What we shipped, what's in flight, what's next

Every executed action is rechecked against the bill at three checkpoints; receipts mark verified, inconclusive or regressed.

IaC ownership, freeze windows, policy protection, and ticket requirements run before any cloud API call.

AWS, Azure, GCP — compute, storage, network, database, Kubernetes, serverless and containers.

AWS RI + Savings Plan + Azure Reserved + GCP CUD recommendations with confidence intervals.

OpenAI / Anthropic / Gemini usage tracked per feature, per end-user, per model — without proxying customer traffic.

24-hour burn-rate anomaly detection with dashboard surfacing.

Manual approval default; per-action approval chains, freeze windows, per-role permissions, audit trail.

HKDF-derived keyring, per-row keyVersion audit column, optional AWS KMS / Azure Key Vault / GCP Cloud KMS.

Postgres row-level security on every tenant-scoped table; automated quarterly drift audit.

Enterprise identity with per-token rate limiting, rotation and soft-revoke endpoints.

Partitioned table with rolling partition-maintenance cron; plan-based retention.

Redis-backed throttling keyed by organisationId; per-IP fallback for pre-auth routes.

HTTP correlation ID flows into job context so worker logs join back to the originating request.

Threshold-flagged DLQ metric in Prometheus, admin digest endpoint with recent failure reasons.

CSV / JSON exports for audit logs, findings, resources, evidence artifacts.

Ticket creation, webhook dispatch, configuration UI with permissions guide.

Plan catalog, checkout, portal handoff, idempotent webhook handling.

Trust Services Criteria mapping done; auditor engagement scheduled.

Vendor selection underway; summary report will be made available to enterprise customers under NDA.

Quarterly cadence starting Q3 2026.

Swagger generation already in place; export to /api/openapi.json and a curated Postman collection.

Second Supabase project + Render region for customers with US-only data requirements.

Per-org daily summary of top recoverable spend, top findings, top pending actions — directly into Slack.

Better ingestion of customer mapping data and clearer cost-per-customer chart in the executive view.

Highlight features where a smaller model would meet observed quality thresholds at a fraction of the cost.

ShedLock-style lock so multiple worker instances can run safely behind a load balancer.

Auto-generated report of PlatformAdmin and per-org-admin members for SOC 2 evidence.