Valoquent: Craft Conversation

Lantern Works, LLC

In-app purchases

About this game

A live feedback loop for the way you think out loud. Live voice conversations with AI re-creations of figures from history, scored as you talk on clarity, depth, follow-through, and presence. The conversation is the lesson. The meter is how you see it.

Every character is an AI re-creation grounded in that person's own writings. Not a recording, not a real person.

The Conversation Meter A live meter on screen scores your conversation as you speak. 22 behavioral signals per utterance: depth, follow-up, reciprocity, coherence, constructive challenge, troll detection, and more. Sub-millisecond updates, on the same screen as the avatar. Five zones: Surface Skim, Exploring, Engaged, Deep Dive, Flow State. Asymmetric momentum smoothing means bad turns hurt faster than good turns help. Built on published engagement-scoring methodology (Ghazarian et al., AAAI 2020): per-utterance engagement is reliably measurable and improves automatic metrics' agreement with human quality judgment.

Three-tier Memory Architecture Each character remembers what you talked about and brings up unfinished threads through structured memory and archival storage. Returning users start at tier-appropriate scores.

Two-axis Disclosure System 16 distinct relationship states per character. Global trust tier (guarded, warming, open, vulnerable) intersects per-topic depth (surface, candid, deep, raw). Three-layer gating: turn count, signal strength, and meter score must all align before a tier advances.

The Tapestry A constellation map of every topic you've explored with each figure. The deeper you go, the more the Tapestry reveals.

Mind Levels & Arc Score Mind Levels move with your average score across all conversations: Novice, Curious, Thinker, Scholar, Intellectual, Sage. Progression rewards depth, not session count, grounded in Self-Determination Theory (Deci and Ryan). Arc Score is the per-session number, calculated from four factors: Conversational Arc, Depth of Exchange, Engagement Quality, Vocabulary Richness. Relationship depth advances per figure: Acquaintance, Familiar, Conversant.

The Conversation Guide An in-app explainer for everything the meter measures. Nothing about the scoring system is hidden.

Custom Intelligence Layer Per Character Hand-built character by character, refined across hundreds of calibration runs. Aurelius does not sound like Curie. 150+ vocabulary terms per character. Each anchored in their own primary sources: Einstein from his 1905 relativity paper, Curie from her radium memoir, Aurelius from his Meditations. The knowledge base is curated, not generated.

Music Grounded in Learning-environment Research Home-screen ambient regulates arousal between sessions. Post-session chimes timed to the zone reveal.

Calibration Discipline Calibrated against a 48-scenario battery across three training tiers: local pipeline, production harness, full production.

Security & Privacy Your conversations are yours. Full in-app account deletion, a web-based deletion request page (all personal data removed within 30 days), secure secrets handling, and Google Sign-In are in place. SOC 2 controls in flight.

Nothing Here is Improvised Grounded in 50+ peer-reviewed studies in conversation, learning, and honest design. Every part of the app, from the meter to the memory to the music, is anchored in published research.

What’s new

Valoquent comes to Android! This build includes includes improved safety protections for sensitive conversations, response reporting, crisis resources, regional referral support, updated privacy disclosures, and Google Play privacy/age-range configuration.