@web-ai-sdk/rewriter

This package wraps the Web's Built-in Rewriter API. It changes the tone or length of text. It also provides session reuse, streaming, and optional result caching.

Docs: https://web-ai-sdk.dev/docs/guides/rewriter/ · React: useRewriter · Production: Checklist

Status

Chrome labels Rewriter a Developer trial in its status table. The public origin trial for Chrome 137–148 has ended. See the Rewriter guide for current localhost flags.

Edge provides a Canary/Dev preview from 138.0.3309.2. Enable "Rewriter API for on-device language model."

Without Rewriter, React reports "unavailable" and rewrite() throws RewriterUnavailableError.

Install

pnpm add @web-ai-sdk/rewriter
# or: npm i @web-ai-sdk/rewriter / bun add @web-ai-sdk/rewriter

The React adapter uses the /react subpath. react is an optional peer dependency.

Vanilla TypeScript / DOM

import { rewrite } from "@web-ai-sdk/rewriter";

const result = await rewrite({
  input: "hey, can u send me that doc when u get a sec? thx",
  tone: "more-formal",
  length: "as-is",
  onUpdate: (text) => console.log("partial", text),
});

console.log(result.output, result.cached);

result.output is the rewritten text (trimmed), or null when the input is empty.

React

import { useRewriter } from "@web-ai-sdk/rewriter/react";

export function Polish({ draft }: { draft: string }) {
  const { status, output } = useRewriter({ input: draft, tone: "more-formal" });

  if (status === "unavailable") return null;
  if (status === "loading") return <p>Rewriting…</p>;
  return <p>{output}</p>;
}

State machine: idle | loading | streaming | done | unavailable. output is the latest text (grows during streaming). fromCache is true when the result came back without invoking the model.

API

rewrite(options): Promise<RewriteResult>

interface RewriteOptions {
  input: string;                  // text to rewrite
  context?: string;               // per-call background info
  language?: string;              // BCP-47; drives input/output hints when supported
  supportedLanguages?: readonly string[]; // default ["en", "es", "ja"]
  tone?: "as-is" | "more-formal" | "more-casual";   // default "as-is"
  format?: "as-is" | "markdown" | "plain-text";     // default "as-is"
  length?: "as-is" | "shorter" | "longer";          // default "as-is"
  sharedContext?: string;
  monitor?: (m: CreateMonitor) => void;
  cache?: "session" | "local" | { get, set };
  cacheKey?: string;
  cacheTtl?: number;      // built-in shortcut TTL in ms; default 1 hour
  cacheRefresh?: boolean; // skip the cache read, write the fresh result
  onUpdate?: (text: string) => void; // cumulative buffer, not deltas
  signal?: AbortSignal;
}

interface RewriteResult {
  output: string | null;
  cached: boolean;
}

isAvailable(): boolean

Feature-detect helper.

checkAvailability(options?): Promise<RewriterAvailability | null>

Forwards to the spec's availability() call. Returns null if the global is missing or the call throws.

Cache controls

import {
  clearRewriterSessions,    // drop every cached rewriter session
  clearRewriterSession,     // drop one cached session by create-options
  configureRewriterCache,   // change the LRU cap (default 8)
} from "@web-ai-sdk/rewriter";

The internal session cache is LRU-bounded (default 8). Evicted sessions have their destroy() invoked when present. Clearing detaches sessions pinned by a lease or an in-flight call and destroys them when the last pin drops.

Prepare and release

prepareRewriter(options) starts native session creation when user intent is clear, before the input exists. It returns a RewriterLease:

interface RewriterLease {
  ready: Promise<void>; // settles when native creation settles
  release(): void;      // idempotent
}

import { prepareRewriter, rewrite } from "@web-ai-sdk/rewriter";

// User opened the polish panel; warm the session now.
const rewriterModel = prepareRewriter({ tone: "more-formal" });

// The matching call reuses the prepared session with no second create.
const result = await rewrite({
  input: "hey, can u send me that doc when u get a sec? thx",
  tone: "more-formal",
});

// User dismissed the panel.
rewriterModel.release();

• prepareRewriter never throws synchronously. Unavailability and creation failure reject ready with RewriterUnavailableError.
• release() is idempotent. The final release destroys the session once no other lease or in-flight call uses it.
• Releasing before creation settles destroys the session after creation succeeds.
• Failed creation evicts the entry, so a later prepare retries.
• Sessions with active leases never evict from the LRU cache.

Reuse requires the same session-affecting options as the rewrite call. PrepareRewriterOptions covers language, supportedLanguages, tone, format, length, and sharedContext. monitor observes creation only and never affects reuse.

Result caching

Off by default; every call hits the model. Pass cache: "session" for sessionStorage, cache: "local" for localStorage, or any { get, set }-shaped object for a custom backend.

Expiry and refresh

The built-in "session" / "local" shortcuts store each entry in a versioned envelope with an expiry time. Entries expire after one hour (DEFAULT_CACHE_TTL_MS) by default. Pass cacheTtl (milliseconds) to override the TTL per call. Expired entries, legacy raw strings, and malformed envelopes count as misses and are removed.

Pass cacheRefresh: true to force a fresh inference. The call skips the cache read, runs the model, and replaces the cached value after a successful run. Failed, aborted, or empty runs leave the cached value in place.

Custom { get, set } caches own their expiry policy. cacheTtl does not apply to them; cacheRefresh still bypasses their read and updates them after success.

// Cache for five minutes instead of one hour.
rewrite({ input: text, cache: "local", cacheTtl: 5 * 60 * 1000 });

// Force a fresh inference; later calls reuse the new value.
rewrite({ input: text, cache: "local", cacheRefresh: true });

Output normalization

The wrapper trims leading/trailing whitespace only, so internal markdown formatting and line breaks the model produces stay intact.

License

MIT © Beto Muniz