MeetingEdge · In development
Capture first. Interpret second.
A local-first Windows workspace for recording, transcription, notes, and evidence-linked AI output.
The problem
A recording has lasting value even when a model is unavailable or its interpretation is wrong. The design separates reliable capture from transcription and interpretation, so those steps can be revisited independently.
Decisions that matter
Keep capture independent
Incremental audio storage and recovery information support a recording path that is separate from the speech and language-model providers.
Choose speech and interpretation separately
Local Whisper or an optional speech endpoint can produce the transcript. Ollama or an optional model provider can interpret it later.
Keep the evidence attached
Structured notes refer back to transcript segments. Unsupported AI output is a review problem, not a reason to invent a citation.
Follow the workflow
- Choose the capture type and devices.
- Record and retain the original audio.
- Transcribe using a selected speech provider.
- Review notes alongside supporting transcript segments.
What is implemented
The source includes a Windows capture layer, local API, React interface, SQLite workspace, and independent provider adapters. Hardware and long-session qualification remain open.
Scope and current limits
Local processing is the default direction; optional cloud providers transmit selected content to those providers. Model integration is not model training.