We built the Tiptap AI Agent to make it easy to integrate AI into a rich text editor, without rebuilding your entire frontend.
If you’ve ever tried wiring up AI inside a document editor, you’ve probably dealt with:
- Extracting context from complex document structures
- Handling prompt input + streamed output
- Supporting undo/redo for AI changes
- Designing UI for accepting/rejecting changes
- Multiplayer session state and conflicts
This new Tiptap toolkit gives you a clean way to define AI Agents that can read and edit rich text based on user-defined tasks. You can trigger agents manually, automatically, or in response to structured input.
Works with OpenAI or your own backend + LLM stack. Built on top of the same multiplayer engine behind Tiptap’s collaboration features.
We also include an AI Changes extension so users can review and accept/reject generated edits, like a built-in code review for content.
There’s a live demo here: https://ai-agent.tiptap.dev/
Developer Docs: https://tiptap.dev/docs/content-ai/capabilities/agent/overvi...
Happy to answer questions or just hear what you’re building :-)
Looks very good! Are these new components exclusively available through the Pro package, requiring at least the paid Starter plan? Are any of the new features open source, or is all of the new functionality restricted to the paid tiers?
Thanks for the positive feedback! You're right. The AI features are part of our paid offerings.
We're currently working on new open source features for the editor core, such as Markdown support, a decorations API, and editor content migrations: https://tiptap.dev/tiptap-editor-v3
We also recently released Hocuspocus V3 (https://github.com/ueberdosis/hocuspocus/releases/tag/v3.1.0).
Thanks for the response! I have a follow-up / suggestion. Looking at pricing, it seems that the only way to get all the Content AI frontend components is to subscribe to the Team account at $149 per month. This includes 5000 documents in Tiptap cloud. For companies that only want to integrate the Editor without Collaboration (only frontend, we need full control of the documents on our backend, and also directly call the LLM ourselves), do you plan to offer any options? Thanks
Great question! We've created a page explaining what counts as a cloud document: https://tiptap.dev/knowledge/what-counts-as-a-document. As long as you don't sync editor documents to our cloud services, you're good to go. The AI extensions can only work on the client side. Hope that helps :-)
From the page:
What does not increase your limit:
Some of Tiptap’s features and common usage patterns do not create new documents. Here are a few scenarios that do not increment your overall count:
Integrating the Editor without Collaboration: You can embed Tiptap’s Editor in your application for local or front-end use without creating any Tiptap Cloud documents. As long as you don’t store documents in the Tiptap Cloud, your count remains unchanged.
Using Content AI without Collaboration: Content AI features, such as autocomplete or content generation, do not require collaboration. If you don’t create new documents or store your AI-generated content in the Tiptap Cloud, you avoid increasing your document total.
Importing other file formats: You can import any number of files in formats like DOCX or MD. This action alone doesn’t raise your count. You only add to the count once you save them as Tiptap documents in the Tiptap Cloud.
Comments and Revisions: Comments and revisions live in the document’s metadata. They never raise your document count, but you must sync the document to the Tiptap Cloud to use them.
If used in a legally binding setting to does your firm accept liability for erroneous output resulting in litigation proceedings or financial loss? Serious question because I’m a human working in a similar niche.
If not, how do you phrase that in your ToS and what is your Errors and Omissions policy language like?
We don't know the use case and content... We're providing the tools to build the editor-AI-UX, which you can then connect to the LLM of your choice.
In our AI extensions, we provide callbacks like onSuccess that you can hook into, to send analytics events and track the AI's responses.
I don't work for Tiptap, but I'm curious as to what your expectations are here.
They're providing tools for integrating Tiptap with AI, including ones for reviewing suggested changes that come from LLMs. I don't really see something that Tiptap should be accepting liability for. As the developer working with Tiptap, I'd argue it's your responsibility to make sure (a) the prompts are giving generally good and useful output, and (b) that you incorporate any necessary UI to put a human in the loop of accepting the AI output. On the latter point, their AI Changes extension is one such example, but there are any number of ways of achieving this.
I gotta say, the focus on all that non-model-specific heavy lifting for AI in editors? That's the real MVP move here. So much of the AI hype is just about the model, but integrating it smoothly is where most projects probably hit a wall.
Agreed, adding Cursor-like AI features to a rich text editor is way harder than it should be... You have to handle prompts, context management, streaming, error states, UI, undo/redo, multiplayer sync, AI UX, it all adds up fast.
If using tiptap as a collaborative editor, are the pending AI Changes (the red/green diff applied to the editor contents) visible to all users, or just the current one?
Ideally it's the latter, but it's non-obvious to me.
In this first version of the AI Agent, the changes are visible to all users as soon as the AI Agent inserts them into the document. However, we are going to implement the other option if there is demand for it. It is in our backlog.
Huh I really like the inline accept / reject button upon clicking
would be nice to customize the diff view colors and button styles
will play with the integration. Very cool stuff!
Thanks for your feedback! It's highly customizable. Feel free to let us know if you need any additional options to meet your specific needs. We're happy to help!
Awesome job Tiptap, looks really powerful.
Is it possible to allow the agent to work on multiple documents for the same user/group?
Yes, absolutely. You can hook the AI Agent to your custom tools that let it browse and switch files. Just like you would do when building a coding AI Agent. It's exciting but also complex to do, that's why we haven't provided an example of how to do it yet. We're still figuring it out, and we'll work on improving multi-file support in the next releases.
Ah, got it. What are the main things that make it complex to work on multiple files?
There are many ways to organize documents: a list, a nested structure of directories, a searchable unstructured directory, etc. Plus, in multi-document setups, AI Agents have many possible ways to index and search files: from simple ls commands to complex vector databases that index files for semantic search. So we're looking for a solution that is simple enough to set up and flexible enough to fit many types of applications.
Is there a github repo to copy from?