Changelog

What shipped

Dated, user-visible changes to the runtime, portal, evidence exports, SDKs, docs, and self-hosting. Each entry is written in the change that ships it, so the date is the date it landed.

Subscribe via RSS

September 2026

11 Sep 2026Gateway
MCP action gateway verified with OpenAI's hosted MCP tool
Agents built on OpenAI's Responses API route their tool calls through the MCP action gateway with no change to agent code. Verified against production on 11 September 2026 — tool discovery, an allowed call, and a blocked call, each with its decision record. The action gateway now answers the standard MCP handshake, so any MCP client that speaks Streamable HTTP connects the same way.
The escalate path was not part of that verification; approvals are demonstrated through the portal queue. Works with OpenAI's Responses API — there is no partnership with, certification by, or endorsement from OpenAI.
10 Sep 2026Portal
API keys keep their history when rotated or deleted
Deleting or rotating a per-user API key now soft-revokes it instead of removing it, so every request the key ever made stays attributable in the audit log. Creating or rotating a provider credential writes a platform audit event, and a key used from a new IP address raises a security audit event.
Part of the runtime attribution work landed the same day: enforcement events, action events, and runs carry the principal that sent the request (user and key), the signed trace chain covers it, and exports carry pseudonymous ids only.
10 Sep 2026Gateway
Provider-residency allowlists enforced on every supported request path, including the fallback
A deployment's allowed provider regions are checked before any provider sees data — on the primary inference API and on the OpenAI-compatible and Anthropic-compatible gateway routes — and the check is applied to the fallback target as well, so a transient error cannot route traffic to a provider outside the allowlist. Every denial is recorded as an enforcement event with the governing policy version.
- 7 September 2026: the actual primary and fallback targets are validated against the region policy before a provider adapter is called.
- 10 September 2026: the Anthropic-native /v1/messages route runs the same residency gate as /v1/chat/completions (403 and an enforcement event before any upstream call).
- Residency is enforced from the operator's tag on the provider, not verified against the endpoint: configured residency, not physical residency evidence.
9 Sep 2026Self-hosting
NVIDIA NIM as a provider, with runtime identity on the record
A NIM you run on your own GPUs registers as a provider type of its own and gets the same input policy, output policy, and per-request trace as a cloud provider. Each request records which runtime build answered and where the runtime reports it, and that record travels into the audit bundle. Verified against a live NIM (llama-3.2-3b-instruct 1.10.1 on a rented NVIDIA A10) on 9 September 2026.
Runtime identity (engine, image digest, NIM version) is recorded when the NIM reports it or the operator annotates it; otherwise the record says not reported. Norcaster does not serve the GPU. This is compatibility with NIM only; Norcaster has no relationship with NVIDIA of any kind.
3 Sep 2026Gateway
Norcaster 2.0.0: the wire rename is live
Quantlix is now Norcaster on the wire. Request and response headers are X-Norcaster-* (with X-Norcaster-Request-Id), the gateway error envelope is error.norcaster, environment variables are NORCASTER_*, and the API answers at api.norcaster.com. Legacy quantlix.ai hosts redirect, and API keys issued before the cutover keep working alongside the new ncr_ prefix.
- PyPI and CLI package: `quantlix` → `norcaster`. npm scope `@quantlix-ai` → `@norcaster`; the offline verifier is `@norcaster/verify`.
- Newly issued API keys use the `ncr_` prefix. Keys issued before the cutover stay valid; leak detectors match both prefixes.
- The chat widget is rebuilt with Norcaster branding and keeps a legacy `#quantlix-chat-root` mount as a fallback.
Norcaster (formerly Quantlix) is the trading name of Navego AB, a Swedish company. We renamed in September 2026 as we expanded beyond the Nordics; another, unrelated company uses the Quantlix name. The product, team, contracts and EU hosting are unchanged.

August 2026

13 Aug 2026Gateway
Organisation policy is a floor, not a default
Policies inherit organisation → project → deployment, and a deployment can now only tighten what the organisation allows. For every key that carries enforcement severity — contextual mode and pack tier actions, pipeline-lock mode, action-policy outcomes, fail-open flags, tool allow and deny lists, amount limits, and monitor deadlines — the stricter value wins, and the request path fails closed when the effective policy cannot be resolved.
- Changing enforcement mode and managing policies now require the matching organisation permissions, so a member can no longer downgrade enforcement on their own deployment.
- The system prompt is governed by the same resolution.
- Follow-up fixes to the fail-closed path landed on 15 Aug 2026.

June 2026

30 Jun 2026Portal
SCIM 2.0 provisioning and SSO group-to-role mapping
Identity providers can provision and deprovision users and groups over SCIM 2.0 at /scim/v2, and SSO groups map to Norcaster roles. An organisation can require SSO for sign-in, and identity configuration has its own admin UI, API, and audit trail.
Configured per organisation. Deprovisioned users lose access on their next request; provisioning and configuration changes are recorded in the audit log.
29 Jun 2026Evidence
Coverage-assurance section in audit bundles
Audit bundles now answer the auditor's hardest question — is this evidence complete? — with a coverage-assurance section that compares governed calls with ungoverned calls detected on connected provider accounts, breaks the figure down per source, and states plainly what the bundle cannot see. Every figure is tagged evidenced or attested, and a full-coverage figure is never presented as proven.
Gateway coverage was first exported into bundles and leadership reports on 12 June 2026; this change aggregates the detection sources into one section with the honesty tags, and exposes the same figures through the coverage API.

Get new entries by email

One email per entry, written in the change that ships it. No digests, no tracking.