Every AI travel assistant promises to make your trips better. Most of them do it by talking too much, knowing too little about what actually matters, and treating your memories like data to optimize. We built Rocky to be the opposite.
Rocky is the AI companion inside Cairn Memories. It's powered by Anthropic's Claude. And the single hardest design challenge we faced wasn't making it smart — it was making it quiet enough to be trustworthy and warm enough to matter.
That tension — between presence and absence, between helpful and intrusive — is the central design problem of any AI travel companion. And most apps get it wrong in the same predictable ways.
The problem with most AI travel assistants
Open any AI trip assistant and the pattern is familiar. The AI wants to talk. It wants to plan your itinerary, suggest restaurants, notify you about attractions nearby, surface tips from other travelers, and generally make itself the center of your experience.
The assumption is that more AI means more value. That the AI travel companion should be companion-forward — always present, always suggesting, always optimizing.
But think about the best travel experiences you've ever had. How many of them involved an app telling you what to do next? How many of them were optimized by an algorithm? The moments that define a trip — the unexpected conversation, the wrong turn that led somewhere beautiful, the quiet morning when you just sat and watched a city wake up — those happen precisely because nobody was directing the experience.
The best AI travel companion isn't one that talks the most. It's one that knows when to talk and, more importantly, when to shut up.
The campfire companion
Early in the design process, we kept circling back to a metaphor that eventually became our north star: the campfire companion.
Imagine you're sitting around a campfire at the end of a long day of travel. Someone next to you — not the loudest person in the group, not the one performing for the crowd — leans over and asks a question. Not a generic question. A specific one. What was going through your mind when you saw the valley for the first time today?
And that question draws out a story you wouldn't have told otherwise. Not because you were hiding it, but because you hadn't thought to articulate it yet. The question created the space for something real to surface.
That's Rocky. Not the tour guide. Not the concierge. Not the algorithm optimizing your trip. The person sitting next to you at the campfire who occasionally asks a good question.
The difference matters enormously. A tour guide tells you what to see. A concierge tells you where to go. Rocky asks what you felt. The first two generate content about a destination. The last one generates memory — the kind that actually persists across years and generations.
Active during journeys, dormant between them
One of the most counterintuitive decisions we made was to design Rocky to be dormant between journeys. When you're not traveling, Rocky isn't pinging you. It isn't sending weekly recaps or "on this day" notifications. It isn't trying to get you to open the app.
This is intentional, not a limitation.
Most AI travel planning tools optimize for engagement — daily active users, session length, notification open rates. The incentive structure pushes the AI to be constantly present, constantly nudging, constantly generating reasons for you to come back.
But Cairn Memories isn't a social media platform. It's a memory system. And memories don't need an AI prodding them every Tuesday. They need to be captured well in the moment and then delivered at the right time — through Location Keys and Time Keys — not on an engagement schedule.
Rocky activates when you begin a journey. It's present while you're building your rock — capturing moments, adding context, recording the stories behind the photos. And when the journey ends, Rocky steps back. Your memories are complete. They belong to you. They don't need an AI hovering over them to have value.
This decision cost us a vanity metric. An always-on AI would drive higher daily engagement numbers. But we're not building for engagement metrics. We're building for memory quality. And memory quality requires an AI that respects the boundary between "I'm traveling and could use a companion" and "I'm living my life and don't need one."
Three density settings: minimal, balanced, rich
People have wildly different relationships with AI. Some want a conversational partner that helps them narrate their entire journey. Others want the AI to be nearly invisible — just capture the ambient layer and stay out of the way. Most people fall somewhere in between, and their preference shifts depending on the trip.
Rocky's density settings address this directly:
Minimal. Rocky captures ambient context automatically — weather, location data, time of day, local events — but rarely surfaces prompts. You're in control. Rocky is essentially a quiet observer, enriching your moments with context you'd otherwise lose but never interrupting the experience. This is the setting for the traveler who wants to be fully present and deal with memories later.
Balanced. Rocky offers occasional prompts and suggestions. After you capture a moment, Rocky might ask a follow-up question: What made you stop here? or Who were you with when this happened? These prompts are designed to draw out the story behind the moment — the context a photo can't capture. You can engage or ignore them. No consequence either way.
Rich. Rocky is a more active companion. It offers more prompts, suggests connections between moments, helps you build narrative threads through your journey, and may offer to draft descriptions or summaries based on what you've shared. This is the setting for the traveler who wants help turning raw experience into a complete, structured memory.
The key design principle: every setting is valid. Minimal isn't "less than" rich. It's a different relationship with the AI, and the memories you create are equally valuable regardless of how much Rocky contributed. We never penalize someone for wanting less AI involvement, and we never push someone toward more.
