Sulu.ai 0.6: Intelligent Search, granular permissions, and a stronger platform

The headline of this cycle is Intelligent Search — our RAG-style search system for Sulu content.

The idea is simple. Structured content shouldn't only power websites. It should become a reusable knowledge layer for search, AI assistants, and future applications. You've already done the hard work of structuring your content; Intelligent Search puts it to work.

How it behaves: content is indexed automatically during publishing. A public JSON API exposes it. Responses stream via SSE. Feedback signals are captured to improve results over time. It works with both Sulu 2.6 and Sulu 3.0.

We're keeping this announcement deliberately high level — the architecture, the indexing model, and the integration options deserve their own deep dive, and we're saving it for a dedicated post.

We're rolling this out gradually, together with early adopters.

Starting in July, we'll open onboarding slots for teams that want to bring Intelligent Search into real Sulu projects. The process is hands-on by design: we want to learn from production use cases while we keep refining the platform.

Want in? You can schedule a meeting with us to talk through onboarding.

As Sulu.ai adoption grows inside larger editorial teams and enterprise environments, permission handling matters more.

With 0.6, every Sulu.ai capability now has its own dedicated sulu.ai.* security context. Administrators can decide precisely which teams and roles get which functionality — editors might get SEO tooling, while translation or media AI features stay limited to smaller groups.

The point: AI now behaves like the rest of the Sulu permission system. It's visible only where it's intentionally enabled.

After upgrading, administrators must explicitly assign the new sulu.ai.* permissions to existing roles.

Roles without these permissions will no longer see AI tabs, buttons, or integrations in the admin interface. This is expected behavior and part of the new permission model - not a bug.

The Media Metadata Generator and the Writing Assistant in the media detail now support SVG files alongside traditional raster formats.

It sounds small. It isn't — not for teams working heavily with brand assets, icon systems, editorial illustrations, or design systems where SVG is often the dominant format. AI-assisted metadata, descriptions, and related workflows now work consistently across vector and raster assets.

We've also refreshed the model lineup across providers to improve quality, efficiency, and long-term compatibility. These updates rolled out in May 2026:

Claude Haiku 4.5 remains available.

As always, Sulu.ai stays provider-agnostic. Teams choose models based on quality requirements, compliance considerations, latency expectations, or budget.

We've updated the Fair Use subscription for members of the Sulu Partner Program.

Every partner now receives one included project automatically as part of their partnership. This lowers the barrier for agencies that want to start integrating AI-assisted workflows into client projects without additional setup costs.

Some platform capabilities have already landed internally and will become more visible over the next releases.

A new Subscription Status API is now part of the platform backend and can already be used programmatically. The admin integration is still in progress and will arrive in a future release.

We're building a Sulu MCP server that lets AI assistants and development tools talk directly to Sulu instances through the Model Context Protocol — think Claude, ChatGPT, and IDE-based AI agents. The first version is targeted for summer (separate from the July Intelligent Search onboarding), with a technical walkthrough once the implementation stabilizes.