Open source · self-hosted

The reactive backend you run yourself.

Write your backend as TypeScript queries and mutations. Subscribe once, and helipod pushes fresh results to every client the moment the data changes. No polling, no refetching.

Read the quickstart →How it works

helipod/messages.tsreactive

// a query is a pure, reactive read
export const list = query({
  args: { channelId: v.id("channels") },
  handler: (ctx, { channelId }) =>
    ctx.db.query("messages", "by_channel")
      .eq("channelId", channelId).collect(),
});

// on the client, re-renders when a mutation commits
const messages = useQuery(api.messages.list, { channelId });

the reactive core

Data changes push themselves

Every query records what it read; every write is checked against it.

01A client subscribes. The engine records its read set.

02A mutation commits in one serializable transaction.

03The result is pushed. Only queries the write touched re-run.

Read about reactivity

8.6ms

Median reactive propagation in a same-substrate benchmark. Run it yourself before you trust it.

See the numbers

components

Auth, scheduler, workflows

Opt-in components: authentication, cron and scheduled jobs, durable multi-step workflows with saga compensation, triggers, and notifications.

storage

SQLite or Postgres, plus files

Zero-config SQLite for local, Postgres when you need it, and built-in blob storage on the filesystem or any S3-compatible bucket.

the client

Optimistic and offline

A typed client with instant optimistic updates and a durable offline outbox that survives reloads, with exactly-once delivery on reconnect.

escape hatches

Actions, HTTP, crons

Actions run outside the transaction for fetch, timers, and randomness. Public HTTP endpoints handle webhooks.

dashboard

A live data browser

Browse and edit tables, watch logs, and run functions from a built-in dashboard that updates reactively.

your infrastructure

Runs where you run

One command in development. In production it is a single self-contained binary, a Docker image, or a Cloudflare deployment, backed by your own SQLite file or Postgres. No managed cloud in the loop, no vendor lock-in.

single binarydockercloudflarepostgres

Self-hosting guide

honest limits

What it doesn't do yet

No searchQuery by index and range. Full-text and vector are reserved seams.
Single-node writesOne writer per shard. Multi-node scale-out is a later tier.
No built-in TLSPlain HTTP. Front it with nginx, Caddy, or Traefik.
In-process functionsA V8-isolate sandbox for untrusted code is a reserved seam.

Get started

Write your first reactive function in a few minutes.

The quickstart takes you from an empty folder to a live, reactive app.

Start building →