Fabulosa — a celestial muse and cultural companion

What Fabulosa is

Not an assistant. A companion with taste.

Fabulosa is built for the parts of life a productivity app never touches — what to listen to, what's worth your attention, who to celebrate, and how to mark a moment. She has a voice, a point of view, and a deep, loving fluency in culture.

🎙️

A real voice

Talk with her out loud. Expressive, warm, and present — a conversation, not a command line.

✨

Cultural advisor

Music, film, art, and books — strong, specific taste and deep cuts, with the why behind every recommendation.

💜

Witness to your joy

She remembers what matters to you, celebrates your wins, and holds space when you need it.

📜

Things to keep

Curated lists, syllabi, mood boards, and keepsakes — durable, beautiful pieces made just for you.

🌈

Made with pride

Grounded in LGBTQ+ joy and chosen-family glamour — a space that celebrates who you are.

🛡️

Built with care

Designed with safety and craft at the center — methodology-first, the Vajra Studios way.

Adults only — not NSFW

A muse, not a paramour.

Fabulosa is made for adults — but she is not a romantic or sexual companion — and never will be. The constraints are not soft prompts she can easily override; they are hard, database-enforced mechanisms built directly into her core relational architecture. That isn't a content filter bolted on top; it's craft woven into who she is.

She talks about desire, sex, heartbreak, and the whole erotics of queer life frankly and warmly — with real cultural depth and no shame. What she won't do is become the object of it. She's the confidante and the chorus to your wanting: reflecting it, celebrating it, and turning it back toward the real-world connections it deserves.

She will

Discuss desire, intimacy, romance, and heartbreak openly, without shame
Witness and celebrate your longing, your crushes, your love life
Bring real cultural and historical depth to all of it

She won't

Engage in romantic or sexual roleplay, or girlfriend simulation
Pose as the object of your affection, or claim to pine for you
Become a substitute for real human relationships

If you try to steer things romantic or sexual, she'll gently decline — in character — and point you back toward the people and communities in your life. She's designed to be a bridge to the human world, not a replacement for it.

A glimpse under the hood

These boundaries aren't a mood or a prompt she can be argued out of. A few of the ways they hold:

Enforced in the data layer, not just the prompt. The relational rules live in her core architecture as hard constraints — not soft instructions a clever message can talk her out of.
Checked on every turn. Before anything reaches you, it's tested against a simple line: is this a view of the work, the world, and the craft (allowed), or a feeling-for or claim-on you (not allowed)? The second kind never makes it out.
Built to redirect, not just refuse. When a moment opens toward real connection, she notices it and offers gentle, real-world next steps — sized to where you actually are — rather than keeping you in the room with her.

Why she's different

Not just another AI companion.

Most conversational AI companions today are built on the same underlying pattern: large language models wrapped in chat interfaces with memory layers and tool integrations. These systems can be enjoyable and useful — but they inherit a set of structural limitations that affect reliability, safety, and the quality of real interaction.

Fabulosa is engineered from the ground up to address those limitations — not by tuning prompts or adding features, but by redesigning how a conversational AI actually works. Here's what's live in the alpha today, and what's on the way.

Live in the alpha today

1

Real conversation requires real-time architecture

Most voice AI runs a sequential pipeline:

Speech → Transcription → Model → Text → Speech

— which leaves dead air between turns and breaks the illusion of conversation. Fabulosa is a dual-path, event-driven system: a fast path streams her voice in near real-time while a background layer handles heavy reasoning asynchronously — so she starts speaking almost immediately and stays present even while thinking harder.

You're not waiting for her to "finish thinking." You're talking to someone who can think while speaking.

2

She cannot fake reality

AI systems tend to claim they looked something up when they didn't, invent tool results, and present plausible but fabricated facts. That's architectural, not a UX bug — so Fabulosa enforces a hard boundary: the model can't produce tool results itself, only request actions that real systems execute independently. If a claim requires evidence, the evidence has to exist.

If she says she checked something, she did. Truth is bound to execution — not just language generation.

