Resonance

Resonance is a local-first AI recorder for Obsidian.

It is built around:

• Web Audio capture
• Whisper for transcription
• Ollama or an optional cloud provider for summaries
• Session storage with a built-in library

The product is desktop-only.

What Resonance Does

• Records from one main microphone plus optional extra audio inputs in the same Web Audio graph
• Writes live transcript updates while recording
• Builds a final summary note after the recording stops
• Stores each session with audio, transcript, summary, and diagnostics
• Lets you inspect, recover, clean up, and bulk-delete session artifacts from the Library

Screenshots

Recorder

Library

First Successful Session

1. Install the plugin in your vault.
2. Open the Resonance settings page.
3. In Capture, choose your microphone.
4. Optionally add loopback or monitor inputs under Additional sources if you want call or desktop audio in the same recording.
5. In Transcription, set the whisper.cpp repo, CLI, and model paths.
6. In Summary, keep Ollama or choose a cloud provider.
7. Run Quick test.
8. Open the recorder and make a short recording.
9. Stop the session and confirm that the transcript and summary appear in the Library.

Setup Overview

Capture

• Microphone device: the voice input you speak into
• Additional sources: optional extra audio inputs such as loopback or monitor devices
• Segment seconds: how often live transcript updates are committed
• Quick test: verifies microphone access and the local pipeline

Transcription

Resonance expects a local whisper.cpp build plus a readable ggml model.

Typical setup:

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
cmake -S . -B build
cmake --build build -j

Then download a model, for example:

cd /path/to/whisper.cpp/models
./download-ggml-model.sh small

Set:

• whisper.cpp repo
• whisper.cpp CLI
• Model path

Summary

Recommended local path:

• provider: Ollama
• endpoint: http://localhost:11434
• model: gemma3

Cloud providers are also supported, but they require API credentials on the machine where the plugin runs.

Library

The Library is the operational workspace for finished and failed sessions.

It supports:

• previewing transcript and diagnostics
• opening transcript and summary notes
• audio playback and export
• recovery actions when transcript or summary is missing
• bulk cleanup with storage estimates
• storage stats for visible sessions

Runtime Artifacts

Each supported session persists:

• session.json
• audio/recording.wav
• audio/segments/
• transcript/live-transcript.txt
• summary/summary.md
• diagnostics.log

The session Library reads these manifests rather than scanning arbitrary files.

Common Failures

Microphone access denied

Symptoms:

• Quick test fails before recording starts
• Capture diagnostics show microphone access denied

Fix:

• Re-enable microphone access for Obsidian in the operating system settings
• Return to Resonance and run Quick test again

whisper.cpp or model missing

Symptoms:

• Diagnostics block recording
• Quick test captures audio but transcription fails

Fix:

• Point Resonance to a working whisper.cpp CLI
• Set a readable ggml model file such as ggml-small.bin

Additional source unavailable

Symptoms:

• A saved loopback or monitor input no longer appears
• Diagnostics warn that extra sources will be skipped

Fix:

• Re-select the source in Capture
• Remove stale additional inputs you no longer use

Local Development

Requirements:

• Obsidian desktop
• Node.js
• whisper.cpp with a local ggml model
• Ollama if you want the default local summary path

Commands:

npm run typecheck
npm test
npm run build

Manual Install in Obsidian

1. Run npm run build.
2. Copy dist/main.js, dist/manifest.json, and dist/styles.css into your vault plugin folder.
3. Enable Resonance in Obsidian community plugins.
4. Open the Resonance settings page.
5. Work through Capture, Transcription, and Summary.
6. Use Diagnostics for health checks and Library for saved sessions.