18th August - 2025 release notes: New features are live

Shared by Ulrik • August 18, 2025

Magnity new features released: Selecting LLM, create new persona, delete persona, add source from manual text and more

🤖 Select your language model: Admins only
Our “burger menu” on the left has been updated so everything pertaining to our back-end engine (rules, persona, crawler, etc.) is now called Content Engine. Under Content Engine, you can find a new option called Models. Click this and select which OpenAI model you want to use for generating content. Right now you can pick between the following:

ChatGPT 4o (the model most have used up until today)
ChatGPT 4.1 – a stronger model than 4o, especially for following instructions
ChatGPT 4.1 Mini – the smaller version of 4.1 that responds super-fast and uses fewer tokens; 4.1 is better, but you can try this to experiment with differences
ChatGPT 5 – the new ChatGPT-5 model with minimal reasoning for fast replies like 4.1 and 4o
ChatGPT 5 Low Reasoning – this model thinks a little harder about things, so slower response time but more reasoning
ChatGPT 5 Mini – a smaller version of ChatGPT-5; again, 5 is stronger but this is faster and you can test to see the difference
ChatGPT 5 Nano – an even smaller and faster version of 5; see above
OpenAI o3 – a reasoning model that always reasons longer about things; good if you have complex problems you are trying to solve
OpenAI o1 Mini – a similar reasoning model, but smaller; o1 Pro is the flagship for pure reasoning models but hugely expensive; this is a mini version of that

If you are not an admin, your admin will decide if and when to use another model. Please bear in mind that changing models influences output, and making sure it is working well is paramount before committing it to an entire account.

IMPORTANT NOTE: ChatGPT 5 is a dual model, which means it can function as a normal chat model that responds in about 10 seconds, but also as a reasoning model that can spend several minutes to think about what you are asking it to do. The fourth on the list is what we recommend for trying ChatGPT 5, since it has minimal reasoning and is still a bigger model than 4o and 4.1. ChatGPT 5 has a parameter we can configure in the back end called effort. Effort means how hard it generally thinks before replying. There are four settings: Minimal, Low, Medium, and High. We have activated the first two, simply because you would go nuts if you had to wait 3–5 minutes for every reply from the two highest settings. There are cases where the higher settings can be helpful, and we would love to go through them with you. For now, try out the two lower models and see how they do. We do not guarantee you’ll like them better than 4o or 4.1, but 5 is significantly better at following instructions and hallucinates significantly less. If you are an admin and want to play with the larger models, contact us.

Pro Tip: Try building an email first with 4o, then save that. Then change any rule slightly by adding a “.” and set it to 4.1. Then build the same email, rinse and repeat with 5. Compare and see what you like best—you can always go back to a model you prefer. Also important: when you change a model, you change it for everyone in your account. So test it when usage is low (off hours) before deciding.

👥 Create & delete personas
You can now create new personas and delete personas. Do not worry—if you delete them you simply unpublish them, and we can restore them for you. We are working on a garbage bin where you can do this yourself, covering projects, personas, and very soon also language and personalized variants.
Find the persona options in Content Engine. Go there to edit any persona or create new ones. You can add images, titles, descriptions, and attributes with as many values as you like. Then save, and you can immediately use your persona in Magnity.

Pro Tip: Personas can be used to experiment with variants of existing personas. For instance, build a persona for three different industries or customer types with small variations in needs or goals. This can help you work even better with personalization. We can also help you experiment with the prompts/instructions for personalization so that it does—not just “rewrite” the default version—but rethinks the entire article for the persona, focusing on more relevant topics from a page. Again, let us know if you have questions.

✍️ Add source/tile from manual text
Now you can create a tile/source from just copying and pasting text. Click the small + sign in your content library where you normally find Add Document, Add Webpage, etc. Here you’ll find a fourth option: Manual Text. Make sure you add the text, give it a title, add a URL it should display in any CTA, and one or more relevant tags. Save it, and you’re good to go.

Pro Tip: This is a very fast way to add—for instance—a campaign brief or a news article that has not yet been published. Use it to get content into Magnity fast, but be aware that this does not support “auto” language detection, meaning you cannot add it across all markets by default. Again, if you have any questions, let your admin or us know.

🐛 Bug Fixes & UI Tweaks

Fixed a bug where Magnity would inadvertently swap around language versions in the interface when creating more than 80 variants. You can now create as many as you like, and navigation has been improved to scroll predictably when switching between variants.
Fixed an issue where images crawled in new pages caused an error in some templates. We do see this persist for one client whose server rejects our bot when scaling images; we will reach out and resolve this together.
Made significant back-end adjustments to how we save projects. Previously, they were stored in Umbraco’s database, which sometimes caused errors with the Lucene search engine and memory leaks. Now all project data is stored in MongoDB—our fast, enterprise database—and this has halved Magnity’s memory use, greatly improving stability.

🔍 Known issues
We still have a few issues with the automatic crawler that sometimes decides to crawl new content it should not. We’re releasing a hotfix Monday or Tuesday night, but should you find pages in your inventory that do not belong, do not worry—we can remove them for you. We are updating all content libraries but do not want to do this at scale before this small issue is fixed. Thanks for your patience. Automatic crawling now supports millions of pages at once, which is complex to manage; we’re working hard to get the very last bit ready.

Note: Manually adding pages works just fine, so you can still use that as needed.

🎯 Why This Matters
By adding these features, we are moving towards making Magnity a true AI agent. Introducing reasoning models is part of our progress toward enabling Magnity to understand your entire content library at once—and allow you to chat with an AI that helps you utilize it better. More on that will come. Also, this update prepares for a simple but helpful feature arriving in a few days: deleting personalized and language variants.

Again, more will come soon. We hope you enjoy the new features.

Questions or Concerns?
Our support team is here to help:
✉️ ulsa@amagnity.ai | 📞 +45 29 35 25 78

Thank you for using Magnity!

— The Magnity Development Team