sailfork / t3code-github / .repos/ alchemy-effect/ website/ src/ content/ docs/ blog/ 2026-10-22-beta-84.md

beta.84 splits alchemy into one package per provider, so you install @alchemy.run/aws or @alchemy.run/cloudflare next to alchemy instead of importing alchemy/AWS. Every AWS, GCP and Cloudflare binding now works from any host, so a Worker can call SQS and a Lambda can write to R2.

It also adds @alchemy.run/domain, a domain-driven design framework that runs on Durable Objects, and rebuilt distilled SDKs that add 5–100× less overhead per call than the official AWS, Cloudflare and GCP SDKs, in bundles a third smaller than before. On AWS, there are Lambda Web Functions with canary deploys. On Cloudflare, alchemy dev for Vite sites now behaves like @cloudflare/vite-plugin, and the runtime bindings use a lot less CPU.

:::caution Breaking changes

  • Providers moved into their own packages (#2348, #2316, #2313, #2311, #2309). Install the packages you use and update your imports. See below.

  • State store commands moved under alchemy state (#2288). The old commands print the new one and exit 1:

    # Cloudflare state store
    -alchemy provider cloudflare bootstrap
    +alchemy state bootstrap --backend cloudflare
    -alchemy prisma generate
    +alchemy provider prisma generate

    In CI, a missing or outdated state store fails unless you pass --yes. AWS.state({ name }) is now a short logical name. Use bucketName to pass a full bucket name.

  • AWS.Lambda.isBindingHost is removed (#2260, #2347). Custom AWS bindings use AWSBindings.Bind and AWSBindings.withCredentials. EC2, LaunchTemplate, AppRunner and Batch hosts now get IAM grants, so they plan new IAM resources on the next deploy.

  • Distilled error classes are no longer Effect Schemas (#2345). Match them with their is guard. Generated request and response schemas are no longer exported; reach them through the operation:

    // match a distilled error
    -Schema.is(TooManyRequests)(error)
    +TooManyRequests.is(error)
    -PutDispatchNamespaceScriptRequest.ast
    +putDispatchNamespaceScript.input.ast
  • Prisma and Neon env values use the runtime wire format (#2244). Reading them through Prisma.Connect or the Neon clients is unchanged. Code that reads process.env directly sees secrets as {"_tag":"Redacted","value":"…"}.

  • AWS framework sites run on Web Functions by default (#2282, #2300). The next deploy creates a WebServer Web Function and deletes the old server Lambda, keeping the CloudFront distribution and domain. Set server: "function" to keep a Lambda Function URL server.

  • Effect 4.0.5 (#2372). Bump effect and @effect/* to ^4.0.5. :::

One package per provider

alchemy keeps the engine, the CLI, and the local and in-memory state stores. Every cloud and vendor now has its own package, and each one brings its own resources, bindings and state store, such as AWS.state() or Cloudflare.state() (#2348, #2316, #2313, #2311, #2309):

bun add alchemy@latest @alchemy.run/cloudflare@latest @alchemy.run/aws@latest
import * as Alchemy from "alchemy";
-import * as AWS from "alchemy/AWS";
-import * as Cloudflare from "alchemy/Cloudflare";
-import * as Neon from "alchemy/Neon";
-import * as Drizzle from "alchemy/Drizzle";
-import * as Postgres from "alchemy/SQL/Postgres";
+import * as AWS from "@alchemy.run/aws";
+import * as Cloudflare from "@alchemy.run/cloudflare";
+import * as Neon from "@alchemy.run/neon";
+import * as Drizzle from "@alchemy.run/drizzle";
+import * as Postgres from "@alchemy.run/sql/Postgres";

The same goes for gcp, fly, railway, hetzner, prisma, planetscale, turso, stripe, github, git, acme, axiom, doppler, infisical, docker, kubernetes, redis and domain. Each package exports its directories and files, so @alchemy.run/aws/S3 and @alchemy.run/aws/S3/Bucket both resolve.

