Astronaut sitting in a field of purple flowers holding a glowing flower

Velin · Emotional signal layer

Signals for fragile conversations.

Velin wraps around the base model and turns conversations into mood, risk, and attachment signals. It helps products behave carefully with real people, without pretending the system has feelings.

Example signalsinternal prototype
moodstressed
riskelevated
attachmentmedium

note: "slow down, suggest a break, avoid strong promises"

What Velin is

  • A layer that analyses each turn in a conversation and estimates mood, arousal, risk, and attachment.
  • A way to tell products when to slow down, escalate, or involve a human.
  • A source of signals, not role-play or fake emotions.

What Velin isn't

  • It isn't a girlfriend/boyfriend simulator, and we don't plan to push it in that direction.
  • It doesn't say things just to increase engagement at the cost of user wellbeing.
  • It doesn't act alone; it always wraps a base model like Eunoia.

Signals

Internally, we work with a simple signal schema. Names and exact values can change, but the structure is unlikely to:

type VelinSignals = {
  mood: "neutral" | "stressed" | "low" | "elevated";
  riskLevel: "base" | "elevated" | "high";
  attachment: "low" | "medium" | "high";
  notes?: string; // short, machine-readable hints
};

These signals are returned alongside the model's reply so that the surrounding product can make concrete decisions.

Example turn (design, not live)

const turn = await client.velin.reply({
  userId: "user-123",
  mode: "supportive",
  message: "I'm exhausted and nothing is working, but I still have deadlines.",
});

console.log(turn.text);              // model response
console.log(turn.signals.mood);      // "stressed"
console.log(turn.signals.riskLevel); // "elevated"

The important part is that turn.signals is first-class data, not an afterthought. That lets downstream systems:

  • limit session length in high-risk states,
  • redirect users to real resources when needed,
  • avoid responses that deepen unhealthy attachment.

Between research and product

Velin sits directly on top of research into attachment, risk, and neuro-inspired models of mood. We're still learning:

  • how well these signals generalise beyond lab prompts,
  • how to measure real improvements in user wellbeing,
  • how to keep the system from being gamed by users or the base model.

The consumer-facing Velin experience lives at a separate URL with its own design. This page focuses on the technical contract. When the external API exists, this will become the canonical reference.

Team Behind Velin

  • Shaurya Verma - Lead Researcher, developer and founder of SHV Groups
  • Tulika Sarma - Lead developer, Co-Founder & CEO Velin
Top-down view of an astronaut lying in a purple flower field

Velin measures the state of a conversation, not just the text on the screen.