What's shipped.

Release notes for the Loupe SDK and console. Alpha builds only — versions below v1.0 may introduce breaking changes without a migration path. Pin your SDK version.

1. v0.6.0highlight

   Apr 20, 2026

   Flaky-test detection in eval suites

   • Eval runs now classify tests as pass / fail / flaky based on the last 10 runs. Flaky tests count against pass-rate but don't trigger regression alerts.
   • Added Anthropic prompt-caching-aware span collapse so repeated system prompts don't inflate token counts.
   • SDK: LangChain integration supports LCEL.
2. v0.5.2fix

   Apr 11, 2026

   Waterfall ordering fix

   • Sibling spans that started within the same millisecond could render in the wrong order. Fixed by stabilizing on trace-insertion order.
   • Fix: prompts tab copy button failed on Safari < 17.
3. v0.5.0

   Mar 30, 2026

   Prompt versioning

   • Prompts are now first-class artifacts. Edit from the console, version on save, roll forward / back per env.
   • Added /docs quickstart page and TypeScript SDK 1.2 with framework-specific integrations.
   • Evals: regression alert banner auto-renders when pass rate drops > 5pp.
4. v0.4.0

   Mar 18, 2026

   Cost + latency breakdown

   • New cost panel per prompt, per model, per endpoint. Daily resolution, 30 day retention on free tier.
   • SDK overhead benchmarked to ~4ms per call on the p99 path.
5. v0.3.0

   Mar 4, 2026

   Eval suites

   • Define eval test cases in your repo; Loupe reruns them automatically on every prompt-version change. Results surface in the Evals tab with a 30 day chart.
6. v0.2.0

   Feb 22, 2026

   Span waterfalls

   • Replaced the flat call list with a Gantt-style waterfall. Parent/child relationships preserved; color-coded per span kind.