A few things moved within packages:

import * as Cloudflare from "@alchemy.run/cloudflare";
-solver: ACME.Cloudflare.ZoneSolver(zone),
+solver: Cloudflare.ACME.ZoneSolver(zone),
-Stripe.Cloudflare.ConsumeEventsLive
+Stripe.ConsumeEventsLive
-GitHub.Cloudflare.RepositoryEventSourceLive
+GitHub.RepositoryEventSourceLive
-import { ServerHost } from "alchemy/Server";
+import { ServerHost } from "alchemy/Runtime/Process";
-import { LibSQL } from "alchemy/SQL/LibSQL";
+import { LibSQL } from "@alchemy.run/sql/SQLite";

The CLI finds each provider's auth, state store and alchemy provider commands once your project depends on its package, through a new Contributions module (#2288). Named state stores come with it:

state: Cloudflare.state({ name: "team-a" }),
alchemy state bootstrap --backend cloudflare --name team-a

Runtime bundles shrank along the way. A Worker no longer bundles the stack builder or CLI auth code (346 KB → 216 KB, #2240), and a Lambda that imports the Cloudflare barrel drops from 3.26 MB to 0.84 MB (#2267). Thanks Rahul Mishra, who did most of this split!

OAuth can now create Account API Tokens

Cloudflare OAuth can now grant Account API Token Provisioning, so alchemy can mint scoped API tokens on your behalf. To add it, re-run the Cloudflare login and pick All Scopes:

alchemy profile edit --reconfigure Cloudflare

Then turn the scope on under Sensitive scopes on Cloudflare's consent page. It's off by default:

Cloudflare's consent page with the Account API Token Provisioning toggle under Sensitive scopes

Login now goes through the Cloudflare CLI's (cf) OAuth client with device auth, the same flow as cf login (#2444). Existing profiles keep refreshing.

Docs: Cloudflare setup.

Every binding works from every cloud

In beta.83, only the S3, Lambda and DynamoDB bindings worked outside AWS, no GCP binding worked outside GCP, and Cloudflare's *Http bindings crashed on any host that wasn't a Worker. As part of the package split, every binding now works on any host: all ~2,850 AWS bindings across 205 services (#2260), all 192 GCP bindings across 57 services (#2347), and every Cloudflare *Http binding (#2267).

A Cloudflare Worker sends to SQS with the same binding a Lambda uses. You provide its *Http layer:

import * as AWS from "@alchemy.run/aws";
import * as Cloudflare from "@alchemy.run/cloudflare";
import * as Effect from "effect/Effect";
import * as HttpServerResponse from "effect/http/HttpServerResponse";

export default class Api extends Cloudflare.Worker<Api>()(
  "Api",
  { main: import.meta.url },
  Effect.gen(function* () {
    const queue = yield* AWS.SQS.Queue("Jobs");
    const sendMessage = yield* AWS.SQS.SendMessage(queue);
    return {
      fetch: sendMessage({ MessageBody: "hello" }).pipe(
        Effect.map(() => HttpServerResponse.text("queued")),
      ),
    };
  }).pipe(Effect.provide(AWS.SQS.SendMessageHttp)),
) {}

On AWS hosts, a binding grants its permissions to the host's own role. On any other host, such as a Worker or Cloud Run, Alchemy creates a least-privilege IAM user and role for it and signs each call with that role. GCP works the same way with a service account and key. The other direction works too: a Lambda writes to R2 through Cloudflare.R2.WriteBucketHttp, which binds a scoped Cloudflare API token. Hosts that run as a role or service account leave this code out of their bundles. Thanks Rahul Mishra!

Docs: Bindings.

Domain-driven design with @alchemy.run/domain

@alchemy.run/domain models your backend in your business's terms (#2176). Commands ask for changes, aggregates decide them, and events record what happened. Views, policies, sagas and projections react to those events. The same model runs in memory for tests and on Durable Objects in production.

