Tiptap AI Toolkit - Empower your AI to directly edit documents in real time.
by•
AI Toolkit is now in production-ready beta — and we’re giving 100 builders a free lifetime license.
Chatbots are easy. Editing inside a rich-text doc in real time is deceptively hard. Skip months of engineering: Tiptap delivers a safe, reliable bridge between your AI and the documents where your team actually does its work.
Sign up for a Tiptap account and join our giveaway: 100 users get a free, lifetime AI Toolkit license (100k tool calls/mo) in exchange for feedback.
Replies
Tiptap
Hey Product Hunt! 👋
We’re a small team of EU-based developers who work on rich text editor frameworks every day. That's Tiptap.
A year ago we hit a problem we couldn't ignore: Everyone wants their AI to work inside documents, not in a chatbot and not with copy-paste. But building that so it actually works is way harder than it looks, because rich text is not plain text. There's a lot going on in how browsers render documents, and one wrong AI edit can break things: complex stuff like tables and charts, interactive stuff like buttons, or just the real-time collaboration happening while people work together.
So we spent months building Tiptap’s AI Toolkit to solve it once, properly, so you don't have to.
The AI Toolkit gives you:
Server-side editing: you don't need an open browser anymore. You can build agents that edit documents even when your users are away, and the changes flow back into their editors through our Collaboration service.
Document awareness: you can tell the AI to edit a specific paragraph or sentence and it knows where that is in the document. It can tell a table from a row from a header, and so on.
Precise edits: highlight any text in your document and ask the AI to change it.
Diff engine: a brand new diff algorithm that can compare structured documents.
This is great for anyone building a smart, AI-driven experience for documents (or document-like things) inside their app.
Check out our launch post at https://tiptap.dev/blog/release-... for more details, and a chance to get a free promo plan if you want to partner with us and give feedback.
Cheers,
Philip
@philip_isik "Rich text is not plain text" — I build AI text editing on macOS and feel that line in my bones. The structured-document diff engine is exactly the piece I'd never have wanted to build from scratch. How does it handle a human editing the same paragraph mid-AI-edit — does the diff reconcile against the live collab state, or snapshot at request time?
Tiptap AI Agent
@philip_isik @rudratosh The AI Toolkit integrates with collaboration, so every edit the AI makes is reconciled with human edits in the same way as two human users writing on the same document.
@philip_isik @arnau_gomez_farell The answer I was hoping for. Treating the AI as just another collaborative participant instead of a special case feels like the right abstraction. It saves every team from reinventing conflict resolution for AI edits. Looking forward to trying this with long-running document workflows. 👏
schema-validated output is exactly where most AI-edits-your-doc tools get sloppy, so that's reassuring. When generated content fails that validation, what happens: does it retry and repair the node, silently drop the edit, or surface the failure to the app so I can handle it? Mostly trying to gauge whether a bad model response can ever reach the collab doc at all.
Tiptap AI Agent
@hi_i_am_mimo Content is always validated, so invalid content never gets inserted into the document. If the AI generates invalid content, the toolkit will first try to repair it (if it's a trivial mistake, like forgetting to add text inside a paragraph) or it will return an error. This error can be handled by the developer or it can be fed back to the AI so it retries the edit.
Building a content tool on my own editor stack right now, so this hits close. The hard part I keep fighting: when AI edits a document live, how do you make the edit boundaries visible enough that a human reviewer trusts the diff? Curious how you handled that in the toolkit.
Tiptap
@vladimir_iudin There are multiple ways. From a UX/UI perspective, one would be to let AI make edits in suggestion mode, like in our demo here: https://template.tiptap.dev/page-ai-toolkit/
Tiptap AI Agent
@vladimir_iudin
That's a very good question. If the AI messes up the document, the user loses trust.
To tackle this, we first built the Smart Diff algorithm, which can accurately compare two documents. See it live here: https://tiptap.dev/docs/ai/ai-toolkit/client/advanced-guides/compare-documents
Then, we built the Tracked Changes extension. It allows you to display changes as suggested edits that users can accept or reject.
Finally, we set up the AI Toolkit so that, every time the AI makes a change to the document, we use the Smart Diff algorithm to find the places where the document changed, and then we use Tracked Changes to display each change as a suggested edit.
