Sulu.ai 0.7: See what your projects are actually doing

Until now, understanding how a project used Sulu.ai meant checking invoices or asking us. Now it's the first tab you open.

Every project keeps a complete request history, giving admins and managers a clear view of what's actually happening. Each entry shows the date and time, the model, the credits it cost, and the expert behind it.

You can filter requests by time, expert, or model, search and sort the list, and export any filtered view as CSV for reporting or customer billing. Access stays with project admins and managers, and every view is scoped to the current project.

The point: the questions you used to send us, you now answer yourself in seconds. "Where did our credits go?" and "What is this expert actually being used for?" are a filter away.

When we introduced Intelligent Search in 0.6, every project could run a single search. That worked well for many projects, but a single assistant rarely fits every use case. Your public website, customer documentation, and internal knowledge base usually shouldn't search the same content.

With 0.7, projects can run multiple Intelligent Searches side by side, each with its own sources, prompts, and experts.

We reworked the navigation to match: every search now lives in its own dedicated section instead of being buried in the general project settings.

A few refinements come with it:

• Experts now resolve placeholders against their own search, so prompts stay correct when multiple searches exist.
• Developers can target a specific search through an optional identifier.
• The demo template now exposes Twig blocks, so you can restyle it in your own design without maintaining a fork.

Along the way, we also fixed webspace filtering, titles that contain line breaks, and source selection that occasionally reached too far. If you want the thinking behind it, we've written about how Intelligent Search turns your existing pages into direct answers. 

• Part 1: From pages to answers: Why website search needs to change
• Part 2: From pages to answers: How Sulu's Intelligent Search works 

Translation quality improves in 0.7 without changing how you work.

DeepL's next-generation model is now the default whenever it's available for a language pair, falling back to the classic model automatically otherwise. The result is more natural translations, especially for longer, connected text.

The biggest difference is in flow. Longer paragraphs stay coherent, terminology is more consistent across a page, and translated content needs less manual polishing before it's ready for review. Your workflow stays exactly the same, but editors spend less time correcting AI output and more time reviewing the message itself.

Not every Sulu project uses the standard editor fields. Many extend Sulu with their own field types, and until now those required custom code to work correctly with AI features.

With 0.7, you can explicitly configure which field types should be treated as plain text and which as HTML. That configuration is shared across translations, full-page translation and the Writing Assistant, so custom fields work consistently throughout Sulu.ai without forking the bundle.

Defaults remain unchanged, so existing projects keep working without additional configuration.

Usage limit notifications now reach the people who can act on them. Instead of sending operational emails to invoice contacts, Sulu.ai now notifies everyone in the organization except users with the invoice role. Projects can also define one additional notification address for a shared inbox or an external partner.

Large Sulu 2.6 installations now ingest much faster, with a much smaller memory footprint. Nodes are scanned once instead of once per locale, documents are counted without hydrating them, and memory stays stable throughout long-running imports. We also fixed published-flag handling, so empty published values now count correctly as unpublished.

The MCP Server we teased in 0.6 has taken shape.

Visibility is only the first step. The next is letting AI assistants work directly inside Sulu. The upcoming MCP Server connects Sulu to assistants like Claude, Codex, and other MCP-compatible clients, so they can work directly on your structured content instead of copying information between tools.

The foundation is already there: pages, articles, and blocks are supported, references resolve automatically, and generated content links straight back into the Sulu admin. The remaining work is authentication and permissions, so an assistant can only ever do what the connected user is already allowed to do.

While the MCP Server itself will be open source, some advanced tools and integrations will be powered by Sulu.ai, bringing AI capabilities directly into your editorial workflows.

We're also working on a refreshed user interface, bulk AI operations for working with large amounts of content, and a number of smaller improvements that make Sulu.ai faster and easier to use every day.

The connection to 0.7 is intentional. Before assistants can actively change content, you need to see what AI is already doing, and Insights lays exactly that foundation.