Skip to content

v4.2.0 — AI Gateway inference and SDK 0.21.0

Choose a tag to compare

View all tags
@cdot65 cdot65 released this 07 Sep 02:45
· 110 commits to main since this release
v4.2.0
852401b

AI Gateway inference and SDK 0.21.0

  • Add airs aigateway inference chat, responses and embeddings with separate runtime endpoint/key configuration.
  • Support bounded streaming text/JSONL, backpressure, cancellation, zero automatic runtime retries and secret-safe diagnostics.
  • Pin the published @cdot65/prisma-airs-sdk@0.21.0 exactly in the manifest and lockfile. SDK publication completed first.
  • Correct unsupported DLP deletion guidance and publish actual sanitized E2E output in Docusaurus.

Verification

A fresh checkout installed from the frozen lockfile passes all 1,033 tests, coverage thresholds, lint, formatting, typecheck and build using the registry SDK rather than a local link. The strict documentation build passes after preserving the existing inference-section anchor. Registry-installed CLI 4.2.0 passes 8/8 live inference checks at 02:52:01 UTC; the published SDK independently passes 10/10 live checks. Independent inventory confirms both temporary keys absent and the read-only credential configuration unchanged. The published CLI's five distribution files match the fresh checkout build byte-for-byte.

Known limitations

Release is explicitly authorized with incomplete AI Gateway coverage: SDK direct upstream contracts remain 137/242 (56.61%), with 28 experimental methods. The complete primary SDK example run has 18 passing and 3 failing scripts (DLP profile update, dictionary upload and network-broker update). Other service/provisioning failures and one unretired unbound DLP fixture remain documented. Gateway E2E uses the documented TLS-verified LAN path, not certified public WAN connectivity. Publication does not claim 99% full AI Gateway coverage or all-green E2E.

CLI inference examples · Release notes · SDK conformance

Assets 2
Loading