" etc.) to canonical merchant entities before categorisation — reduces "Unknown" category assignments by 67%
  • ★ Real-time feedback loop: corrections submitted via the Corrections API are now incorporated into per-account model weighting within 15 minutes (previously next daily batch)
  • ★ Confidence band annotations: every categorise response now includes a `confidence_band` field ("HIGH", "MEDIUM", "LOW") alongside the numeric score for easier threshold configuration
  • ★ Bulk export API: GET /categorise/export now supports JSONL and Parquet formats with configurable date range and category filter parameters
  • ↑ Python SDK updated to 4.3.0: added async/await support via `trustinera.aio` subpackage, compatible with Python 3.10+
  • ↑ TypeScript SDK updated to 4.3.0: full ESM support, tree-shakable exports, Zod schema validation on all request/response types
  • ⚠ `POST /v1/classify` endpoint deprecated — use `POST /v2/categorise`. v1 will be removed March 2027 (12-month notice period)
v4.2.0 Minor Release

TrustFlow 2.0: ML Pipeline Orchestration with Zero-Downtime Model Swap

Zero-downtime model promotions
  • ★ Model promotion gates: define automated evaluation criteria (accuracy threshold, latency p99, error rate) that must pass before a new model version is promoted to production traffic
  • ★ Blue/green model deployment: TrustFlow 2.0 routes a configurable percentage of live traffic to a shadow model for comparison before cutover
  • ★ Pipeline versioning: full git-style lineage for pipeline configurations — diff, rollback, and compare pipeline versions via API or UI
  • ★ BYOM v2: bring-your-own-model now supports ONNX and TorchScript format in addition to the existing Python callable interface
  • ↑ Ingestion throughput increased from 8,000 to 22,000 transactions/second per pipeline instance via internal batching optimisation
  • ✓ Resolved memory leak in long-running pipeline workers processing streams >72 hours continuously
v4.1.0 Minor Release

Sentrise: Real-Time AML Screening + FCA Audit Pack v2

FCA audit packs in < 4 hours
  • ★ Real-time AML screening: transactions are screened against OFAC, HMRC, and combined entity watchlists within 200ms of ingestion — configurable block/flag/monitor actions per rule
  • ★ FCA Audit Pack v2: auto-generated audit documentation now includes Consumer Duty outcome evidence, transaction sampling methodology, and model decision rationale summaries in FCA-legible format
  • ★ SAR (Suspicious Activity Report) drafting: Sentrise auto-populates SAR fields from transaction clusters that match suspicious pattern templates; compliance team reviews and submits
  • ★ Governance dashboard: new Sentrise UI surface showing policy rule coverage, exception rates, AML hit rates, and model drift indicators
  • ↑ OpsIQ anomaly detection model updated — false positive rate reduced by 38% on UK open banking feeds
v4.0.0 Major Release

Platform v4: Unified API Surface, Multi-Tenant Architecture, On-Premise GA

All modules on a single API surface
  • ★ Unified API v2: all modules (Categorise, Reconcile, OpsIQ, Sentrise, TrustFlow) now share a single API surface at api.trustinera.ai/v2 with consistent authentication, pagination, and error schemas
  • ★ On-premise GA: Helm charts are now generally available for all Enterprise customers — tested on EKS, GKE, AKS, Rancher, and OpenShift
  • ★ Multi-tenant isolation: data partition keys, encryption key separation, and compliance boundary enforcement are now configurable at the account level
  • ★ SSO/SAML: enterprise identity providers (Okta, Azure AD, Google Workspace) can be configured as the authentication source for platform access
  • ⚠ v1 API endpoints will be removed in v5.0.0 (planned Q4 2026). Migration guide published at docs.trustinera.ai/migration/v1-to-v2

Stay updated

Subscribe to the release newsletter. New version? You'll know first.

Book a Demo Read the Docs