Hey everyone! We've got a packed update this week with some really useful new features, better performance across the board, and a bunch of quality-of-life fixes. Here's what's new:
We've made a major upgrade to how Juma generates PDFs and web pages. Behind the scenes, we've introduced purpose-built skills for both PDF creation and front-end design that guide the AI toward producing output that actually looks professionally crafted — think better color palettes, more thoughtful typography, and layouts that don't scream "an AI made this."
Even better, Juma now respects your brand guidelines when generating these deliverables. If you've set up brand voice in your project or provided branding context directly in the chat, the AI will pull from those guidelines to match your colors, tone, and visual identity. If neither is available, Juma defaults to its own curated color palette and styling — so your output always looks polished, regardless of whether you've configured branding or not.
A new "Attach existing" dropdown lets you quickly pull in instructions you’re already using in other projects or created as standalone ones. It's a small change that saves a lot of clicking around when you're building out a conversation or project.
You can now also attach multiple info & instructions per project which are all taken into account by the AI when chatting.
Chat now supports a message queue, so if you're sending follow-up messages while the AI is still working, they'll be lined up and processed in order rather than getting lost.
You can add a queued message by typing your prompt in the chat box while AI is still generating a response and pressing Enter or the “Add to queue” button.
Need your brand fonts available in Juma? You can now upload custom font files (.woff2, .ttf, .otf) directly into your projects. These fonts are also available inside the sandbox environment, so your generated content stays on-brand. On top of that, we've added a Google Fonts resolver with a curated font pack — giving you access to popular typefaces out of the box.
The command hub search is now significantly faster thanks to parallel entity queries — you'll notice the difference when searching across your workspace. We've also improved our LinkedIn and Instagram tools performance with faster discover queries and fixed timeout issues that some users were experiencing with MCP integrations. The prompt builder no longer hangs during AI generation, which was causing delays of up to 8 minutes in some cases.
We tidied up several UI details: the chat input no longer shows a rectangular glow artifact, the scroll-to-bottom button behaves correctly and no longer overlaps with the queue panel, and the deliverables panel now properly closes when you delete a thread or message. The settings menu integration list is capped at 7 items to prevent overflow, and we've fixed an issue where project knowledge items weren't appearing correctly via the External API. HTML previews now work for all File tool actions, and you can edit or delete instructions that were created through app management.