A command declares its target, input, events, rejections and reply:

import * as Domain from "@alchemy.run/domain";

export class MoneyDeposited extends Domain.Event("MoneyDeposited", {
  amount: Schema.Int,
  balanceAfter: Schema.Int,
}) {}

export class DepositMoney extends Domain.Command("DepositMoney", {
  target: AccountId,
  input: { amount: Schema.Int },
  events: [MoneyDeposited],
  rejects: [NotOpen],
  reply: { balance: Schema.Int },
}) {}

The aggregate decides it against its current state:

DepositMoney: (s, cmd) =>
  s._tag !== "Open"
    ? new NotOpen()
    : [new MoneyDeposited({ amount: cmd.amount, balanceAfter: s.balance + cmd.amount })],

A story tests it in memory: given this history, when this command, then these events.

story(
  Story.given(account, new AccountOpened({ owner: "sam" })),
  Story.when(account, new DepositMoney({ amount: 100 })),
  Story.then(account, new MoneyDeposited({ amount: 100, balanceAfter: 100 })),
  Story.replied({ balance: 100 }),
);

On Cloudflare, provide DomainCloudflare.DurableObjects and every aggregate, view key and policy gets its own Durable Object. Views are security boundaries, and clients can watch them live over WebSockets. The package ships an alchemy-domain skill so your coding agent can build a model with you.

Docs: Domain · Tutorial · Cloudflare Durable Objects.

Smaller, faster distilled SDKs

Alchemy's cloud SDKs come from distilled. This release picks up a rebuilt distilled (distilled#873, #2345) that dedupes schemas, builds them lazily, and replaces Effect Schema with a lighter schema tree. Medians over 98 one-auth, one-call programs:

BeforeAfterChange
Bundle size (gzip)57.1 KB38.3 KB−33%
Added latency per call (p50)19.5 µs4.0 µs−79%
Import heap2.15 MB0.57 MB−74%
First call (cold)4.9 ms3.3 ms−34%

Importing a whole provider is where it shows most. AWS has 438 services, GCP 318 and Cloudflare 129:

ProviderImport heapImport time
AWS227 MB → 13.9 MB1.61 s → 0.14 s
GCP111 MB → 9.4 MB0.75 s → 0.09 s
Cloudflare76 MB → 3.2 MB0.63 s → 0.05 s

The packages you install from npm shrink too:

PackageDownload (tgz)Unpacked
@distilled.cloud/aws23.3 MB → 19.3 MB211 MB → 143 MB
@distilled.cloud/gcp42.3 MB → 36.1 MB345 MB → 262 MB
@distilled.cloud/cloudflare9.5 MB → 5.8 MB112 MB → 58 MB

Faster than the official SDKs

The same calls through the official SDKs: AWS SDK v3, the Stainless-built cloudflare package, and @google-cloud/storage. Each official program imports one client. Each distilled program imports the whole provider and is still smaller:

CallAdded latency per callBundle (gzip)Import heap
GCP storage.getBuckets674 → 6.2 µs239 → 40 KB5.1 → 0.7 MB
S3 ListBuckets140 → 19 µs157 → 61 KB4.4 → 1.1 MB
EC297 → 16 µs265 → 55 KB12.2 → 1.5 MB
DynamoDB87 → 14 µs145 → 54 KB3.3 → 1.0 MB
Lambda79 → 15 µs141 → 55 KB3.6 → 1.1 MB
Cloudflare workers.scripts.list15.5 → 3.0 µs125 → 56 KB15.3 → 0.7 MB

Distilled adds 5–100× less overhead per call, ships bundles 2–6× smaller, and imports with 3–22× less memory. Its first cold call is faster too, except on Cloudflare, where it takes 3.4 ms against the official SDK's 1.9 ms.