What Rocky does not do
Defining what an AI travel companion shouldn't do turned out to be more important than defining what it should. Here's where we drew the lines:
Rocky does not spam notifications. No "You're near a popular landmark!" alerts. No daily digests. No push notifications designed to drive app opens. When Rocky communicates, it's because you're actively on a journey and the prompt is contextually relevant to what you're doing right now.
Rocky does not generate content without permission. Every piece of AI-generated content — whether it's a suggested description, a narrative summary, or a connection between moments — is opt-in. Rocky might offer: Want me to draft a summary of today's journey? But it will never generate that summary and present it as your memory without asking. Your rock is yours. Rocky contributes only when invited.
Rocky does not share data between users. Your moments, your stories, your locations — none of that is visible to other users unless you explicitly share it by placing your rock on a Shared Cairn or Event Cairn. Rocky doesn't aggregate data across users, doesn't surface "what other travelers did here," and doesn't treat your memories as training data.
Rocky does not optimize for engagement. There is no algorithmic feed. There are no streaks. There are no gamified incentives to capture more moments or use the AI more. Rocky's success metric is memory quality — specifically, whether the moments you captured are rich enough to be meaningful when you or your family revisit them in five, ten, or thirty years.
Rocky does not pretend to be human. Every AI-assisted element in Cairn Memories is transparently labeled. If Rocky helped draft a description, it says so. If the ambient layer was populated by AI, it says so. The trust equation is simple: you should always know what was written by a person and what was suggested by an AI. Blurring that line would undermine the entire purpose of a memory platform.
The trust equation
Trust is the only currency that matters for an AI travel companion. If you don't trust the AI, you won't share real stories with it. If you won't share real stories, the memories it helps you capture will be shallow. And shallow memories don't persist across generations — which is the entire point of Cairn Memories.
We arrived at a principle we call the trust equation: every AI feature earns trust by being transparent about what it is.
This sounds obvious. It isn't. The industry trend is toward seamlessness — making AI so integrated that you can't tell where the human ends and the AI begins. That's fine for a writing assistant or a code editor. It's dangerous for a memory platform.
When your granddaughter reads the rock you built from a trip to Kyoto, she should know which words are yours and which were suggested by an AI. Not because the AI's suggestions are less valuable, but because the distinction matters for the integrity of the memory. Your voice is the thing being preserved. Rocky is the tool that helped you articulate it. Collapsing those two things into one undermines both.
So we label everything. AI-suggested descriptions are marked as AI-assisted. Ambient layer data is clearly separated from your personal narrative. Draft summaries are presented as drafts, not finished products. The seams are visible on purpose.
Why Rocky is powered by Claude
We chose Anthropic's Claude as the foundation for Rocky for a specific reason: alignment with our design values.
Claude is designed to be helpful, harmless, and honest. That maps directly to what a memory companion needs to be. Helpful — able to ask the right question at the right time, enrich context, and support storytelling. Harmless — never generating content that distorts memories or pressures users. Honest — transparent about its limitations, clear about what it knows and what it's guessing.
The alternative would be building on a model optimized for engagement, persuasion, or content generation volume. Those are fine objectives for other products. For a multigenerational memory platform, they're the wrong objectives entirely. We needed an AI foundation that could be genuinely quiet — that didn't have a built-in incentive to maximize output.
Rocky's real job
Strip away the technical details and Rocky's job is actually very simple: help you say the thing you wouldn't have said otherwise.
Most people don't narrate their experiences. They take a photo and move on. The story — why this moment mattered, what they were feeling, who they were thinking about — stays locked in their head. It fades. And eventually it disappears.
Rocky exists to create small openings for those stories to come out. A well-timed question. A gentle prompt. A suggestion to add a voice note while the feeling is still fresh. These aren't sophisticated AI tricks. They're the digital equivalent of what the best conversationalists do naturally: create space for other people to share something real.
The design philosophy behind Rocky isn't about what AI can do. It's about what people need. And what people need, when they're somewhere meaningful with people they love, isn't an algorithm. It's a nudge to pay attention. A reminder that this moment — this one right here — is worth remembering.
Rocky is the nudge. Nothing more, nothing less.
What this means for the future of AI travel companions
We think the next wave of AI travel assistants will split into two camps. One camp will keep optimizing for engagement — more notifications, more suggestions, more AI-generated content, more reasons to stay in the app. That camp will win on metrics and lose on meaning.
The other camp will optimize for what we're optimizing for: memory quality. AI that earns trust through restraint. AI that knows its role is to support human experience, not replace it. AI that measures success not by how much it contributed, but by whether the memories it helped capture are still meaningful decades from now.
Rocky is our bet on the second camp. A quiet companion. A campfire question-asker. An AI that does its best work by knowing when to stay silent.
That's what an AI travel companion should do. And more importantly, that's what it shouldn't.