Silo – Private AI Chat App

A privacy-first AI chat app with end-to-end encryption and multi-model access. 35k+ downloads through ASO and editorial placements, zero ad spend.

Role: Product Designer (full end-to-end on iOS/Web)

By Bishal Mishra, Product Designer

Private AI Chat

A privacy-first AI chat app available on iOS and web

Silo gives users access to multiple AI models with full end-to-end encryption. No logging, no training, no data collection. The core design challenge: privacy products have a trust problem.

Outcomes

I shipped Silo on the iOS App Store with zero data collection and full end to end encryption, grew it to 35k plus downloads through ASO and editorial placements.

Challenges

Privacy products face a credibility gap. Even when you clearly state "encrypted" and "private" in the interface, users often remain skeptical. With Silo, the challenge was to make security feel effortless rather than intimidating, while giving access to multiple AI models without exposing unnecessary technical complexity. At the same time, power users expected precision. They wanted a clear explanation of what privacy meant in practice, what data was shared, and what was not.

Process

Shipped a working mockup on day one to unblock engineering Ran a parallel research track, auditing privacy UX patterns and trust signals across competitors Turned research into PRDs defining scope, user flows, and success criteria Built visual moodboards to align the team on design language Iterated through high-fidelity designs across iOS and web Owned handoff to engineering, and where it made sense, built and shipped features directly

We killed the name

How do you make users believe their data is actually private? We tried leading with privacy as the headline, but it felt like VPN marketing. Instead, we made privacy the foundation, not the pitch. Show, don't tell.

Explorations

Privacy products default to locks, shields, and dark interfaces. We explored everything from 3D padlocks to botanical illustrations to cosmic landscapes. Most felt either generic or desperate. The shipped direction needed to feel calm and confident, not loud about security.

Onboarding as a growth engine

Onboarding wasn't just about getting users in. We turned it into a conversion machine. Strategic paywall placement, upsell screens, and a free lifetime promo created urgency and drove sharing. 18K downloads, no ad spend.

Framing models by outcome

Users don't know the difference between DeepSeek and Llama. We tried an expert mode with full model specs. Overwhelming. We tried auto-select. Felt like a black box. We shipped task-based labels that tell users what each model is good at, not what it's called.

Embeded trust

Instead of banners and badges, we built trust into the interaction layer. Secure chat indicators, encryption status in the message bar, thinking animations that reinforce on-device processing. Users feel safe without ever being told they're safe.

Discovery

I owned growth beyond the product itself. Optimized App Store screenshots and metadata for conversion, and secured editorial placements that drove organic downloads. 17K downloads, no ad spend.