Back to Blog
Product
BrandOS: context engineering for your brand
BrandOS converts everything your brand is — visual identity, voice, language, AEO/SEO rules, and legal guidelines — into a structured data model that both humans and AI can work with directly. It sits at the center of Magi as the bridge between content creators, AI systems, and the brand itself.
The problem we kept seeing
Most marketers who've gone deep on AI have already tried to solve this. A brand voice doc in the system prompt. A SKILL.md file. A custom GPT built on a weekend. It works — for a while.
Then the positioning shifts. A competitor launches. The ICP evolves. And the context you encoded quietly stops being true. The AI keeps generating from it anyway, resurrecting messaging the team deliberately moved away from. The model isn't being stubborn. It's following what looks like current guidance. The guidance is just stale.
This is how brand drift happens.
The biggest marketing pain point that often surfaces on our customer calls is brand consistency:
Our primary use case is building campaigns with consistent branded messaging. And so that's what we're really looking to solve for.
-David VanEveren, Vice President, Demand Generation, Ring Central
BrandOS is the brain we built to fix that. It is the living source of truth that evolves as your brand does. Every content piece that you generate using Magi reflects the current version of your brand. You can keep BrandOS up-to-date by editing or adding any positioning shifts, product launches, or ICP change.
Introducing BrandOS
BrandOS converts what it means to have a brand into a structured data model.
Think of it as a neural network built around your brand. The brand is the core: your identity, your purpose, your positioning. Voice and tone, visual identity, language rules, keywords and prompts, compliance, and content guidelines are the interconnected nodes around it.
Each node does something specific on its own. Collectively, they learn from each other and reinforce the whole. The more completely the network is built out, the stronger every individual output becomes.
This is what separates BrandOS from a style guide or a brand document. It gets smarter the more it's used. Traditional brand guidelines sit in PDFs. Brand guidelines for AI marketing need to be structured, queryable, and connected. BrandOS is built for that. It gets smarter the more it's used.
One important detail about how it works: Magi does not pass the entire ontology into every generation step. The ontology is exhaustive by design, so instead, agents intelligently fetch only the parts relevant to the current context. If an image is being generated in an illustration style, there is no reason to include photography guidelines in that context. The system selectively pulls only what matters. This keeps context cleaner, reduces wasted tokens, and improves both quality and performance overall.
Components of BrandOS
Brand Identity
The foundation everything else builds on. Brand Identity is where you define who you are, why you exist, and what makes you distinct, in enough detail that the AI can reason from it.
It covers six structured areas:
Identity | Brand name, primary domain, short description, and long brand story. Includes allowed forms ("Magi", "MagiHQ") and disallowed forms ("MAGI") — so the AI never renders your name incorrectly. |
Brand Archetype | A universal character type (e.g. Magician, Hero, Sage) that gives the AI a personality framework to draw from across all content. |
Brand DNA | Mission statement (why you exist) and vision statement (what future you’re building toward). These inform the “why” behind everything the AI writes. |
Core Values | Name and description pairs — e.g. “Never Settle for Good Enough,” “Move Fast. Learn Faster.” Each value includes context on what it means in practice, so the AI reflects values in substance, not just tone. |
Key Differentiators | What sets you apart from alternatives — structured as named differentiators with descriptions. The AI uses these to lean into your positioning in competitive content, comparisons, and sales enablement. |
This is where the difference between generic AI output and brand-specific AI output is decided.
A tool that doesn’t know your differentiators writes content that could belong to any competitor. A tool grounded in your Brand Identity writes content that only you could have published.
What this changes: Content is rooted in why your company exists, what it stands for, and what makes it distinct — not a generic AI interpretation of your industry. And because Brand Identity defines exactly what's allowed and what isn't. |
Visual Identity
Every visual decision your designer made — icon style, illustration language, typographic hierarchy — lives in a Figma file the rest of the team can't use. The standards exist. The intent exists. Without a way to encode it into the system, every new request starts a conversation with design, and every output without that conversation drifts.
Upload your colour palette, fonts, and logo library. Add your iconography set, illustration style, and photography direction. Crucially, add a text description alongside each asset. This is how Magi learns to apply your visual intent rather than just recognise your files.
The visual configuration now goes significantly deeper — users can define illustration styles, add visual references, and configure how each asset type should behave across different contexts. Once that's done, content generated through Magi starts reflecting those patterns much more consistently. Logos come out correctly. Styling stays closer to the provided references. Outputs feel aligned with the brand rather than approximating it.
