plasma

What plasma is

plasma is a TypeScript sync engine you drop into a Cloudflare Workers + D1 stack. You define your schema and mutation functions once; they run:

• on the browser, optimistically, against IndexedDB — for instant reactive UI
• on the Worker, canonically, against D1 (or Postgres via Hyperdrive) — for the source of truth

plasma owns the /sync/* endpoints, the change log, the pull cookie, the outbox, the rebase, the WebSocket poke, the R2 blob uploads for file() columns, and everything else in between.

Isomorphic DSL

Types flow end-to-end. db.select().from(todos).where(eq(todos.done, 0)) means the same thing on the browser and on the Worker, and returns the same row shape.

Optimistic + convergent

Mutations apply locally the moment you call them. The sync loop pushes them to the server, rebases on top of concurrent changes, and — with CRDT columns — converges without conflict resolution boilerplate.

Files, encryption, CRDT

Declare a file() column and R2 handles the bytes. Declare .encrypted() and the client wraps at-rest. Declare crdtOrSet<string>() and concurrent tabs merge additions and removals losslessly.

Cloudflare-native

Ships with SyncCoordinator (Durable Object with hibernation-aware WebSocket fan-out), SequencerDO (per-region monotonic id source), Miniflare + workerd test harness, and R2 blob storage adapter.

When it fits

plasma is designed for the shape of app where each user has a manageable set of rows they own or share with a small group: todos, notes, chats, kanbans, drafts, contacts, small collaborative documents.

It is not designed for public feeds (Twitter timeline shape) or analytics-heavy workloads. The change log grows with writes, not with reads, so a heavy-read / light-write app is fine; a shape where a single row is fanned out to thousands of subscribers is not.

Next

• Install
• Quick Start — a todo app in 5 minutes
• Mental model — the three layers plasma is built on, and why they let you write one thing and get two runtimes.