The Cloudflare package now signs R2 presigned URLs with the standalone @distilled.cloud/aws4signer, so it no longer depends on the AWS SDK (#2370). Thanks Michael K!

Plan and apply scale linearly

Plan and apply used to do work proportional to the square of the stack's resource count. Apply rebuilt every resource's outputs once per resource, and plan scanned every node to find each node's dependents. Both are now linear (#2419):

Stack shapeResourcesBeforeAfter
Independent4,0003.4 s1.7 s
Independent8,0009.5 s3.8 s
20 dependencies each1,0001.2 s1.1 s
20 dependencies each2,0003.3 s2.1 s

Timings are engine overhead only, measured with no-op providers. Thanks gpu004 for profiling it!

Foldkit SSR and prerendering

Foldkit apps can now render on the server and prerender pages, on every website provider (#1313). Turn on ssr.build in the Foldkit Vite plugin:

// vite.config.ts
export default defineConfig({
  plugins: [
    foldkit({
      ssr: {
        serverEntry: "/src/entry.server.ts",
        clientEntry: "/src/entry.ts",
        build: { prerender: true },
      },
    }),
  ],
});

Alchemy deploys the server Foldkit generates and serves prerendered pages as static files. On Cloudflare, Fly, Hetzner, Railway, Neon, Prisma and GCP, the build decides whether the site is an SPA, static, SSR or a hybrid:

const site = yield* Cloudflare.Website.Foldkit("Web");

On AWS, you choose between a server and a static site:

const app = yield* AWS.Website.Foldkit("Web", { output: "server" });
const docs = yield* AWS.Website.Foldkit("Docs", { output: "static", errorPage: "404.html" });

Foldkit pairs with @alchemy.run/domain for the front end of a domain model. Thanks Filip Falcon!

Docs: Foldkit on Cloudflare · Foldkit on AWS.

Lambda Web Functions

AWS.Lambda.WebFunction runs your HTTP server on Lambda Web Functions (#2243). Each environment serves up to 128 requests at once, responses stream, and every function gets an HTTPS endpoint. An Effect Web Function returns fetch, and existing bindings work unchanged:

export default class Api extends AWS.Lambda.WebFunction<Api>()(
  "Api",
  { main: import.meta.url, authType: "ApplicationManaged" },
  Effect.gen(function* () {
    const bucket = yield* AWS.S3.Bucket("Uploads");
    const putObject = yield* AWS.S3.PutObject(bucket);
    return {
      fetch: Effect.gen(function* () {
        yield* putObject({ Key: "hello.txt", Body: "Hello!" });
        return HttpServerResponse.text("Saved");
      }).pipe(Effect.orDie),
    };
  }).pipe(Effect.provide(AWS.S3.PutObjectHttp)),
) {}

Or bring your own server (Hono, Express, node:http). It listens on AWS_LAMBDA_HTTP_ENDPOINT:

yield* AWS.Lambda.WebFunction("Api", {
  main: "./src/server.ts",
  authType: "ApplicationManaged",
  regions: ["us-east-1", "eu-west-1"], // one global domain
});

canary sends a share of traffic to each new deploy (#2280). Edit the weight to promote or roll back:

canary: { weight: 10 }, // the new revision gets 10%, the previous one 90%

Every server-rendered AWS.Website (Nuxt, SvelteKit, Astro, React Router, SolidStart, TanStack Start, Waku, Octane, Vinext) now runs on a Web Function by default (#2282). Next.js runs as its own standalone server with an S3 cache, without OpenNext, SQS or DynamoDB (#2300). Existing sites roll forward on the next deploy.

Docs: Web Functions · Next.js.

Effect workflows on every engine

alchemy/Workflows runs effect/workflow workflows on Durable Objects, Cloudflare Workflows, and Lambda durable functions (#2252, #2268, #2390).

