The Behavior Graph: Why the Future of Connection Is Built on What You Actually Do
"Other apps ask what you like. Anima captures what you actually do. That behavioral truth is what makes real connection possible."
The intention-behavior gap
Ask someone what they are interested in and you will get a list: running, mindfulness, literature, design, travel, being a good friend, learning Spanish. Ask them how they actually spent last Tuesday and you will hear about a different day entirely. They checked email for two hours before their first meeting, sat through three calls they forgot by Thursday, scrolled a feed they did not enjoy, and went to bed later than they planned. The person and the profile are not the same person. Everyone who has ever filled out a dating app or a LinkedIn bio knows this intuitively, but the platforms do not.
This matters because real connection is not built on shared declarations. It is built on shared reality. The friend who knows what your Tuesday actually looked like, who has seen the version of you that did not live up to the aspiration, who still reaches out: that relationship is made of behavior, not of bio. Most digital relationships are built at the level of the bio because that is all the platforms can see. The deeper layer, the one where life actually happens, is invisible to them.
The consequence is a social internet that has grown enormous in scale and thin in meaning. People have thousands of connections and very few relationships. They follow people whose curated output they admire and rarely know how any of those people actually spend their time. The gap between profile and life is the design flaw of the entire category.
Anima started from a different premise. The product is a voice journaling life RPG. You talk about your day, and the app listens for how you actually spent your XP across seven stats. The output is a running record of your lived behavior, updated every day you speak. That record is the behavior graph. It exists because of the primary product, not as a bolt-on. And it turns out to be the missing substrate for a different kind of connection.
The behavior graph: what Anima actually captures
Every entry in Anima is a few minutes of spoken reflection on your day. You are not filling out a form, checking a box, or selecting an interest. You are describing what happened. The app's job is to listen for real activity: did you train, did you rest, did you read, did you create, did you spend time with another person, did you notice your own inner state, did you reflect on the shape of the day. The seven stats are the framework, but the signal comes from what you actually said you did.
Because the signal is behavioral, it has properties that a self-reported profile cannot have. It is continuous, so it captures trends rather than snapshots. It is honest about effort, because a week where you said you would train four times and actually trained once looks like what it is. It is comprehensive across life, not specialized to a single vertical like fitness or productivity. And because the raw input is voice journaling that the user chose to record for themselves, the resulting signal is the closest thing to ground truth about a person's life that any app has ever held.
The seven-stat shape is the unit of identity on the behavior graph. Two people with identical interests on paper can have very different shapes. A self-described creative who has not made anything in three months has a different Creativity velocity than a so-called analyst who paints on weekends. A person who describes themselves as social but whose Empathy stat is quiet for weeks is a different entity on the behavior graph than a quiet introvert who calls their sister every Sunday. The shape tells the truth that the self-description obscures.
This is not surveillance. The raw voice is the user's own private journal, captured voluntarily for their own use. The stat signal is the byproduct, computed locally on behalf of the person and displayed back to them. The behavior graph, when the user chooses to participate, shares only the stat-level layer, never the raw recording. What is captured belongs to the person, and what is shared is a small, abstract, aggregate view.
Three connection primitives built on behavior
The behavior graph unlocks three new ways to be connected to other people. They are simple to describe and hard to fake, which is the point. Each corresponds to a specific kind of relationship that the existing graphs cannot produce because they cannot see the underlying data.
Mirrors: people living like you
A Mirror is a person whose recent stat distribution resembles yours. Not someone who says they like the same things, but someone whose actual weeks have the same shape. If you have trained four times this week, read most evenings, spent real time with your partner, and journaled consistently, your Mirror is another person whose week looked like that.
The experience of finding a Mirror is hard to get from any other product. It is the intimacy of being seen without explaining. You do not have to introduce yourself, justify your rhythms, or convince the other person that what you value is valuable. They are already living it. The shape of their week is the shape of yours. Mirrors are not about becoming closer friends by default. They are about the relief of knowing that the life you are trying to live is a life other people are also trying to live.
