macOS · Local AI · Private by default

Real-time transcription.
AI that understands context.

Capture live audio, get instant summaries, extract action items, and ask deep questions — all running locally on your Mac.

Download for Mac View on GitHub

Requires macOS 13+ · Powered by Ollama · Your data stays on device

What it does

Live Transcription
Continuous, low-latency transcription of any audio playing on your Mac — meetings, lectures, calls.
Session Overview
AI-written paragraph summaries and key topic extraction that update automatically as the session progresses.
Deep Dive Analysis
Structured thematic breakdown: named topics with evidence and depth — built up incrementally over the session.
Context Awareness
Provide a background briefing before a session. The AI applies that context to every summary and analysis it generates.
To-Do Extraction
Action items and follow-ups pulled directly from the conversation — tagged by owner, sorted by urgency.
Model Switcher
Choose from installed Ollama models or download new ones in-app. Switch models between sessions — memory reclaimed automatically.

How it works

01
Record
Press record. scribeLight captures system audio in real time — no extra setup, no external mics required.
→
02
Transcribe
Speech-to-text runs locally, streaming words to screen as they're spoken. No audio leaves your device.
→
03
Understand
A local LLM reads the transcript continuously — surfacing summaries, topics, and action items as they emerge.

Download & install

Download scribeLight for Mac v1.0 · Apple Silicon · macOS 13+ Intel Mac download
1
Install Ollama
scribeLight uses Ollama for on-device LLM summarization. Install it from ollama.com, then pull a model:
ollama pull qwen2.5:3b
2
Install Python dependencies
Transcription runs via Faster Whisper — a local Python process scribeLight manages automatically. Python 3.9+ required.
pip install faster-whisper fastapi uvicorn
3
Open the DMG and drag to Applications
Launch scribeLight. On first run, grant Microphone and Screen Recording permissions when prompted — both are needed to capture audio.
4
Start recording
Press Record. The Whisper model loads once on first use (~5 seconds), then transcription is instant. Everything stays on your device.

Plans

Free
$0 / month
Get started with no commitment.
  • 60 recording minutes / month
  • 10 web searches
  • Live transcription
  • Session summaries & deep dive
  • All local AI features
Download Free
Pro
Pro
$12 / month
For power users who live in long sessions.
  • Unlimited recording minutes
  • 1,500 web searches & AI answers / month
  • Everything in Free
  • Priority model inference
  • Early access to new features
Get Pro