Workflows.layer implements a workflow. Pass it an Effect that resolves bindings, here an R2 bucket, and return the handler that uses them:

// src/Onboard.ts
import * as Cloudflare from "@alchemy.run/cloudflare";
import * as Workflows from "alchemy/Workflows";
import * as Effect from "effect/Effect";
import * as Schema from "effect/Schema";
import { Activity, DurableClock, Workflow } from "effect/workflow";

export const Profiles = Cloudflare.R2.Bucket("Profiles");

export const Onboard = Workflow.make("Onboard", {
  payload: { userId: Schema.String },
  success: Schema.String,
  idempotencyKey: ({ userId }) => userId,
});

export const OnboardLive = Workflows.layer(
  Onboard,
  Effect.gen(function* () {
    const profiles = yield* Cloudflare.R2.WriteBucket(Profiles);
    return Effect.fn(function* ({ userId }) {
      yield* Activity.make({
        name: "create-profile",
        execute: profiles
          .put(`${userId}.json`, JSON.stringify({ userId, plan: "trial" }))
          .pipe(Effect.orDie),
      });
      // the run suspends for a week and costs nothing while it waits
      yield* DurableClock.sleep({ name: "trial", duration: "7 days" });
      yield* Activity.make({
        name: "end-trial",
        execute: profiles
          .put(`${userId}.json`, JSON.stringify({ userId, plan: "free" }))
          .pipe(Effect.orDie),
      });
      return `onboarded ${userId}`;
    });
  }),
);

The Worker starts runs with Workflows.RunWorkflow and reads them back. Give it the implementation, the R2 binding and an engine, here Durable Objects:

// src/Api.ts
import * as Cloudflare from "@alchemy.run/cloudflare";
import * as Workflows from "alchemy/Workflows";
import * as Effect from "effect/Effect";
import { HttpServerRequest } from "effect/http/HttpServerRequest";
import * as HttpServerResponse from "effect/http/HttpServerResponse";
import * as Layer from "effect/Layer";
import { Onboard, OnboardLive } from "./Onboard.ts";

export default class Api extends Cloudflare.Worker<Api>()(
  "Api",
  { main: import.meta.url },
  Effect.gen(function* () {
    const onboard = yield* Workflows.RunWorkflow(Onboard);
    return {
      fetch: Effect.gen(function* () {
        const request = yield* HttpServerRequest;
        const url = new URL(request.url, "http://localhost");
        if (request.method === "POST") {
          const userId = url.searchParams.get("user")!;
          const run = yield* onboard.start({ userId });
          return HttpServerResponse.text(run.executionId);
        }
        const runId = url.searchParams.get("run")!;
        // NotStarted | Running | Completed | Failed | Interrupted | Died
        const status = yield* onboard.get(runId).status;
        return HttpServerResponse.text(status._tag);
      }).pipe(Effect.orDie),
    };
  }).pipe(
    Effect.provide(
      OnboardLive.pipe(
        Layer.provide(Cloudflare.R2.WriteBucketBinding),
        Layer.provideMerge(Cloudflare.Workflows.DurableObjectEngine),
      ),
    ),
  ),
) {}

Swap DurableObjectEngine for Cloudflare.Workflows.NativeEngine to run it on Cloudflare Workflows instead. On AWS, host workflows in an AWS.Lambda.WorkflowFunction and start them from other functions with AWS.Lambda.DurableFunctionEngine.

Docs: Cloudflare · AWS.

Vite dev that matches @cloudflare/vite-plugin

A Website.Vite Worker now behaves in alchemy dev as it does deployed, and as it does under @cloudflare/vite-plugin (#2367, #2365, #2144). Durable Objects from other Workers, Workflows, Containers, WebSockets, ctx.access, Node.js polyfills, HTTPS dev servers and Wrangler's /cdn-cgi/local/* endpoints all work in dev. Asset routing follows your htmlHandling, notFoundHandling and runWorkerFirst:

yield* Cloudflare.Website.Vite("App", {
  main: "./api/index.ts",
  assets: { notFoundHandling: "single-page-application", runWorkerFirst: ["/api/*"] },
});
// dev before: /api/some-file.txt → the file on disk
// dev now:    /api/some-file.txt → the Worker, as deployed

A Vite config that keeps cloudflare() from @cloudflare/vite-plugin now runs, so RedwoodSDK, Flue and the Agents SDK build as documented. Tests run the official plugin next to Alchemy's on the same projects and compare config, build output and runtime behavior. The full list is in PARITY.md.