The philosophy here is important. The opposite of performance is recognition. Social media rewards the former and starves the latter. Mirrors reverse that. Nothing is performed, because the signal was collected as a byproduct of private reflection. Nothing is curated, because the stat distribution is just the truth of the weeks that happened. What is recognized is the lived pattern, and the user does not even have to speak to be seen.
Guides: people strong where you are weak
A Guide is a person who has opted in to mentor others in a specific stat where they are notably strong. If your Empathy sits at tier 2 and theirs is at tier 8, and they have volunteered to teach in that dimension, the system can introduce you. The match is specific. It is not a generic mentor, it is a person whose actual behavior in that stat has been consistent enough, long enough, to offer something real.
What makes Guides different from the mentor relationships on existing platforms is that the qualification is behavioral, not credentialed. A LinkedIn mentor is someone whose title or company says they should be qualified. A Guide on the behavior graph is someone whose weeks, over months, have demonstrated the thing you are trying to learn. There is no CV. There is a stat history. And because the Guide chose to be discoverable in that specific dimension, the consent is explicit and the scope is narrow.
The relationship is reciprocal by design. The same person who guides you in Empathy may be a student of yours in Strength or Creativity. The behavior graph is high-dimensional, and almost no one is strong in every dimension at once. This keeps the dynamic from calcifying into a permanent hierarchy of teacher and student. It is closer to the peer guild of a good dojo than to the one-way professional mentorship of most networks.
Twins and complementary pairs
A Twin is a pact between two people whose strengths cover different dimensions. Your high Strength combined with their high Empathy forms a growth pair where each of you can lean on the other to build the dimension you are weaker in. You are not cloning one person's life. You are forming a two-person system where complementary shapes reinforce each other.
The twin dynamic replaces the most common failure mode of accountability partnerships, which is the checklist relationship. Traditional accountability is one person reporting their habits to another and being gently scolded when they fall behind. That structure tends to die after a few weeks because it is one-directional and mildly humiliating. A Twin pairing avoids this because both people are simultaneously a guide and a student. The structure is mutual and the stakes are shared.
Twins also produce something rare in adult life: a partner for a specific stretch of growth. Most friendships are about the relationship itself, which is good and necessary. A Twin relationship is different. It is time-bound, intentional, and scoped to a mutual project of becoming. When the project is done, the relationship may continue as a friendship or quietly return to its baseline. Neither outcome is a failure. The structure was honest from the start about what it was for.
See your shape first
Before the network comes the mirror. Take the quiz to see which of the seven stats drives your character today.
Find Your CharacterThe privacy model: opt-in, stat-level, no raw data
The behavior graph is powerful precisely because the signal is honest, which means the privacy model has to be equally honest. A product that asked people to journal into a microphone and then quietly fed the raw content back into a social feed would destroy trust instantly and deserve to. The model Anima is committing to is the opposite of that.
The first rule is that the core product works without the network. You can use Anima for years as a private voice journal and never encounter another user. The seven stats, the mandala, the character system, the long-term graph of your own life: all of it is yours, alone, by default. The network is a separate feature that a user has to reach for and switch on. If the user never switches it on, the product still does its primary job.
The second rule is that the network never sees raw voice or transcripts. What leaves the device, only after explicit opt-in, is a derived layer: your current tier in a given stat, your recent velocity, and a coarse timestamp of activity. That is the minimum information needed to match two behavior patterns. Anything more would be surveillance dressed as connection. The design discipline is to keep the sharing layer as small as the matching layer requires, and no larger.
The third rule is that disclosure is tiered and granular. A user can choose to be discoverable as a Mirror without being discoverable as a Guide. They can volunteer as a Guide in one specific stat without exposing their full seven-stat shape. They can pair with a Twin without being visible to the general network. Every level of participation is a separate decision, reversible, and expressed in the most specific way the feature permits. This is not the flat privacy of most social products, which is either public or invisible. It is a ladder of choices, and each rung is meaningful.