Two layers of visual context work together at generation time. The Visual Prompt is defined once at the brand level — it captures mood, colour language, lighting behaviour, and the forms your brand prefers. It applies to every visual Magi generates.
The System Prompt is defined per medium — it captures how that specific format should be made: composition, subject treatment, technical specs. A blog cover might use a halftone dot-pattern. Icons might be built on a 24×24px grid. Each medium gets its own rules because each format requires different treatment.
When Magi generates a visual, it stacks both: Visual Prompt + medium-specific System Prompt + user's request. The visual prompt holds the brand mood constant. The system prompt makes sure the right kind of asset gets made for the context. The result is a landing page design or a blog header that draws from your actual icon library and illustration style — not a generic stock interpretation of your brand.
You define what on-brand looks like for each context once. Magi applies it every time.
What this changes: The designer's POV stops living in Figma and starts living in every output, whether or not design was involved in making it. When a salesperson needs a one-pager or a marketer is spinning up a campaign without a designer in the loop, the output reflects the same intent as if design had been there. |
Voice & Tone
Multiple people like the founder, marketer, or salesperson can use Magi to create content in their voices and the content adapts to their tone and voice.
Voice is who you are. It's the consistent personality that runs through everything you publish — the point of view, the way you frame ideas, the things you'd never say. It doesn't change based on what you're writing or who's writing it.
Define your brand voice once in Magi. Every output starts from that foundation.
Tone is how your voice adjusts to the room. A founder's LinkedIn post reads differently from a support email, even when both are unmistakably the same company. More direct in sales. More considered in thought leadership. Warmer in the community.
Multiple people can use Magi to create content in their own voice. The founder, the marketer, the salesperson. The content adapts accordingly. Create named voice profiles for everyone who publishes, and a tone library with situational "when to use" notes for each. Set defaults that apply automatically to new content, so nothing goes out in the wrong register.
Or skip the manual setup: import a voice directly from a landing page or blog post and let Magi learn it from real examples.
What this changes: Users can now define precisely how they want the brand to sound — and the system does a significantly better job of maintaining that consistency across content. The result is a consistent brand voice across channels without a briefing doc at the start of every session. Voice and tone get more accurate over time. Every piece of content created under a voice profile teaches Magi more about how that person actually writes: their cadence, their instincts, their defaults. |
Messaging
Define your positioning statement, taglines, and messaging pillars once. These aren't just reference material, they're the narrative architecture Magi builds every content piece around.
Messaging pillars are the core themes your brand amplifies across every marketing programme: launches, demand gen, thought leadership. Each pillar has a name and a description of what it means in practice.
Magi uses them as the underlying logic when it picks an angle, frames a story, or structures an argument — so every programme, regardless of format or channel, is pulling in the same direction.
Each tagline is paired with context for when to use it, so the right line surfaces in the right moment — not just the most recent one someone remembers.
What this changes: When messaging pillars are structured into the system, Magi already knows what good looks like — and produces to that standard from the first draft. |
Language & Grammar
Most AI tools default to their own grammar preferences. Magi defaults to yours.
Language & Grammar lets you define the exact writing rules your brand follows and applies them automatically across every output.
Set your language variant, preferred and avoided terms each with a reason why, and capitalisation rules per context. Grammar preferences like Oxford comma, contractions, active voice, and spaced em dashes are simple toggles. Date format, quotation mark style, hyphenation rules, and punctuation style round out the layer.
The “why” behind each term is as important as the term itself. When the AI understands the reasoning, it can make better judgment calls in contexts you haven’t explicitly covered.
What this changes: Every output reads like it was written by someone who actually knows your style guide because the rules are structured into the platform. |
Keywords and Prompts
Keywords are the search terms your buyers use when researching a problem you solve, organised by persona and topic. When you add keywords in Magi, it ensures all your content is optimised for search engines across channels and content types. You can add them manually or upload a CSV of your keyword universe.
Prompts go further. These are the exact questions your buyers type into Perplexity, ChatGPT, Claude, and Gemini — conversational queries that don't show up in a keyword tool but decide whether your brand gets cited in an AI answer. Organised the same way: by persona, by topic, manual entry or CSV.
Both layers feed directly into the content workflow. When Magi generates a blog post, landing page, or comparison piece, it references the keywords and prompts mapped to the relevant persona — so every piece is written to rank and to be cited, not just to exist.
What this changes: You stop optimising for a search index that's losing share and start optimising for the surface where your buyers actually research now. |
Content Guidelines
Define the structural rules and real examples that govern how every content type is produced.
