Edge vs cloud: design trade-offs for Voco, a privacy-first voice assistant

AI · 6 min read

Voco entered a crowded assistant market with a privacy promise: local-first voice processing. Early technical prototypes showed that on-device ASR and intent classification reduced data leakage but struggled with large language tasks and multi-turn context beyond a short window. Cloud processing gave richer responses and continued context, but undermined the privacy positioning that differentiated the product.

Designers and engineers developed a hybrid model where local models handle wake-word detection, private commands, and short-context tasks, while opt-in cloud extensions power richer capabilities like long-form summarization and advanced planning. The UX surfaced this trade-off transparently: when a request required cloud resources, Voco explains what will be sent, why, and provides a one-tap preview of the data. Defaults favor local processing, and users can set per-skill preferences in a privacy dashboard.

This architectural choice influenced product discovery and pricing: the team introduced a free tier with fully local features and a subscription that unlocks cloud-enhanced skills. Early metrics indicated strong conversion among privacy-conscious early adopters and lower latency-related drop-offs in core interactions. Voco's approach illustrates how startup product design must make trade-offs explicit in the UX, aligning technical constraints with user expectations rather than hiding them behind marketing language.