The fourth rule is consent for introduction. No Mirror, Guide, or Twin pairing is formed without both parties agreeing to it. The system can propose a match, but a connection only becomes real when both sides accept. That keeps the behavior graph from feeling like a dragnet, even at scale. It is a network of invited relationships, not an algorithmic imposition.
Why this cannot be built on top of existing graphs
It is tempting to imagine that the behavior graph could be built as a layer on top of an existing network. Take Instagram's data, add a behavioral model, produce the seven-stat shape. In practice this is not possible, not because of technical limits but because of what the underlying signals actually measure.
Facebook and Instagram record the version of your life you chose to post. That is a performance graph. It captures peaks, not baselines. A person who is quietly consistent with their training gets no signal on Instagram. A person who posts a single gym selfie per month dominates the signal. The platform rewards edited life, not lived life, and no amount of modeling on top of that stream will recover the behavior graph underneath. The data is wrong at the source.
LinkedIn records a professional credential graph. Job titles, endorsements, companies, certifications. It is useful for hiring and for professional reputation. It is silent on almost everything else. A LinkedIn profile cannot tell you whether a person has been present with their partner this month, whether they have moved their body, whether they have read anything that changed them. Those are not weaknesses of the product. They are deliberate scope decisions. LinkedIn is not trying to be a behavior graph, and it could not become one without destroying its professional utility.
Dating apps like Bumble and Hinge record stated preferences for romantic fit. They are built on exactly the kind of self-declared interest that the intention-behavior gap undermines. The platforms know this, which is why their matching algorithms quietly lean on implicit behavior, like who you actually swiped on, rather than on what you said you wanted. But even that implicit signal is a narrow slice: swipe behavior within the app. It says nothing about how you spend your Tuesday, how you treat your friends, how you move through a real week.
The behavior graph requires a product whose primary purpose is to observe lived behavior across all seven dimensions, continuously, in a way that the user has consented to because the observation is useful to them first. Voice journaling fits this shape. The user is the primary beneficiary of the reflection. The seven-stat signal is useful to them every day, whether they ever connect with another user or not. The data is right at the source because the source is a person telling the truth to themselves. A performance network cannot produce that. A credential network cannot produce that. A dating network cannot produce that. The behavior graph has to be grown from different soil.
Roadmap: today, next, later
The roadmap is deliberately slow, because the trust model is the product. A behavior graph that rushes to add social features would become the surveillance layer we said we would not build. The sequencing below trades speed for fidelity.
Today: the anonymous Mirror. Every Anima user can already see a de-identified distribution of all users on each stat. You can locate yourself within that distribution: what percentile you are at in Strength, where you sit on the Empathy curve, how your Creativity velocity compares. No individual user is visible. No connection is proposed. The only thing you see is where your shape fits inside the shape of the whole community. Even this version of Mirror produces something rare, the feeling of not being alone in the pattern you are living.
Next: named Guides. The next step is to let a user invite a specific other user to guide them in one stat. The inviter has to reach out. The invitee has to accept. The scope of the relationship is narrow, just the one stat, and the duration is whatever both people want it to be. The system's job is to suggest compatible Guides based on the user's current gaps and on the pool of users who have opted in to guide in that dimension. The rest is between two people.
Later: complementary Twins. The final layer is the Twin pairing, and it is the most demanding to get right. A Twin relationship is a mutual growth pact between two users whose strengths complement each other. The system has to identify compatible pairs, respect each person's opt-in state in multiple dimensions, and provide a container for the relationship that makes it easy to be honest without it becoming another inbox. This is a deep design problem and it will not ship until the Mirror and Guide layers have produced enough trust for people to want it.
No dates are attached to these stages, because dates in a roadmap are promises a small team cannot always keep. What is being committed to is the order: Mirror first, Guide second, Twin last. Each step adds power to the user and requires a new kind of consent from them. If any step cannot be shipped in a way that preserves the privacy model, it will not ship.