What Alchemy offers that we don't use

Research, October 2026. Sam Goodwin (Alchemy's author), on hearing we use it as our deployment factory:
"Alchemy is about much more than deployment". This maps that "much more" against what
packages/cli/src/gen/alchemy.ts generates today. Sources: the package we install (alchemy@2.0.0-beta.77, read
under node_modules/.bun/alchemy@2.0.0-beta.77+*/node_modules/alchemy/src), the docs at
alchemy.run (also served to agents as llms.txt), and the
release notes.

The short answer: v2 is an Effect application framework, and deploying is one part of it. It also runs the app
locally against emulators, tests it against real throwaway stages, gives every PR its own preview, notices
when the account drifts, and ships libraries that run on Workers: a git server, Better Auth, typed RPC
between Workers. Everything below is beta; the README says "Expect breaking changes".

What we use now

The generator writes .rustybuns/alchemy.run.ts from rustybuns.config.ts, and the CLI runs alchemy plan,
deploy, destroy and dev on it. In that file:

• Cloudflare: Worker (domain, routes, crons, observability, placement, limits, assets, workersDev),
  DurableObject, D1, KV, R2, Queues (queue and consumer), Vectorize (and metadata indexes),
  Hyperdrive, RateLimit, Images, Email.SendEmail, Container, Artifacts, Workflow, WorkerLoader,
  AnalyticsEngine, Workers.AI, InferEnv for the typed env.
• Elsewhere: Command.Build, Hetzner.Server/Volume/Service, Railway.Project/Service/Volume,
  Neon.Project/Branch, Alchemy.makeRandom for minted secrets.
• Policies: AdoptPolicy.adopt, RemovalPolicy.retain (protect: true), renamedFrom.
• State: Alchemy.localState(), which packages/cli/src/state.ts symlinks into the git common dir and guards
  with its own lock so every worktree on one machine sees one state per app.

That is the classic async-Worker style: our app code never imports Alchemy or Effect. Most of what follows
keeps it that way; the rows that don't say so.

Before anything else: the version

We pin 2.0.0-beta.77; npm's latest is 2.0.0-beta.84 (October 24, 2026). beta.84 is a
breaking package split: Cloudflare, AWS,
Git, Drizzle and others moved to their own @alchemy.run/* packages, so the generator's
import * as Cloudflare from "alchemy/Cloudflare" changes, as does the pin in packages/cli/src/glue/deploy-deps.ts.
beta.78 also took Alchemist out of the root export. The bump is its own PR, with a plan diff on codeSplitter's
stack that shows no replaces.

beta.77 also lists seven exports in package.json with no file behind them (Construct, ContentType,
Endpoint, Process, TUI, Util/Clank, Cloudflare/Live); nothing there to adopt. Nothing called Clank
appears in the repo or docs either.

Worth taking, ranked

Also there, no use yet

• Libraries that run on Workers: @alchemy.run/better-auth (Better Auth with a D1 layer, six-part tutorial),
  @alchemy.run/domain (commands, aggregates, events; beta.84), Effect AI LanguageModel, AI Search, Effect SQL.
• Cloudflare: Pipelines, Browser, Flagship (feature flags), Stream, SecretsStore, Pages,
  Website.* (Vite, StaticSite, Astro, Next, Nuxt, SvelteKit, TanStack Start), WorkersForPlatforms, Tunnel,
  LoadBalancer, WAF Ruleset, Alerting, Email Routing, R2 event notifications, MCP Portal.
• Other providers: AWS (with local emulation), GCP, Docker, Kubernetes (Helm, kind), Fly, PlanetScale,
  Prisma Postgres, Turso, Stripe, and the rest of Hetzner (Firewall, Network, DNS) and Railway (Postgres, Redis,
  Bucket, domains). Neon branches per PR go with Fix desktop builds on Windows for apps with "use server" actions #3.
• Data: Drizzle and Prisma with committed migrations applied on deploy, SQL.Migrations, Neon.Migrations.
• Tooling: Alchemist (the CLI as an Effect API: Stack, Drift, Logs, State; the hook for an agent or the office
  to drive deploys without shelling out), the alchemy-test runner (agent-friendly output, one Bun process),
  alchemy state list|read, alchemy profile, KeyPair, custom providers and "Actions" for anything missing.
• alchemy unsafe nuke deletes every resource the providers can list, whatever the stack or state. It is hidden
  from --help so agents don't find it; ours must never run it (--local only wipes the emulators).

Things to know either way

• Telemetry is on by default. The CLI sends a random id, hashes of the git origin and branch, the repo's root
  commit SHA, and the command, resource types and operations to otel.alchemy.run (no secrets, values or source).
  DO_NOT_TRACK=1 or ALCHEMY_TELEMETRY_DISABLED=1 turns it off, and our sandbox Dockerfiles already set
  DO_NOT_TRACK; rustybuns doesn't when it runs alchemy. The Cloudflare state store's Worker sends its own
  signals even then. (privacy)
• Agent mode. The CLI notices CLAUDECODE, CURSOR_AGENT and CODEX_CLI, prints plain output and needs
  --yes to apply. alchemy.run/llms.txt and every page as .md are written for agents.
• No hosted service, no pricing. Everything runs in our own accounts; Alchemy is open source.

Suggested order

1. Bump to beta.84 alone (#2235), with the import rename and a plan diff on codeSplitter's stack. Set
   DO_NOT_TRACK=1 in the env rustybuns gives alchemy, or say in the docs that it's on.
2. rustybuns drift (updating for hetzner and readme #1), #2236: small, read-only, and it shows whether the bump or anything else moved the account.
3. postgresState on the Neon we already make (Restore desktop.host, desktop.headers, desktop.native #2) (#2237), then Worker Previews per PR (Fix desktop builds on Windows for apps with "use server" actions #3) (#2238) on top.
4. rustybuns test on the test harness (add plans for things CF just released! #4), run against those preview stages.
5. Read @alchemy.run/git (update init #6) next to codeSplitter's merge code and file what's worth taking.
6. The rest as the apps need them. Every row that creates resources costs money at scale; price it first and
   keep the smallest tier (packages/cli/src/costs.ts, COSTS.md).