Today we are releasing a major stepping stone toward a multi-agent network that can operate as part of a B2B marketing operating system: the Rules Agent.
Magnity has always offered a huge degree of freedom and flexibility through its layered rule system. That flexibility is powerful, but it also comes with complexity and responsibility. So we decided to build an agent specifically to help you understand, navigate, optimize, and safely update your rules.
This release also includes updates to the Content Agent, the Compliance Agent, a new home page, and a few important fixes.
The Rules Agent is an AI reasoning agent attached to your rules.
It understands:
• How your rules are structured
• How they relate to each other
• How merge fields work
• How rules affect content generation across Magnity
• How to safely suggest and apply updates
It comes with:
• A general chat for all rules
• A dedicated chat for each individual rule
You can use it to:
• Ask questions about your rules
• Optimize all rules at once
• Optimize a category of rules
• Optimize a single rule
• Ask for recommendations
• Discuss changes before applying them
• Get suggested edits
• Let the agent perform the changes for you
The Rules Agent uses a set of specialized skills to reason about your rules and decide what to do.
From the current setup, these skills include:
• How rules are organised
Helps explain the overall structure of the rule system and how the different rule layers fit together.
• Key relationships
Explains how rules influence each other and what dependencies exist between them.
• Proposing rule updates
Helps draft improvements, rewrites, or optimizations to rules based on your intent.
• Reviewing rule quality
Evaluates whether a rule is clear, useful, redundant, too vague, or potentially conflicting.
• Using merge fields
Explains and helps update merge fields correctly so rules use the right dynamic inputs.
• Version history and rollback
Understands how rule versions, snapshots, and rollback work so experimentation stays safe.
• Element generation order and dependencies
Explains how rules affect content generation order and how certain rules depend on output from others.
• External research and inspiration
Helps reason about outside inspiration and how to translate intent into better rules without breaking your existing setup.
We have redesigned the rules interface to make it much easier to understand, use, and navigate.
When you open your rules, you now land in the All rules view, and the chat on the right tells you that you are currently chatting with all rules.
You can:
• Filter by source and medium
• Search rules
• Sort rules alphabetically
• Sort by latest updated

This makes it much easier to find the rules you just changed, or the rules you want to review next.
We have updated the names of rule categories to better reflect their purpose and context.
The changes are:
• Global rules → Brand rules
• Local rules → Market rules
• Base prompts → Element rules and Element prompts
You can now also work with:
• Task rules
This is where personalization and translation rules now live.
• Flow rules
These now live under Element rules.
When you expand a rule, you can now see:
• A short description of what it does
• The rule itself
For element, market, and task rules and prompts, you can also see the merge fields used by the rule.
Merge fields are the dynamic pieces of text Magnity inserts during generation, such as:
• The URL of the tile you used
• The selected language
• The scraped source text
• Other runtime context

If you are unsure what they do, the Rules Agent can explain them to you.
When you click a specific rule, the chat for that rule opens.
Next to the chat, you will now find a new tab: Rule Stack.
This shows you what other rules influence the rule you are working with. That matters when making changes, because rules rarely live in isolation.
The Rules Agent understands the Rule Stack as well, so when it suggests or applies changes, it takes the surrounding rule context into account.

In the top right corner you can now access Rules History and see the changes that have happened.
You can also click a new function called Snapshot, which lets you save a snapshot of all your rules before experimenting.
This is especially helpful before making changes across multiple rules.
In addition, each individual rule still supports rollback to its five latest versions, so if you only want to undo one rule change, you can do that quickly.

When the Rules Agent suggests an update, it will always offer to show you a diff.
The diff makes the change transparent by showing:
• Removed text in red and struck through
• New text in green
• The changes side by side
It will also offer an Action Block, where you can see the full updated rule and either:
• Accept
• Dismiss
If you accept, a new rule version is created and saved automatically.
This gives you a much safer and more flexible way to optimize rules and experiment with rule sets.
There are more details to the Rules Agent than we can reasonably cover here. We will go through those in an upcoming training video.
If you would like a more detailed walkthrough, we are also happy to offer a full 1-hour presentation free of charge.
Please reach out to your admin.
The next step for the Rules Agent is to allow it to make proactive rule suggestions. We are working toward that now, but there are a few things we need to build first.
This is one of many steps toward an agentic OS for B2B marketing.
We have made a few useful additions to the Content Agent.
You can now copy text directly from the Content Agent canvas using a small copy icon in the top right corner.
In the chat, you can now choose between:
• Replace
This replaces the text in the canvas with the selected chat response.
• Add to canvas
This adds the selected text to the bottom of the canvas.
This is especially helpful when building campaign plans or other structured documents that expand as you work.
We have made two important improvements to the Compliance Agent.
You can now run multiple analyses and choose which analysis you want to export findings and remediations from as comments.
This means you can analyze a PDF in multiple steps and still get all suggestions collected into one exported document.
A new Manual edit button makes it easier to manually replace a found issue directly. This is a fast and practical way to edit content when you know exactly what needs changing.
We have also fixed the chat in the Compliance Agent so that it can now apply the fixes you reach through discussion in the chat itself.
We are very happy to announce that Magnity is now using OpenAI’s new Image Model 2.0 for both image creation and editing.
You can use it:
• In the Content Agent
• In the image cropper
If you have not tried the new image capabilities yet, give them a spin.
For example:
“Give me an image of my brand’s top product as a blueprint infographic. Search for product details.”
Here is ours :D

You will also notice that the start page has been redesigned.
The new home page now allows you to:
• See the last four projects you worked on
• Pick between the available agent builders you have access to
• Read the latest news
This, together with the Rules Agent, is an early look at the broader redesign we need to undertake as we move toward an agentic OS.
We are currently making backend updates so that more of Magnity can be transformed into agents.
This new approach also demands a new design direction, which we have wanted for a long time. We are now actively working on that.
We are going to redesign Magnity. As part of that redesign, several smaller but highly useful features will arrive, including:
• Folders
• Comparison views
• Review features
• Other quality-of-life improvements
This will take a few weeks, but it should make the interface significantly better, more helpful, more intuitive, and more flexible.
Stay tuned for more.
We have fixed a couple of important issues:
• Previously, if you used a high thinking model on a very large task, Magnity could appear not to respond. This happened because the underlying Umbraco backend timed out if no response was returned within four minutes. This is no longer a problem, and the model can now think as long as needed before responding.
This means you can now use high thinking models for tasks such as translating large volumes of text.
• We have also fixed the chat in the Compliance Agent so that it can now apply any fixes you discuss your way toward in the chat.
Our support team is here to help:
✉️ ulsa@magnity.ai | 📞 +45 29 35 25 78
Thank you for using Magnity.
— The Magnity Development Team