Migrate infra to alchemy v2 (Effect-based Stack)

[Makisuo]

What

Migrates the entire deploy layer from alchemy v1 (^0.93.12) to alchemy 2.0.0-beta.61 (the Effect-based rewrite), following https://v2.alchemy.run/migrating-from-v1/ in the full-Effect style:

• Root alchemy.run.ts is now a single Alchemy.Stack("maple", …) Effect program; the seven app files (api, alerting, chat-flue, electric-sync, web, landing, local-ui) became factories returning Effect.gen programs. v1→v2 mechanics applied throughout: entrypoint→main, bindings→env, alchemy.secret→Redacted.make, compatibility: "node"→{ date, flags: ["nodejs_compat"] }, per-resource adopt: true→ deploy-wide --adopt.
• Binding names are preserved verbatim (verified against v2 source) — no worker source changes anywhere.
• Deploy order change: chat-flue deploys before api, which service-binds it directly (CHAT_FLUE); the v1 WorkerStub cycle-breaker is deleted. Enabled by giving pr stages static custom domains (api-pr-N/chat-pr-N/sync-pr-N.maple.dev in stage.ts) because v2 resource attributes are lazy Outputs that can't be string-interpolated into env.
• web/landing/local-ui deploy as Command.Build + Worker(assets: {directory, hash}) — all three are static SPA builds with a tiny assets-fallback worker, so v2's Vite pipeline isn't needed. Web's VITE_* env flows through the Build's memo hash, so per-stage URL changes re-run the build. The v1 alchemy/cloudflare/vite plugin (and the now-unused @cloudflare/vite-plugin dep) are removed.
• api: Hyperdrive.Connection (origin from MAPLE_PG_URL), KV, queue + standalone Queues.Consumer (same batch settings), workflows keyed by physical name; alerting keeps its cross-script workflow ref via scriptName.
• chat-flue: Command.Build replaces inline execFileSync; bundle: false + module rules for the prebuilt Flue bundle; the AI binding now comes from a Cloudflare.AI.Gateway resource (v2 has no bare Ai() descriptor for async workers).

Why

v2 is the maintained line. v1 state is incompatible with v2 — cutover per stage is: abandon v1 state, redeploy with --adopt (identical physical names re-adopt live resources). Never run a v1 alchemy destroy against a live stage.

Reviewer notes

• prd is intentionally blocked: v2 has no HyperdriveRef equivalent for the dashboard-managed maple-prd Hyperdrive. The root stack fails fast on --stage prd; prd deploys keep using a v1 checkout until that gap closes (or we adopt maple-prd as a managed Connection with creds in the prd deploy env).
• effect patch extended (patches/effect@4.0.0-beta.93.patch): bun's isolated linker gives the alchemy CLI and the stack program two path-distinct effect module instances; effect v4's redactedRegistry is a module-local WeakMap, so every Redacted env binding crashed with "Unable to get redacted value". The patch shares the registry via a Symbol.for-keyed global (what effect v3's globalValue did). Candidate for upstreaming to effect-smol.
• Secrets/state: ALCHEMY_PASSWORD/ALCHEMY_STATE_TOKEN are dead. State is v2's Cloudflare state store (one-time alchemy bootstrap cloudflare per account, or first deploy --yes); ALCHEMY_LOCAL_STATE=1 opts into local file state for dev experiments. CI auth is CLOUDFLARE_API_TOKEN + CLOUDFLARE_ACCOUNT_ID; the legacy CLOUDFLARE_DEFAULT_ACCOUNT_ID name from Infisical is bridged in the root stack.
• Deploy scripts gained --yes (v2 prompts otherwise); the pr deploy also passes --adopt. CI workflow YAML needed no changes (env comes from Infisical).
• New tsconfig.alchemy.json typechecks all eight alchemy.run.ts files, wired into bun run typecheck.

Validation

• bun run typecheck green (29 turbo tasks + the alchemy tsconfig).
• CI=true ALCHEMY_LOCAL_STATE=1 bunx alchemy deploy --dry-run --stage dev-claude --env-file .env.local plans all 18 resources to create (5 workers, 2 SPA workers, 3 builds, Hyperdrive, KV, queue + consumer, 2 workflows, AI gateway).
• Not yet deployed to any cloud stage. First real validation should be a pr-N preview deploy (state-store bootstrap required first), checking: queue consumer settings in the dashboard, both workflows triggerable, chat-flue DO roundtrip + env.AI runtime behavior, api→CHAT_FLUE service call, alerting cross-script workflow start, and PR-close destroy not touching stg/prd zone objects.

🤖 Generated with Claude Code

---

[pullfrog]

Your Pullfrog Router balance is exhausted.

You have a card on file but auto-reload is disabled, so runs paused once your balance went past the overdraft buffer.

Top up balance → · Enable auto-reload →

  ^{｜ ⚠️ this action is pinned to a commit SHA, which freezes the cleanup step — switch to @v0 or keep the SHA fresh with Dependabot ｜ Rerun failed job ➔ ｜ View workflow run ｜ via Pullfrog ｜ Using Claude Opus ｜ 𝕏}