The Wire · Showcase
VERCEL QUEUES SHIPS IN PYTHON SDK, TURBOPACK GETS SMARTER MEMORY EVICTION
By RepoJournal · Filed · About Vercel
The Python SDK now ships a production-ready Queue client alongside cache improvements that unlock Redis-like backends, while Next.js fixes critical bundle bloat regressions and Eve bumps Anthropic prompt caching efficiency to 45%+.
Vercel-queue hits the SDK with a full async/sync API surface: `send()`, `@subscribe`, and `poll()` primitives for publishing and consuming from Vercel Queues [1]. This ships alongside cache strict mode [2], which now raises errors instead of swallowing them when you need the cache to behave like a real backend. Both features unlock the split-out `vercel.cache` and `vercel.internal.telemetry` packages [3], dropping the full SDK dependency for teams building queue task result storage. On the Next.js side, Turbopack's memory eviction gets an `'auto'` default mode [4] that only purges cache after snapshots when enough memory has actually accumulated to make eviction worthwhile. In parallel, the team shipped critical fixes splitting remaining "client-node"-only modules into `.browser` variants [5], closing regressions where server code leaked into client bundles. React upstream bumped [6] to pick up six new patches from Facebook. Eve's Anthropic integration now places the cache breakpoint on the last message of each request [7], pushing observed prompt cache rates to 45-48% from broken placements that looked cached but weren't. Eve 0.22.1 exports `callSlackApi` and `resolveSlackBotToken` [8] so external code stops hand-rolling fetch calls against Slack's Web API. Workflow's trace viewer gets a minimap [10], and the local-world runtime dropped two companion performance regressions: AbortSignal latency per step cut from ~500ms to near-zero [9], and chunk polling swapped from O(world) to constant-time [9].
Action items
- → Ship vercel-queue integration: full async/sync API is production-ready vercel/vercel-py [plan]
- → Update Next.js to pick up client-bundle regression fixes before next deploy vercel/next.js [immediate]
- → Upgrade Eve if running Anthropic workflows; cache efficiency doubled vercel/eve [plan]
- → Monitor Workflow beta pre-release: minimap added, local-world latency fixed vercel/workflow [monitor]
References
- [1] Implement Vercel Queue SDK ↗ vercel/vercel-py
- [2] vercel-cache: Add strict mode ↗ vercel/vercel-py
- [3] Split out `vercel.cache` and its dep `vercel._internal.telemetry` ↗ vercel/vercel-py
- [4] [turbopack] Don't evict when there is little memory to save ↗ vercel/next.js
- [5] Split remaining "client-node"-only modules into .browser variants ↗ vercel/next.js
- [6] Upgrade React from `23def8fd-20260706` to `12a4baec-20260707` ↗ vercel/next.js
- [7] fix(eve): place the Anthropic cache breakpoint on the last message of each request ↗ vercel/eve
- [8] [email protected] ↗ vercel/eve
- [9] [world-local] Fix per-step AbortSignal latency and O(world) chunk polling ↗ vercel/workflow
- [10] Add minimap to the trace viewer ↗ vercel/workflow
FAQ
- What changed in Vercel on July 9, 2026?
- The Python SDK now ships a production-ready Queue client alongside cache improvements that unlock Redis-like backends, while Next.js fixes critical bundle bloat regressions and Eve bumps Anthropic prompt caching efficiency to 45%+.
- What should Vercel teams do about it?
- Ship vercel-queue integration: full async/sync API is production-ready • Update Next.js to pick up client-bundle regression fixes before next deploy • Upgrade Eve if running Anthropic workflows; cache efficiency doubled
- Which Vercel repositories shipped on July 9, 2026?
- vercel/vercel-py, vercel/next.js, vercel/eve, vercel/workflow