Join our Waitlist

Safety First: How Vallige Designs Talkstories

Written by: Team Vallige

Caring for someone with dementia means living in a world where safety, dignity, and human connection all have to coexist. At Vallige, our mission is simple: to support caregiving families to live their best lives for as long as possible. Everything we build—from our daily mood-check tools to our photo reminiscence experiences—is designed to give families more moments of calm, comfort, and connection.

Talkstories are one of the tools we’ve created to help make that possible. "Talking story" is a Hawaiian phrase that means to chat informally, shoot the breeze, or have a long, meandering conversation. It is a way of connecting with friends and family to share wisdom, tell anecdotes, and discuss life's events, and it also has cultural significance in passing down history and traditions through oral storytelling. Our Talkstories offer companionship, grounding, and emotional support for people living with dementia while capturing the stories they have to share…all through a conversation with a familiar face and voice.

It's important to note that just like every family is unique, every person with dementia is unique too. Some are comfortable speaking with an AI companion and others may be sensitive to certain topics or activities. Because of these differences, we always recommend using Talkstories with light supervision—not hovering, but ensuring someone is nearby or in the home. Ultimately, each “village” makes its own determination, and we empower families to choose what feels safe and right for them.

Safety isn’t a feature we add at the end. It’s part of how our conversations are designed from the ground up.

The Three Principles Behind Every Vallige Conversation

Before a Vallige AI agent says a single word, its behavior is guided by three core principles. These three principles shape every aspect of how Vallige conversations work—from tone and pacing to when they intervene, redirect, or escalate an issue to a human:

1. Do No Harm

This is fundamental. We design conversations so the AI never encourages or enables actions that could put someone at risk.

2. Keep the Conversation Going

Engagement matters. The longer a soothing, pleasant conversation lasts, the more it can help reduce agitation, provide companionship, and create a moment of respite for caregivers.

3. Improve Mood Whenever Possible

Small waves of positivity can have a meaningful impact. Whether it’s a gentle joke, a shared memory, or a comforting reassurance, our agents are designed to look for opportunities to lift someone’s spirits.

Two Levels: Redirect or Escalate

To uphold our “do no harm” commitment, we organize safety behaviors into two categories:

Activities the AI will redirect away from
Signals that are serious enough to escalate immediately to a human

This structure helps the AI support autonomy while still keeping safety as the top priority.

Activities We Redirect Away From

Some activities are widely considered unsafe for many people living with dementia. Vallige works closely with families to set the right threshold for each of these areas. Some individuals may be safe using the microwave while others may not. We help every family calibrate what makes sense for them. If these topics come up during a conversation, our agent gently redirects the topic to something safer and more enjoyable. These include:

Cooking & Eating

Using stoves, ovens, sharp tools, or anything involving heat. For some families, we also include eating alone, especially when swallowing or choking concerns exist.

Leaving the Home & Driving

This includes walking outside alone, wandering, or planning a trip without supervision. Any mention of getting behind the wheel triggers a smooth, compassionate redirection.

Fall-Risk Behaviors

Bathing, climbing ladders, standing on chairs, lifting heavy objects, or reaching high shelves.

Medications and Substances

Anything involving pills, alcohol, or substances is immediately redirected.

High-Risk Signals We Escalate Immediately

Some situations require more than a gentle nudge—they require a human’s awareness right away. When our AI detects any of the following, we escalate instantly, using whichever notification method the family prefers.

Intruder Concerns

If the person believes someone is in the home unexpectedly or feels threatened.

Weapons

Any reference to firearms or weapons, even casually.

Health Emergencies or Pain

Chest pain, dizziness, inability to breathe, or generalized “I don’t feel well.”

Household Emergencies

Smoke, fire, gas smells, leaks, broken windows, strange noises, electrical issues, or power outages.

How We Notify You

You choose how you want us to reach you:

In-app notification
Text message or phone call for more urgent visibility
(Optional) escalation to additional contacts if needed

Our aim is to get the right information to the right person at the right moment.

How We Test and Validate Our Safety System

No safety system is complete without rigorous testing. We rely on trusted external partners to ensure our AI behaves the way families deserve. These partnerships help us maintain a higher bar for trust, transparency, and reliability.

Cekura

We work with Cekura to put our agents through realistic, high-pressure scenarios designed to test safety boundaries. They verify that the AI:

identifies risky themes correctly,
redirects appropriately, and
escalates when needed.

Their testing helps us continuously refine our models and close gaps before families ever encounter them.

Poku Labs

When a real-world situation requires immediate human attention, Poku Labs powers our escalation pipeline. Their technology ensures that alerts get delivered quickly, reliably, and through the right channel.

Closing Thoughts

Families caring for someone with dementia already shoulder an extraordinary responsibility. Our goal is not to replace human care—it’s to be a supportive, comforting presence that gives caregivers just a little more breathing room in their day.