Content Hierarchy sets the information presentation order for each format. Structural rules define formatting conventions.
CTA Templates give Magi a library of approved calls-to-action — “Book a demo,” “Try Magi,” “Start your 7-day trial” — so the right action phrase appears in the right context, consistently.
Content Type Prompts let you add system-level instructions per format, and you can set intro requirements and ideal bullet counts for formats where structure matters most.
The most powerful part is Content Examples. Upload real pieces of your best content organised by content type.
Magi uses these as reference outputs to understand the structural patterns that define how your brand actually writes at its best.
The more examples you add, the more precisely Magi can replicate the output quality you’ve already proven works.
What this changes: Every piece Magi produces follows the structure your best content already uses, not a generic template, but the specific patterns your audience has already responded to. |
Social Media
Define the platform-specific rules that govern how your brand shows up on LinkedIn and X, so every post Magi creates is sized, tagged, and formatted for the platform it's going to, not just written in the right voice.
Set always-on hashtags that apply across platforms, and platform-specific tags for LinkedIn and X separately. Platform Guidelines go beyond character counts to format-level guidance post length by content type, visual dimensions, and imagery rules per platform.
What this changes: Posts come out in the format your brand already uses. |
Legal & Compliance
Every brand has rules about what it can and can't say — financial claims, healthcare references, region-specific disclaimers, competitor comparisons. Most teams enforce these in review, which means content gets written, circulated, flagged, and rewritten. It's slow, and it depends on someone remembering to check.
Legal & Compliance lets you set those rules inside Magi so they're enforced at the point of creation. Define what to detect, what to do when it's found, and who it applies to. Magi flags the issue and suggests the fix before anything leaves the platform.
Magi ships with 12 default rules covering the most common patterns — earnings disclaimers, cookie disclosure, ROI substantiation — so you're protected from day one without building anything from scratch.
What this changes: Legal review stops being a bottleneck. Compliance is enforced where content is created, not after it’s already been shared for approval. |
The bridge between users, AI, and the brand
BrandOS is the neural network your brand thinks through.
Marketers draw from it when they brief, write, or review. AI agents query it when they research, ideate, or draft. Brand and compliance teams strengthen it when they add rules, refine voice, or update positioning. Every interaction adds a signal. Every node reinforces the others. The network gets smarter the more it's used.
This is fundamentally different from how most AI tools handle brand context.
A pasted style guide or a markdown doc gets read the same way the model reads everything else, as flat text, with no structure, no hierarchy, no connections between ideas. BrandOS gives AI the right brand context for content generation, which means structured fields, typed rules, and direct asset references and the relationships between them.
It understands how every part of your brand connects to every other part and generates from that understanding.
Why BrandOS matters especially for image outputs right now
A significant reason BrandOS feels especially powerful at this moment is the generation quality of newer models.
Models are far better at understanding structured context than their predecessors and that capability compounds when combined with a well-configured ontology.
What this means in practice: image outputs now come back with the right visual hierarchy, correct logo placement, and on-brand color usage, without having to recreate the brief from scratch every single time.
BrandOS tells the model what the brand looks like and how its assets are allowed to be used. The model does the rest.
What customers are seeing with BrandOS
The qualitative improvements from customers have been meaningful.
For instance, below is a comparison of visual generation in Magi before and after the BrandOS update.
AccuKnox had always liked the copy Magi generated, but images were the gap because the brand wasn't configured tightly enough to get them right. After the BrandOS release, that changed: logos rendered correctly, themes held consistent, and the output became reliable enough that they now ship generated images without editing them.
NexaSure saw the same shift even faster. Before their brand setup was fully configured, their visual identity was already dialed in enough that generated images felt on-brand from the start. Within about a week of onboarding, they were posting generated content on LinkedIn on a consistent cadence.
This is the first time the combination of structured brand context and model capability has been good enough to produce on-brand AI content at scale.
Getting started
Maintaining brand voice with AI starts with what you already have.
Most teams import any of the below and use them as the starting point.
ICP deck
Brand guide
Positioning doc
The platform maps those into the ontology structure and surfaces the gaps.
For teams that haven’t formalized these things yet, filling out the ontology is itself a valuable exercise. It forces decisions many early-stage teams have been deferring:
Who exactly is the buyer
What does the brand sound like in writing
What’s the single thing every piece of content should convey
Already on Magi? Open Brand under Design in your workspace. Set it up, run a piece of content, and see the difference firsthand.
