hybrid-compute · local, worker, remote

Keep the main thread free.
Run each task where it fits.

Register a task once and let runTask() send it to the main thread, a Web Worker or a remote service over HTTP or WebSocket. Your code stays the same; only the strategy changes.

Get started API reference GitHub

Frames

Main thread

runTask('resizeImage') blocks the main thread

Worker

idle

Remote

idle

Runs on

local main

Main thread

Blocked

Frames

Dropped

strategy 'local', and the local backend can run it

runs on the main thread

Illustrative timeline, not a measurement.

Packages

Use core with any mix of the three backends.

Package	Runs on	What it does
@hybrid-compute/core	anywhere	The orchestrator. Picks a backend for each task by strategy; with auto it falls back from worker to local to remote.
@hybrid-compute/local	main	Runs registered tasks on the main thread. The right place for small, cheap work.
@hybrid-compute/worker	Web Worker	Runs tasks in a Web Worker, so heavy work never blocks rendering or input.
@hybrid-compute/remote	remote service	Sends tasks to a service over fetch or WebSocket, as JSON messages.

One call, any backend

Change the strategy, not the code.

// app.ts
const compute = createHybridCompute({ local, worker, remote });

await compute.runTask('resizeImage', blob);            // auto: worker, local, then remote
await compute.runTask('resizeImage', blob, 'worker');  // always off the main thread
await compute.runTask('summarise', text, 'remote');  // on your server

Why route tasks

A UI that never stutters

Parse, resize and transform in a worker; keep the main thread for input and rendering.

Data that stays on the device

Run sensitive steps locally and send only what a remote service really needs.

Works offline

Prefer local and worker backends, and reach for remote only when it is there.

Keep the main thread free.Run each task where it fits.