See a demo here: https://tiptap.dev/docs/ai/ai-toolkit/agents/tracked-changes
@arnau_gomez_farell Ok this answers it completely, thanks. Smart Diff finding the changes and Tracked Changes turning each one into accept/reject is exactly what client approval needs, the reviewer sees what the AI touched instead of trusting it blindly. Reading the docs now, my editor might end up on this after my launch settles.
Tiptap
Hey PH! When OpenAI shipped the Assistants API in 2023, we built a first prototype of AI agents editing documents. It technically worked, but the agent did things to the document, not with you.
Two and a half years later, this is the version where it finally feels like collaboration: agents make precise tracked changes users accept or reject, edits stream in live, and server-side operations keep running after the tab closes.
It's headless just like our editor: no UI, no required agent framework. You own your agentic loop and models and the AI Toolkit handles the document operations.
If you've tried building this yourself, I'd love to hear about your experience!
Plugged it into a side project and had a collaborative doc running by lunchtime. The extension API feels really intuitive compared to other editors I've wrestled with.
Tiptap AI Agent
@smet0b2o Very happy to hear that 😁
The server-side editing piece is what grabs me — agents that keep working after the tab closes is exactly the gap in most "AI in docs" setups. Question on document awareness: for a long, structured doc (nested tables, embedded nodes), what do you actually feed the model so it can target "this paragraph" reliably — a semantic/structural map of the doc, or the full serialized content? Curious how you keep that from eating the context window as documents grow.
Tiptap AI Agent
@hung_tran_from_notebook_os
To help AI understand the custom nodes of the document, the AI Toolkit generates a systemPrompt string that explains the different elements of the document and how they are encoded. You can attach this string to the system prompt. The systemPrompt string always has the same value, so you'll benefit from your AI provider's caching and save token costs.
For large documents, the AI Toolkit reads them chunk by chunk so the document does not overflow the context window. The chunk size can be configured by the developer. We're also looking into dedicated search tools, to help AI navigate long documents. All of this should help you keep token costs low.
The AI Toolkit unlocks a new frontier of collaborative working for your users, don’t miss out on the launch and give it a try now 🚀
Tiptap AI Agent
It's been a few months working on this, so I'm happy for the launch. This is just the beginning though. In the following weeks, we're going to post many videos and demos about what the AI Toolkit can do. Stay tuned for our updates on LinkedIn and our website.
And props to @bdbch for working on Tracked Changes. Without it, the AI Toolkit would not be possible.
Real-time AI edits on a live document are a hard distributed systems problem. You're essentially layering an AI agent on top of CRDT or OT-based sync, and conflicts aren't always commutative. We've thought about this pattern when building structured data capture, and the ordering of writes matters a lot. How does the toolkit handle AI and human edits that conflict on the same node simultaneously?
Tiptap AI Agent
@anand_thakkar1 The AI Toolkit integrates with Tiptap collaboration and version history. Any conflicting edit made to the AI works the same way as edits made by two users simultaneously to the same document. If you're interested in using the AI Toolkit but and you have certain requirements you'd like it to cover for your application, write to us at humans@tiptap.dev and we'll work to make sure these requirements are met.
the reconciliation and diff answers cover the single-agent case really well, curious about the multi-agent one: if I've got several different AI agents with different jobs (one summarizing, one restructuring, one fact-checking) all hitting the same doc through the server-side API, is there any way to scope an agent to only certain nodes or sections, or does anything with API access get edit rights to the whole document and you just have to trust your own orchestration to keep them in their lane
Tiptap AI Agent
@galdayan In the current version, the AI can edit any node in the document.
One of the items in our roadmap is to allow developers to define allowed/protected nodes, to restrict what content the agent can modify.
We'd like to know more about your requirements. You can contact us at humans@tiptap.dev and if it makes sense, we can add this feature to our product
The part that resonates is treating rich text as not plain text - an AI that blindly string-replaces will happily corrupt a table or a live collab cursor. Concretely: are the AI edits applied as ProseMirror transactions/steps so they merge with concurrent user edits through your collab layer, or as a diff that can clobber in-flight changes? And is the model constrained to your schema so it cannot emit invalid nodes, or do you validate and repair after the fact?
Tiptap AI Agent
@hi_i_am_mimo Thanks for your questions Valeria. Edits are converted into transactions and sync with the collab server. All AI output is validated so that generated content conforms to the schema.