Give visitors a host they want to spend time with
Characters visitors genuinely want to talk to keep them engaged longer and bring them back. Every destination gets a cast of voiced AI personas with their own backstory, personality and memory.
A living host for every place
A Roman centurion at the forum. A Victorian inventor in the workshop. A local legend on the harbour wall. Each character carries a defined backstory, personality and speaking style, so the same fact lands completely differently depending on who's telling it.
Characters greet visitors, narrate stops, set missions and answer questions in real time — staying in persona throughout, because their identity and the destination's content live in a vector store the AI reasons over.
"Curiosity is the only tool you can't buy." Eleanor ran a clockwork workshop in 1880s and lost three patents to thieves — she'll tell you about each one if you find the right stop.
Four technologies, one believable host
Backstory, voice, face and memory combine so characters feel less like an app feature and more like a guide who's genuinely there with you.
Persona & backstory
Define each character's history, personality and speaking style in the admin. That identity drives every line they speak and keeps tone consistent across an entire tour.
ElevenLabs voice narration
Natural, characterful voice for every stop and reply. Choose accent and tone per character, then batch-generate narration for the whole tour from the admin.
HeyGen avatar intros
Short avatar video intros put a face to each character at the start of a tour or stop — generated in batches, no film crew required.
Qdrant semantic memory
Real-time chat grounded in a vector database, so characters recall what a visitor has seen and answer from the destination's real content — not generic guesswork.
Conversations that actually go somewhere
Visitors ask anything — and the character answers from the destination's real content, in voice and in character. Semantic memory means follow-up questions build on what came before instead of resetting every time.
Characters can nudge visitors toward the next stop, drop hints about hidden content and react to context like mood and time of day — turning a one-way guide into a back-and-forth that keeps people exploring.
See how chat ties into rewardsMemory, relationships and mood
Three systems most "AI guides" don't have — and the reason Wander characters feel like people, not chatbots.
Semantic memory
Responses are pre-indexed in Qdrant (text-embedding-3-small, 0.65 similarity); matched answers return instantly, with a live GPT-4o-mini fallback otherwise. Characters store and recall a visitor's name and details across visits. Why it matters: "good to see you again" is the hook that drives the return visit.
Character relationships
Characters hold opinions of each other across 8 types — trust, distrust, rival, admire, fear, contempt, affection, neutral — each with a 0–100 strength that drives whether they warn about or recommend another. Why it matters: a living social web makes a destination's cast feel like a community.
Mood system
9 moods shift with weather, time of day and how the conversation is going, colouring tone and what a character will share. Why it matters: the same stop feels different on a rainy evening than a sunny morning — a reason to come back.
Characters built for scale
Every persona is generated, voiced and indexed in one pass — then talks back in real time with memory.
Book a demoCommon questions about characters
Each persona is grounded in your destination's real content, indexed in a vector store, with defined anachronism rules so a Victorian character won't mention smartphones. Live answers fall back to GPT-4o-mini constrained by the same persona and content.
Yes. A conversation-memory service stores a visitor's name and details, so on a return visit the character can greet them by name and reference earlier questions — the foundation of repeat-visit engagement.
Voices use ElevenLabs (a distinct voice per character) and avatar video intros use HeyGen at 1080p. Both are batch-generated from the admin — no recording booth or film crew required.
Batch generation produces roughly 300 in-persona Q&A responses plus matching audio and Qdrant indexing in a single job, so a fully voiced character can be ready in a day rather than weeks.
Cast the perfect guide for your destination
We'll help you design characters, voices and stories that fit your place. Book a demo to see them in action.