What is Brillix?
Brillix is an AI system built around one idea: a general-purpose AI that knows nothing about you is less useful than a specialized AI that learns how you think. Every conversation you have on Brillix is used to build a deeper understanding of your preferences, your communication style, and your goals — so the next answer it gives fits you better than the last one.
Unlike a single AI model that does everything at an average level, Brillix routes your questions to the right specialist for the job. A financial question reaches a system trained exclusively on financial intelligence. A coding question reaches a system trained exclusively on software engineering. The result is answers that go deeper than any generalist can.
Specialist AI Systems
Purpose-built AI for finance, code, research, and more — each trained on its own domain exclusively.
Perspectival Memory
Brillix learns your preferences, vocabulary, and thinking style over time. It adapts — not just to what you ask, but how you ask it.
Truth Mode
Every claim is tagged: confirmed fact, disputed, or opinion. You always know what you're getting.
Live Research
Real-time web search, source retrieval, and grounded answers — with full citations you can verify.
How It Works
When you send a message, Brillix doesn't just forward it to one AI and return the result. It routes the question through a coordinated network of specialized systems — each contributing what it knows best — and synthesizes a single, coherent answer.
The routing happens automatically. Brillix reads the nature of your question, classifies its intent, picks the right specialist systems to engage, and blends their outputs. You see one clean answer. Behind it, multiple systems contributed.
Most AI products use a single large model for everything. That model has to be good at finance, code, creative writing, medical questions, and casual chat all at once. Specialization always wins in a focused domain. Brillix routes to specialists rather than relying on generalist averages.
The synthesis layer is what makes the answer feel unified rather than stitched together. It resolves conflicts between systems, weights confidence levels, and formats the output for readability — all before the response reaches you.
Specialist Modes
The pill row at the top of every Brillix conversation shows the available specialist systems. You can click any of them to direct your question specifically, or leave it on the default and let Brillix route automatically.
Each pill routes your question to a different system with a different training focus and response style. The active system is highlighted. Clicking a pill mid-conversation switches context immediately — Brillix keeps the conversation history but adjusts the specialist handling the next response.
If you don't select a specialist, Brillix classifies your question automatically and picks the best system for it. Most users never need to manually switch — the routing is accurate enough that it just works.
Atlas — Financial Intelligence
Atlas is Brillix's financial specialist. It is trained exclusively on financial knowledge — personal budgeting, credit scores, debt management, investing, tax strategy, retirement planning, and real-world money decisions. It is not a generalist model that happens to know some finance. Finance is all it knows.
Atlas speaks plainly. It gives direct, actionable guidance the way a trusted advisor would — without hedging everything into uselessness. It knows the difference between a question that has a factual answer and one that depends on your specific situation, and it tells you which is which.
| Topic | What Atlas covers |
|---|---|
| Credit | Score factors, utilization, hard vs. soft inquiries, disputing errors, building from zero |
| Debt | Payoff strategies (avalanche vs. snowball), collections, charge-offs, settlements |
| Budgeting | 50/30/20 rule, zero-based budgeting, emergency funds, cash flow management |
| Investing | Index funds, ETFs, 401k, Roth IRA, compound interest, risk tolerance |
| Taxes | Deductions, credits, estimated payments, 1099, W-4 adjustments, IRS notices |
| Legal money | Bankruptcy basics, wage garnishment, statute of limitations on debt |
Atlas provides financial education and information — not licensed financial advice. For decisions involving significant sums, consult a licensed financial advisor.
Vulcan — Software Engineering
Vulcan is Brillix's software engineering specialist. Every code answer it gives has been run through a logic-validation layer that checks for common failure modes before the response reaches you. It doesn't just generate plausible-looking code — it checks its own work.
Vulcan's training data comes from two sources: real functions extracted from MIT-licensed open-source repositories, and a synthetic generation pipeline that produces verified examples across edge cases that rarely appear in standard training sets.
| Category | Coverage |
|---|---|
| Languages | JavaScript, TypeScript, Python, Go, Java, SQL, Bash |
| Frameworks | React, Node.js, Express, FastAPI, Next.js, Vue |
| Algorithms | Sorting, searching, dynamic programming, graph traversal, string ops |
| Systems | Concurrency, async patterns, caching, queues, databases, APIs |
| Security | Auth, JWT, encryption, injection prevention, secure defaults |
| DevOps | Docker, CI/CD pipelines, environment management, deployment |
NEXUS
NEXUS is Brillix's full-synthesis mode. Where individual specialist systems go deep on their domain, NEXUS engages all of them simultaneously and synthesizes a unified response that draws from every system at once.
