New
Built-in provider resolvers
One-line adapters for OpenAI, Anthropic, and Azure OpenAI. Bring your own key; keys never leave your server. Also works with OpenAI-compatible endpoints (Groq, Together.ai, local Ollama) via a BaseUrl override.
AI Toolkit
Bring AI where content happens.
Ask AI, AI Chat, inline preview, and a reviewable suggestion drawer - all wired into the editor. Writers stay in the document, reviewers keep control, and engineers swap in their own resolver when they are ready.
Live preview
Ask AI, suggestion, review - all inline.
RichTextEditor
Our new release improves performance by 34% while shipping a smaller bundle.
AI rewrite · proofread
The new release improves performance by 34% and ships with a smaller bundle.
ApplyAcceptReject
In the base license
The AI Toolkit at a glance
Everything below ships in the base license. No premium tier, no usage credits to buy. See the full release notes for version-by-version detail.
New
Streaming responses
Tokens render as they arrive — same feel as ChatGPT / Claude. SSE-based wire format; editor.aiToolkit.streamRequest({ onDelta, onDone }) on the client, with XHR fallback for old browsers. Abortable mid-stream.
New
DOCX export
Server-side HTML → .docx via MIT-licensed DocumentFormat.OpenXml. Client helper editor.aiToolkit.exportDocx() triggers a browser download. No cloud SaaS subscription.
New
Dictation (Web Speech)
Toolbar mic button; interim transcripts render live under the cursor. The only major editor to ship dictation out of the box. Opt-in via dictationEnabled: true.
New
Shared review ledger
AI suggestions + human Track Changes + threaded comments all flow through a single editor.reviewLedger store. One drawer, interleaved entries, accept / reject regardless of source. Competitors silo these in separate plugins.
Preview
Yjs concurrent typing
Opt in with textSync: true to bind the editor HTML to a shared Y.Text. CRDT merges concurrent inserts / deletes at character level. Per-node binding is next on the roadmap.
Three surfaces, one toolkit
Meet the writer where they already are.
Different editing tasks want different surfaces. Ask AI for quick actions, Chat for a conversation about the document, Review for batch decisions. They share state, so what you preview in one is reviewable in the others.
Ask AI
Toolbar-first dialog. Proofread, rewrite, shorten, expand, summarize, or translate the selection or the whole document.
Ask AI · proofread
SOURCE
hello world ,bad spacing
RESULT
Hello world, bad spacing.
AI Chat
Docked panel for multi-turn work. Scope-aware: uses the selection or the whole document, with quick prompts built in.
AI Chat · document
Summarize this for the release notes.
Drafted a 2-line recap. Apply to the document?
Tighten the tone to be more neutral.
Revised the recap, neutral phrasing. Ready to apply.
AI Review
Queue of pending, accepted, and rejected suggestions. Preview before applying, batch-accept or reject, or sync with your server ledger.
AI Review · 3 pending
Proofread intro paragraphPending
Translate heading (Spanish)Accepted
Shorten section 2Pending
Why it matters
AI suggestions should stay inside the editing workflow.
Instead of copying content to a separate AI window and pasting results back, RichTextEditor keeps AI actions attached to the document. Writers can select text, ask for a rewrite, preview the result inline, and review suggestions without leaving the editor.
For product teams
Keep users in your app with an editor-native AI experience that feels integrated, reviewable, and consistent with your content workflow.
For engineers
Start with the built-in demo resolver, then swap in your own model or provider when you are ready. Same UI, new backend.
Quick actions
Proofread · rewrite · translate
Seven one-click modes plus your own: proofread, rewrite, shorten, expand, summarize, translate, justify. Each returns a rationale.
Structured plan
Operation contract
Resolvers return one or many operations (preview, replace, insert, comment). Users apply step-by-step or as a batch.
Inline preview
Accept or reject in place
Before/after view with change highlighting, operation summary, and accept/reject controls. Nothing commits until a human says so.
Persistent ledger
Review drawer + sync
Pending, accepted, and rejected suggestions persist per editor. Hook a remote ledger endpoint to sync across sessions.
Scope control
Selection or document
Every action knows what it is acting on. Selection-aware by default, with a one-tap switch to whole-document scope.
Multi-turn chat
Conversation with context
Docked panel with quick prompts, message history, and copy / apply / insert / replace actions on every AI response.
Custom actions
Register your own modes
Use registerAction to add domain-specific actions (legal clause check, SEO rewrite, tone matcher). They plug into the same UI.
Demo mode
No API key required
Built-in deterministic resolver ships with the plugin. Evaluate the UX first, wire a real model later - same front-end.
Developer surface
Bring your own resolver in ~15 lines.
The aiToolkitResolver hook receives a structured request (action, source text, scope, snapshot) and returns a result or a list of operations. The editor handles rendering, preview, and review for you.
JavaScriptaiToolkitResolver
var editor = new RichTextEditor("#editor", {toolbar: "full",aiToolkitResolver: async function (request) {// request: { mode, source, scope, snapshot, language, ... }const reply = await fetch("/api/ai", {method: "POST",body: JSON.stringify({mode: request.mode,text: request.source})}).then(r => r.json());return {result: reply.text,reason: reply.explanation,operations: [{ type: "preview-suggestion",text: reply.text }]};}});
Pair with structured content
AI gets stronger when the document model is structured.
Use the structured content bridge alongside AI Toolkit to move between HTML, semantic JSON, Markdown, static HTML rendering, and validation. That makes it easier to save, inspect, render, and process AI-assisted content in real applications.
Ready to evaluate?
Start with the demo, then connect your backend.
Open the AI Toolkit demo in a new tab, run the built-in resolver locally, and decide how you want to wire a real provider. No account, no download.