The dev server also refuses .env, .dev.vars and .alchemy/, and no longer forwards a request like //127.0.0.1:47667/x to another local port (#2381). Thanks Arman Charan!

Faster Cloudflare runtime

Dan van der Merwe measured the runtime bindings on deployed Workers and Durable Objects and cut their CPU time:

WorkloadMean CPU before → after
Copy a 16 MiB R2 object with get + put (#2276)630 → 3.8 ms
Prepare five 1 MiB Workers AI images (#2274)270 → 10.6 ms
1,000 workflow result round trips (#2271)358 → 32 ms
2 MiB through the RPC codec (#2272)85 → 20 ms
Pack 500 queue messages (#2275)235 → 92 ms
Drain 1,000 Durable Object scheduled events (#2277)33 → 8.4 ms
2,000 env reads in a Durable Object (#2279)29 → 4.9 ms

SQL cursors allocate only what you use (#2281). RPC streams whose values arrive after the method returns now deliver every value (#2273), and durable callbacks that keep failing now back off and park after eight attempts instead of waking forever (#1908).

RDS migrations everywhere

migrations now works on every PostgreSQL, MySQL and MariaDB DBInstance, standalone or Aurora (#2125, #2289). Aurora runs them over the Data API. A public database connects directly, and a private one runs them from a Lambda in the VPC:

const runner = yield* AWS.RDS.MigrationRunner("MigrationRunner", {
  subnetIds: network.privateSubnetIds,
  securityGroupIds: [runnerSecurityGroup.groupId],
});

const db = yield* AWS.RDS.DBInstance("Db", {
  engine: "mysql",
  // ...
  migrations: "./migrations",
  migrationsTransport: runner,
});

Thanks gegi!

Docs: RDS · Drizzle on Aurora.

GCP state store

GCP.state() keeps deploy state in a Cloud Storage bucket in your project (#2379):

export default Alchemy.Stack(
  "MyApp",
  { providers: GCP.providers(), state: GCP.state() },
  program,
);
alchemy state bootstrap --backend gcp

AWS.state() now locks each stack and stage too, so two concurrent deploys of one stage can no longer mix their state (#2145). Thanks Arman Charan! The GCP docs are rebuilt around a five-part tutorial (#2374).

Docs: GCP · GCP setup.

Fly remote builds

Fly.Service images now build on Fly's remote builders, the same ones fly deploy uses (#2472). Only your bundled app leaves your machine: the builder pulls the base image, builds, and pushes to registry.fly.io right next to it. Deploys no longer upload the whole ~86 MB node:26-slim image for every Service, which made them slow on home and CI connections. Nothing to change in your stack:

// builds on a Fly remote builder
yield* Fly.Service("Api", { main: import.meta.url });

// build and push with your local Docker engine instead
yield* Fly.Service("Api", { main: import.meta.url, build: { remote: false } });

Docs: Fly services.