Use NEXUS for complex, multi-part questions that span more than one domain. A question like "should I quit my job to freelance?" touches finance, career strategy, psychology, and practical planning. NEXUS coordinates across all of those instead of defaulting to whichever domain the question most resembles.
NEXUS is available to subscribers. The NEXUS pill is always visible in the interface — clicking it shows a brief explanation and a path to upgrade if you're not yet subscribed.
Truth Mode
Truth Mode is how Brillix handles the difference between fact, uncertainty, and opinion — without making everything feel like a legal disclaimer. Every factual claim in an answer is tagged with its epistemic status.
| Tag | What it means |
|---|---|
| Confirmed | Verifiable fact — backed by data, consensus, or primary source. Can be checked. |
| ⚠ Disputed | Contested claim — experts disagree, evidence is mixed, or context changes the answer. |
| · Expert view | Informed perspective or professional consensus — not a universal fact, but widely held. |
Truth Mode activates automatically on answers that contain factual claims, statistics, or advice. Can be toggled off in settings if you prefer untagged output.
Perspectival Memory
Perspectival Memory is the system that makes Brillix feel different the longer you use it. It observes patterns across your conversations — not by recording everything you say, but by extracting signals: how you communicate, what you care about, how deeply you want things explained.
This is not a simple "remember this" feature where you manually tell the AI things about yourself. It's passive, continuous inference — the same way a person who knows you well picks up on your preferences without you having to list them.
Communication style
How formal or casual you are. How long your messages tend to be. Whether you prefer bullet points or prose. Brillix mirrors your register without being asked.
Topic context
Recurring subjects, ongoing projects, and domains you ask about frequently — so Brillix has background context without you repeating it every session.
Depth preference
Whether you want the quick answer or the full explanation. Brillix calibrates response depth to match what you've historically engaged with.
Your data
Memory is stored per account and never shared across users. You can view and clear your memory at any time from account settings.
Signal Core
Signal Core is Brillix's personal knowledge graph — a live, evolving map of what you actually know. It builds from your conversations and tracks your demonstrated understanding across domains, topics, and subtopics.
The result is that Brillix stops explaining things you already know. If your Signal Core shows strong familiarity with a topic, Brillix skips the basics and responds at the level where you actually are. If it detects a gap, it fills it without being asked.
Perspectival Memory
Shapes how Brillix talks to you — tone, depth, format, register.
Signal Core
Shapes what Brillix assumes you already know — adjusting the starting point of every explanation.
Signal Core builds automatically in the background. There is nothing to set up or configure. The longer you use Brillix, the more precisely it calibrates to where you actually are.
Sources & Research
When you ask a question that benefits from current information, Brillix runs a live web search alongside the AI response. Sources are shown inline, linked, and visible — not buried or hidden behind the answer.
The Research specialist mode expands this into a full deep-research workflow: multiple search queries, cross-referenced sources, synthesis across findings, and a structured output with citations. It works the way a researcher would — not just the first result, but a surveyed answer across many.
After most answers, Brillix surfaces suggested follow-up questions based on what you just asked. These aren't generic prompts — they're generated from the specific content of the answer, pointing toward natural next steps in the inquiry.
Code Studio
Code Studio is Brillix's interactive coding environment. It opens alongside the chat and gives you a live editor, an output panel, and a running environment where code executes in real time. You don't copy-paste to test — you run it directly.
Code Studio is available on desktop. It integrates with the Vulcan specialist — when you're working in Code Studio, Vulcan handles all code-related responses automatically, without you having to switch modes.
Living Documents
When you ask Brillix to research a topic, you get a detailed, sourced document. A Living Document stays current. It monitors the factual claims inside it and alerts you when something changes — so you're never working from outdated research.
Mark any saved document as Living and Brillix registers the factual claims inside it for monitoring:
Claim tracking
Each verifiable fact in the document is registered. When a claim becomes outdated or contradicted, you get an alert with the new information.
Auto-refresh
You can trigger a full document refresh at any time. Brillix re-researches the topic, compares with the original, and updates what changed.
Living Documents are built for research that you act on over time — competitive analysis, market overviews, regulatory summaries, ongoing due diligence. The research happens once. The maintenance happens automatically.
Knowledge Base
The Knowledge Base is a personal intelligence layer that builds a model of what you know — derived passively from every conversation you have on Brillix. You never have to maintain it. It maintains itself.
As you explore topics, ask questions, and refine your thinking, Brillix tracks which beliefs you've stated, which subjects you've touched, and which areas you've never engaged with despite clear relevance to your interests. Over time, this produces four outputs:
Established Knowledge
Facts and beliefs you've stated across conversations. Brillix monitors whether they're still accurate — and flags them when evidence shifts.
