LocalMeet
2026
LocalMeet — Offline AI Meeting Assistant
A privacy-first desktop app that records meetings, transcribes them, and turns them into summaries, decisions, and action items — entirely on-device. No cloud, no account, no telemetry.
Role
Product Designer & Developer
Team
Solo
Duration
4 weeks to MVP
The challenge
Meetings generate decisions, action items, and context that teams lose the moment the call ends. AI assistants promise to capture it all — but they send confidential audio and transcripts to the cloud, which is a non-starter for teams and professionals who handle sensitive conversations.
LocalMeet was built around a single question: can a meeting assistant be genuinely useful while never leaving the user's computer? The answer had to be a polished desktop app, not an engineering demo — no account, no upload, no telemetry, and the AI itself running locally.
The value is not the biggest model. The value is that useful meeting intelligence runs privately on your computer.
Product screens
Meeting view — summary, decisions, and action items
Transcript — searchable, timestamped, linked to evidence
Recording — minimal controls, privacy indicator visible
Settings — model downloader, storage, and privacy toggles
Design process
Define the privacy promise
Turned 'no cloud' from a technical detail into the central product narrative and a visible UI feature.
Simplify the recording flow
Reduced the core journey to five steps: create, choose mic, record, stop, review. No onboarding walls.
Structure AI output
Designed a scannable dashboard that separates summary, decisions, action items, questions, topics, and participants.
Make trust auditable
Every action item links to transcript evidence, so users can verify what the AI inferred.
The solution
I designed and built the entire product. The AI is compiled directly into the binary — speech-to-text via whisper.cpp and analysis via llama.cpp — so there are no cloud dependencies and models can even be swapped or downloaded from within the app.
The core journey is intentionally minimal: record, stop, and the app transcribes the meeting and structures it into a summary, decisions, action items with owners and deadlines, unresolved questions, key topics, and participants — with a searchable, timestamped transcript.
Trust is a visible feature rather than a hidden one. A persistent on-device indicator, privacy toggles, storage visibility, and action items that link back to transcript evidence all make the privacy promise tangible and auditable.
The result is a working, offline-first meeting assistant that proves useful AI intelligence doesn't require sacrificing privacy.
Key design decisions
Calm, focused UI
A minimal palette, strong typography, and restrained motion keep the interface professional and distraction-free.
Privacy as a feature
The 'local processing enabled' indicator and clear privacy toggles make security visible, not hidden.
Auditable AI
Action items link back to transcript evidence so users always know why the AI made a decision.
Five-step workflow
Create, choose microphone, record, stop, review. No onboarding walls or complicated configuration.
Structured output
Summary, decisions, action items, questions, topics, and participants are separated for quick scanning.
On-device AI
Whisper and Llama inference is compiled directly into the Rust binary, with models managed through an in-app downloader — no external executables or cloud calls.
Tools & methods
Design
Figma
AI collaboration
Claude
Backend
Rust
Frontend
HTML, CSS, JavaScript
This is the first meeting tool that doesn't make me read a privacy policy.
— Beta Tester
Coming soon
Download LocalMeet
A privacy-first meeting assistant that runs entirely on your computer. No cloud, no account, no telemetry.
LocalMeet is a private, offline-first product currently in beta