Browser runtime

Import the browser-safe facade from smthrs/browser. It runs the real Smithers driver, scheduler, renderer, graph extractor, schema validation, and task dependency semantics in a lean, Node-free bundle.

import React from "react";
import { z } from "zod";
import {
  Task,
  Workflow,
  createBrowserRuntime,
  createBrowserSmithers,
  defineBrowserWorkflow,
} from "smthrs/browser";

const resultSchema = z.object({ answer: z.number() });
const workflow = defineBrowserWorkflow(() =>
  React.createElement(
    Workflow,
    { name: "browser-example" },
    React.createElement(
      Task,
      {
        id: "answer",
        output: "results",
        outputSchema: resultSchema,
        agent: {
          async generate() {
            return { answer: 42 };
          },
        },
      },
      "Return the answer.",
    ),
  ),
);

const runtime = createBrowserRuntime();
const smithers = createBrowserSmithers({ workflow, runtime });
const run = await smithers.run({ input: {}, runId: "browser-demo" });

console.log(run.status); // "finished"
console.log(await smithers.getOutputs(run.runId));

createBrowserSmithers creates a fresh driver and scheduler session per run() call. getRun(runId) and getOutputs(runId) read through the selected storage adapter. runBrowserWorkflow(workflow, options) is the one-shot equivalent when you don’t need to retain the wrapper.

Browser-safe exports

Export	Purpose
`defineBrowserWorkflow(build, options?)`	Define a portable workflow from a context-to-React builder.
`createBrowserRuntime(options?)`	Create the default browser `RuntimeAdapter`.
`createBrowserSmithers({ workflow, runtime?, runtimeOptions? })`	Create a reusable runner with run and output lookup methods.
`runBrowserWorkflow(workflow, options?)`	Run one workflow once.
`Task`, `Workflow`, `Sequence`, `Worktree`	Browser-safe workflow primitives.
`RuntimeCapabilityError`, `RUNTIME_CAPABILITY_UNAVAILABLE`	Typed failure for unavailable host capabilities.
`RuntimeAdapter`, `BrowserRuntimeOptions`, `BrowserWorkflow`, `BrowserRunOptions`, `BrowserSmithers`	Public TypeScript contracts.

Runtime options

The default runtime uses Date, performance, abortable setTimeout, Web Crypto UUIDs, and Map-backed run/output storage. Override any portable seam:

type BrowserRuntimeOptions = {
  clock?: RuntimeClock;
  storage?: RuntimeStorage;
  uuid?: () => string;
  executeTask?: (
    task: TaskDescriptor,
    context: TaskExecutorContext,
  ) => Promise<unknown> | unknown;
};

Provide storage when state must outlive a page refresh, and executeTask to route execution through an application-owned worker, provider client, or policy boundary. The default executor supports static and compute tasks plus in-process agents exposing generate, execute, run, or call, and validates Zod-like outputSchema values before storing outputs.

Capability boundary

The browser adapter fails closed for filesystem, subprocess, worktree, and sandbox: calling one of those throws a RuntimeCapabilityError with code RUNTIME_CAPABILITY_UNAVAILABLE, plus the runtime name, capability, and operation. The exported Worktree primitive is portable at graph-render time, but a browser run can’t resolve a real worktree unless the application supplies a custom RuntimeAdapter that owns that capability. The browser facade omits Node database factories, server helpers, CLI agents, subprocess code, bundled tools, and the full Node component barrel. Use the main smthrs entry point for those surfaces.

Source browser engine · Tests browser runtime tests · See also Run workflow, Testing workflows, Package configuration

Overview

Core API

Runtime

Types & errors

Gateway RPC

Browser-safe exports

Runtime options

Capability boundary

​Browser-safe exports

​Runtime options

​Capability boundary

Browser-safe exports

Runtime options

Capability boundary