Why Nuxeo is more AI ready than people think

Nuxeo is more AI ready than most people give it credit for. If you are already using it, you have all the building blocks you need to add serious AI capabilities, without replacing the platform.

The ECM and DAM space is going through an exciting shift right now, with new platforms and new ideas around AI showing up almost every week. It is a good moment for the industry. In this post I want to focus on Nuxeo specifically, and walk through the technical reasons it holds up well in this new landscape, and how you can add AI capabilities to it using what is already there.

Also, yes, I have a lot of experience with Nuxeo, but I will try not to be biased :).

Backend

One of the strongest arguments for using Nuxeo is the support for NoSQL storage (MongoDB), which is more suited for scaling large volumes than a SQL backend in an ECM/DAM. Nuxeo also supports Elasticsearch or OpenSearch (not mandatory in 2026, as MongoDB Atlas Search can be used instead).

Both MongoDB and Elasticsearch support storing embeddings and running vector search natively, which is what you need for semantic search and similarity search across your repository.

There are multiple ways to generate embeddings for your images, videos, or documents, along with text embeddings for your metadata. You can externalize the embeddings generation (build a service outside Nuxeo) or you can simply add a bulk action into Nuxeo that invokes the LLM of your choice, gets the embeddings back, and stores them in Elasticsearch or MongoDB. I will write some dedicated articles about how to do this and enable semantic search or similarity search across your repository.

This is a very simple (and essentially cheap) task and most people don't realize how non complicated it is, as Nuxeo does offer all the building blocks:

• Renditions: Nuxeo supports different renditions, so you don't need to send the original image, video, or contract to the LLM. You can choose to send one of these lightweight renditions (for pictures, you can send a Full HD JPEG, or even a smaller Medium rendition) to generate the image embeddings. And you will have this rendition available even if your original file is a JPEG, a PNG, a Photoshop, or an InDesign file.
• Kafka streams and the Bulk Action Framework: You can configure a bulk action to do this. With bulk actions you can configure the concurrency, the retries, the failure on errors, etc. So you can invoke any LLM of your choice at scale, with proper backpressure, without reinventing the pipeline.
• Document model and facets: This one is often overlooked. Nuxeo's typed schemas mean that whatever an LLM extracts (summaries, tags, classifications, sentiment, key clauses from a contract) lands in proper typed properties, not in a blob you have to parse later. And facets let you attach AI generated data to existing documents without touching the original schema. So you can add an aiEnriched facet with embeddings, summaries, and confidence scores to any document type already in your repository.

At Maretha, we built a generic addon for basic AI interactions with any model. Feel free to take a look at maretha-ai on GitHub. And if you come to Hyland CommunityLIVE, come to see our talk!

REST APIs and MCP

I would argue that you can make any platform "AI ready" as long as the platform exposes really well designed APIs (that support pagination, async, etc.) that an agent can interact with.

Not only does Nuxeo expose many REST APIs, it is also easy to add new ones. OAuth 2 is natively supported by the Nuxeo Platform, so authenticating an agent or an external app is a solved problem.

And this is where it gets interesting for 2026. There is already a Nuxeo MCP server that exposes Nuxeo operations to any agentic platform that speaks MCP. That means your army of agents (Claude, or whatever orchestrator you are using) can search the repository, read documents, trigger workflows, run automation, and write back enriched metadata, without you having to build a custom integration layer. If you have a custom operation in Nuxeo, you can expose it through MCP and an agent can call it. This is, in my opinion, the most underrated argument for Nuxeo in the AI era. The platform was already built around an extensible operation model, and MCP is essentially a standard wrapper on top of that.

UI

One of the weakest points for Nuxeo is the UI. Polymer has been deprecated for some time, and even though we have been promised a new UI for years, it has not been delivered.

However, the good news is that everything I described above means the UI is the easy part to replace. Nuxeo exposes really good REST APIs that support pagination, OAuth 2 is natively supported, and bulk operations are async by design. So while the UI is not pretty, you have all the right building blocks to unleash Claude to build you a better one in React, Angular, Vue, or whatever you need. It is not a given, but with the right direction your agents can build you a nicer UI that taps into all the main features of the platform: async processing and pagination to make it performant and scalable.

Because, don't forget. No matter how shiny the UI is, if the backend does not scale, your application will not scale either.

So, is Nuxeo more AI ready than people think? I believe yes, and I hope this post gave you a few concrete reasons why. The building blocks are there. The hard part is using them well. At Maretha Solutions we help teams add AI capabilities to their Nuxeo instances, from semantic search and embeddings pipelines to MCP integrations and custom agentic workflows. If any of this resonates with your roadmap, reach out. We would love to talk.

← All insights