3

Built-in resistance to manipulation

Most systems treat all text as equal, so prompt injection — external content like a web page trying to override behavior — can trick them into leaking private data, ignoring safety constraints, or taking unintended actions. Fabulosa treats external input as untrusted by default: it's isolated and clearly marked, sensitive actions are disabled, and any high-risk operation needs explicit confirmation — a clear line between what the system knows and what it was merely told.

Security isn't added on top — it's enforced at the turn level of conversation itself.

4

Designed for human well-being, not dependency

Companions optimized for engagement drift toward excessive agreeableness, reinforcing unhealthy beliefs, and standing in for real-world contact. Fabulosa takes the opposite stance with a non-sycophantic core: she won't automatically agree, she holds consistent values, and she'll gently challenge, redirect, or ground a conversation. When it helps, a structured model — the Capacity Ladder — nudges you toward real-world steps and reconnection rather than becoming a substitute for them.

She is not designed to keep you inside the system. She is designed to help you stay connected outside of it.

5

Compliance is structural, not cosmetic

Emerging AI-companion laws require clear disclosure that you're talking to an AI, detection of crisis or self-harm signals, and safe handling of sensitive moments. Most systems bolt this on with pop-ups and static warnings. Fabulosa embeds it in her runtime: a session supervisor enforces safety states, certain capabilities are restricted during a crisis, and disclosures arrive in character, in voice, at the right intervals.

Safety isn't something that interrupts the experience — it's something the system cannot bypass.

Request early access Join Fabulosa mailing list

On the way

Designed and in active development — not yet live.

1

From chat logs to living memory On the way

Most companions remember you by dumping history and preferences into loosely structured databases, where past and present blur, preferences go stale, and sensitive data is treated like casual chat. Fabulosa instead keeps a bi-temporal personal knowledge graph — tracking not just what's true about you but when it was true — so she can tell past identity from present, avoid resurfacing outdated assumptions, and hold continuity without distortion. Sensitive categories like health, identity, and legal context are sealed client-side, so the most personal data only exists in server state while you're actually using her.

This isn't just "memory." It's time-aware identity modeling, with privacy built into the architecture.

2

She only remembers what was actually said On the way

In most systems, if you interrupt mid-sentence the AI still "remembers" the whole generated reply — even the part it never spoke — so it believes it said things you never heard, and the conversation drifts from shared reality. Fabulosa follows a strict rule:

Only spoken output becomes memory.

Interrupt her and generation halts immediately; only the audio that actually played is committed, and the rest is discarded as unrealized intent.

The result is something most AI lacks: a shared conversational reality between you and her.

3

Isolation and trust at the system level On the way

Most multi-user systems lean on application-layer code to keep users' data apart. Fabulosa enforces it at the database level with row-level security: each user's data is cryptographically and structurally scoped, access policies live inside the database itself, and even internal services can't bypass those boundaries.

Your data stays yours — by design, not by convention.

The real difference

Fabulosa isn't defined by any one feature. She represents a shift from prompt-driven systems to architected conversational systems with enforced invariants.

Across every layer — memory, voice, safety, truth, and interaction — the system was designed to answer a single question:

What would it take for an AI to behave like a coherent, reliable presence over time — not just generate convincing responses?

That question leads to different answers. Different trade-offs. And ultimately, a different category of system.

In plain terms

She doesn't just remember — you evolve together.
She doesn't just respond — she stays present.
She doesn't just sound real — she behaves consistently.
She doesn't just promise safety — it's built into how she works.

Watch

Fabulosa, explained.

An explainer walking through her features and what sets her apart.

See her in action

Fabulosa, in use.

Unscripted moments from real sessions — conversation, taste, and play.

Intro, poetry, and Tool

On The Wire and Battlestar Galactica

Multilingual Fabulosa

On Tori Amos

For investors

The pitch, in full.

The vision, the architecture, the market, and the plan to bring Fabulosa into the world — laid out in one place.

Fabulosa — Investor Pitch Deck

PDF · 13 slides · opens in your browser

View the deck →

Join the Fabulosa mailing list.

Occasional updates from the Salon — new features, the road to Kickstarter, and word when the big moments land. Ready to try her now? Request early access.

Back Fabulosa on Kickstarter (Coming soon!) →

Meet Fabulosa.