Also in this release

  • GCP.IAM.ServiceAccount and ServiceAccountKey resources (#2347).
  • Cloudflare.Hyperdrive.lookup (#1345) binds an existing Hyperdrive configuration without managing it. Thanks Franz Rebaza!
  • ctx.props and ctx.exports in Effect Workers and Durable Objects (#2251), and hibernation tags on Cloudflare.upgrade() (#2265). Thanks gegi!
  • Container placement jurisdiction and regions are typed (#2253), local containers can mount FUSE filesystems (#2362), and detached or adopted container applications redeploy cleanly (#2378, #2283). Containers built from inline Dockerfile content deploy (#2341), and monitor() fails with the same typed placement errors as start() (#2340). Thanks Jakub Sieczczyński, Jonathan Beckman, Yannic Charlon and Drew Powers!
  • localUploads on Cloudflare.R2.Bucket (#2424).
  • Cloudflare fixes: a failed ruleset read no longer wipes a zone's other rules when a Worker sets domain.redirects or version.affinity (#2339), WorkerEntrypoint targets a cross-stack Worker.ref (#2414), a rotated MtlsCertificate uploads the new PEM (#2428), and a Hyperdrive config replaces when its CA pin is dropped (#2429). Thanks Drew Powers!
  • Cloudflare website builds receive env, so NEXT_PUBLIC_*, VITE_* and similar values reach the client bundle (#2238), and a full Worker deploy returns its versionId and deploymentId (#2241). Thanks Kevin!
  • Local dev fixes: /.well-known/ assets serve (#2212), large D1 migrations and importFiles apply (#2380, #2258), publicDir files serve through ASSETS (#2290), StaticSite takes dev.port (#2331), tail consumers receive events and each Worker gets its own Cache store (#2383), port probes stay on loopback (#2237), and early responses no longer reset the connection (#2242). Thanks birbprophet, Altay, Kevin and Matthew Ruiters!
  • Cloudflare deploys on Node 26 use HTTP/1.1 for API calls, avoiding stream resets (#2155), and Durable Object RPC retries through deploy propagation (#2294).
  • metadataOptions on AWS.EC2.Instance and LaunchTemplate (#2426), Redacted userData on Instance (#2421), and a VPC whose create was interrupted is recovered instead of leaked (#2416).
  • AWS.RDS.Aurora accepts Outputs for kmsKeyId and migrations (#2364). Thanks gegi!
  • AWS fixes: Lambda.KafkaEventSource grants the EC2 permissions MSK mappings need (#2401), EMRContainers.VirtualCluster waits for a new EKS cluster's connection (#2412), and DataZone.EnvironmentBlueprintConfiguration removes the Lake Formation admins DataZone added (#2411).
  • ~200 more AWS resources run on floci in alchemy dev (#1803).
  • AgentCore Runtime V2 with platformVersion (#2301), ECS dev watcher keeps the recorded task definition (#2091). Thanks Henry Marks and TBIQ007!
  • Railway ephemeral environments (#2117) and a 768 MiB Prisma Compute upload cap (#2257). Thanks Mikkel and Aman Varshney!
  • GitHub.consumeRepositoryEvents generates a webhook secret when you omit one, so every delivery's signature is checked (#1932). GitHub.Ruleset takes allowedMergeMethods and requireExtraApprovalForUnattributedChanges (#2397) and adopts an existing ruleset by name (#2129), and GitHub.BranchProtection deploys with requiredStatusChecks.checks (#2076). Thanks Jason Kuhrt, Patt O'Brien and Arman Charan!
  • Core fixes: declaring a logical id already used by another type fails with DuplicateLogicalIdError (#2063), afterAll(destroy(Stack)) no longer hangs under Vitest (#2142), a Docker.Context missing from a fresh CI runner is re-created (#2187), log redaction no longer leaks the tail of a secret that extends a shorter one (#2418), tarballs keep long non-ASCII file names intact (#2420), and alchemy/State type-checks under exactOptionalPropertyTypes (#2413). Thanks Arman Charan and Patt O'Brien!
  • The CLI runs on Node 22.22.3+ again (#2403), and alchemy profile refresh --provider matches case-insensitively (#2410). Thanks Rahul Mishra!
  • alchemy profile no longer crashes with Cannot find package 'react' (#2295), and Workers bundle less core code (#2240). Thanks birbprophet and Juliaan!

Where to go next