Configuration

All Verba settings are configured in VS Code's settings.json.

Settings Reference

Setting	Type	Default	Description
`verba.audioDevice`	String	`""`	Audio input device name. Leave empty for system default.
`verba.templates`	Array	8 built-in templates	Prompt templates for post-processing. See Templates.
`verba.terminal.executeCommand`	Boolean	`false`	If `true`, sends Enter after inserting text into the terminal.
`verba.glossary`	Array	`[]`	Terms preserved during transcription and cleanup (limit: ~80 terms).
`verba.transcription.provider`	String	`"deepgram"`	Transcription provider: `deepgram` (API) or `local` (whisper.cpp).
`verba.transcription.localModel`	String	`"base"`	Whisper model for local transcription: `tiny`, `base`, `small`, `medium`, `large-v3-turbo`.
`verba.contextSearch.provider`	String	`"auto"`	Context search provider: `auto`, `grepai`, or `openai`.
`verba.contextSearch.maxResults`	Number	`5`	Number of context snippets per dictation (1-20).

Audio Device

By default, Verba uses the system default microphone. To select a specific device:

Run the command Verba: Select Audio Device — a Quick Pick dialog lists all available devices.
Or set verba.audioDevice manually in settings.json:

{
  "verba.audioDevice": "MacBook Pro Microphone"
}

Tip

On Windows, you can list available devices by running ffmpeg -list_devices true -f dshow -i dummy in a terminal.

Glossary / Dictionary

Define terms (product names, technical jargon, abbreviations) that must be preserved exactly during transcription and cleanup. Glossary terms are sent as hints to Deepgram and as protection instructions to Claude.

Global terms are configured in settings.json:

{
  "verba.glossary": ["Kubernetes", "Visual Studio Code", "PostgreSQL", "gRPC"]
}

Project-specific terms are defined in a .verba-glossary.json file at your workspace root:

["Verba", "CleanupService", "TranscriptionService", "ffmpeg"]

Both sources are merged automatically. For best results, keep the combined glossary under ~80 terms (~300 Deepgram keyword tokens). If this limit is exceeded, a warning is shown and excess terms may be ignored by Deepgram.

Tip

Place .verba-glossary.json under version control so that all team members share the same glossary. Changes to the file are picked up automatically.

Offline Transcription (whisper.cpp)

By default, Verba uses the Deepgram Nova-3 API for transcription. You can switch to local, offline transcription via whisper.cpp for full privacy and zero API costs.

Setup

Install whisper.cpp: brew install whisper-cpp
Download a model: Run Verba: Download Whisper Model command
Switch the provider:

{
  "verba.transcription.provider": "local",
  "verba.transcription.localModel": "base"
}

Available Models

Model	Size	Speed	Quality
`tiny`	~75 MB	Fastest	Lower accuracy
`base`	~148 MB	Fast	Good balance
`small`	~488 MB	Moderate	Better accuracy
`medium`	~1.5 GB	Slow	High accuracy
`large-v3-turbo`	~1.6 GB	Slowest	Best accuracy

Models are downloaded to VS Code's global storage and shared across all workspaces.

Tip

Start with the base model. If accuracy is insufficient, upgrade to small or medium. The large-v3-turbo model provides the best quality but requires significant disk space and processing time.

Note

Offline transcription currently supports macOS. Linux and Windows support is planned for a future release.

Context Search Provider

For context-aware templates, Verba needs a search provider:

Provider	Setup	Speed
`grepai`	Install grepai, run `grepai init`	Fast
`openai`	Run Verba: Index Project command	Moderate
`auto` (default)	Uses grepai if installed, otherwise OpenAI Embeddings	—

{
  "verba.contextSearch.provider": "auto",
  "verba.contextSearch.maxResults": 5
}