Knowledge Gaps
Adjacent topics you've never explored despite clear logical connection to your interests. Surfaced as prompts you can investigate immediately.
Challenged Assumptions
Positions you've held that new evidence contradicts. Brillix shows you the conflict without telling you what to think.
Recall Challenges
Periodic checks on things you researched weeks ago. Helps you retain knowledge, not just acquire it.
Each knowledge gap shows an "Explore with Brillix →" button. Clicking it opens the main interface with that topic pre-loaded — so you can start learning about it immediately, without having to type anything.
General Intelligence
Atlas and Vulcan get a lot of attention because they represent something genuinely new: AI trained exclusively on a domain rather than averaged across everything. But most conversations on Brillix are general — writing, brainstorming, planning, analysis, research, creative work, explanation.
The general intelligence layer handles all of it. It is a large-context, instruction-following model optimized for the full range of what people actually use AI for day to day.
Writing & Communication
Drafts, edits, rewrites, emails, presentations, reports — written to match your voice and adjusted using Perspectival Memory.
Research & Analysis
Deep research with live sources, structured summaries, side-by-side comparisons, and grounded conclusions.
Planning & Strategy
Breaking down complex decisions, mapping tradeoffs, building step-by-step plans across business, personal, and creative domains.
Learning & Explanation
Any topic, any depth. Brillix calibrates the explanation to what you already know — not a fixed level for everyone.
The specialist systems sit on top of this general capability — routing to Atlas or Vulcan when the domain warrants it, and falling back to the general layer for everything else.
Open Information
Most AI products make a choice you never see: they decide what you should believe before they answer. Questions with contested answers get smoothed into a consensus. Alternative perspectives get omitted. The AI acts as a filter on reality rather than a window into it.
Brillix works differently. When you ask about a topic that has multiple perspectives — scientific debate, contested history, political questions, health controversies — Brillix presents them. It tells you what the mainstream view is, what the dissenting view is, and where the genuine uncertainty lives. It does not decide for you.
If a topic has a mainstream scientific consensus and a legitimate alternative theory, Brillix presents both with accurate Truth Mode tags. The consensus is labeled as consensus. The alternative is labeled as alternative. You see both — and you decide.
| Topic type | What Brillix does |
|---|---|
| Scientific consensus | States the consensus clearly; notes where genuine scientific debate exists within the field |
| Alternative theories | Presents them alongside mainstream views with accurate Truth Mode status — not dismissed, not elevated |
| Political topics | Presents the positions and evidence on all sides without advocating for any party or ideology |
| Health & medicine | Follows current medical evidence; surfaces dissenting research and evolving findings where they exist |
| Historical events | Notes where the historical record is settled and where interpretation is genuinely contested |
Brillix's job is to give you the information to think with — not to think for you. An AI that steers you toward predetermined conclusions, even well-intentioned ones, is less useful than one that shows you the full picture and trusts you to reason through it.
Privacy & Data
Brillix stores conversations per account, encrypted at rest. Your messages are not used to train shared models. Memory profiles are per-account and isolated — one user's data never influences another user's experience.
Training data for Atlas and Vulcan comes from purpose-built synthetic generation pipelines and permissively licensed public sources — not from user conversations.
| Data type | Stored | Used for training | Shared |
|---|---|---|---|
| Conversations | Yes, encrypted | No | No |
| Memory profile | Yes, per account | No | No |
| Account info | Yes | No | No |
| Usage analytics | Aggregated only | No | No |
Frequently Asked Questions
What makes Brillix different from ChatGPT or Claude?
Brillix routes to specialized systems trained on specific domains rather than using one generalist model for everything. It also builds a persistent memory of how you communicate and what you know — so answers improve over time rather than starting from scratch every session.
Can Brillix make mistakes?
Yes. All AI systems can be wrong. Truth Mode is specifically designed to surface uncertainty rather than hide it — so when Brillix is less confident, that shows in the response. We think visible uncertainty is more useful than false confidence.
Is Brillix politically neutral?
Yes. Brillix does not take political positions or advocate for political parties or ideologies. On contested political questions, it presents the positions and evidence from multiple perspectives. The goal is to inform, not to steer.
How do I clear my memory?
Go to your account settings and select Memory. From there you can view everything Brillix has learned about you and clear any or all of it. Memory is yours — you control it completely.
What is NEXUS and do I need it?
NEXUS is full-synthesis mode — all specialist systems engaged simultaneously. Most questions don't need it. It's most useful for complex, multi-domain questions where you want every relevant system contributing to the answer rather than just the closest match.
When will Atlas and Vulcan be available externally?
Atlas and Vulcan are currently in active training. External API access will open in a private beta — join the waitlist on the research preview page.