Concept study — not a client engagementAI-powered mobile assistant
Pocket Assistant
A voice-and-text assistant for iOS and Android that remembers context, manages tasks and works across the user's calendar and notes.
9:41
Assistant
Listening for context
Summarize my 3pm call and draft the follow-ups.
Summary
Agreed on a pilot in two regions. Pricing review next week.Email pilot scope to Dana
Book pricing review · Tue
Share deck v3
Send the first one.
Ask anything…
- Platforms
- iOSAndroid
- Technologies
- React NativeLLM APIOn-device speechVector memory
The challenge
Mobile assistants often feel stateless and slow. The goal of this study was an assistant that keeps useful context between sessions, responds quickly on a phone connection and asks permission before it acts.
Our approach
- 01Streaming responses so the first words appear quickly while the rest generates.
- 02A lightweight memory layer that stores user-approved facts as embeddings and retrieves only what's relevant.
- 03Explicit confirmation sheets before any action touches the calendar, reminders or messages.
Architecture
- Mobile client
- Edge API gateway
- Orchestrator
- LLM + tools
- Memory store
Have something similar in mind?
Let's scope the real version together.