Configuration

Set default values for model size, output directories, and more. Disable tools you don’t need.

Augent loads user defaults from ~/.augent/config.yaml on startup. Every setting is optional — if the file doesn’t exist, Augent uses sensible defaults.

Per-call arguments always override config values.

---

Example config

# ~/.augent/config.yaml

# Whisper model size: tiny, base, small, medium, large
model_size: tiny

# Default download directory
output_dir: ~/Downloads

# Default directory for notes, clips, and TTS output
notes_output_dir: ~/Desktop

# Seconds of padding around clips
clip_padding: 15

# Words of context in deep_search / search_memory results
context_words: 25

# Default TTS voice
tts_voice: af_heart

# TTS speed multiplier
tts_speed: 1.0

# Tools to hide from your MCP client (list of tool names)
disabled_tools: []

---

Disabling tools

If you don’t use certain tools, you can hide them from Claude (or any MCP client) so they don’t clutter the tool list:

disabled_tools:
  - text_to_speech
  - identify_speakers
  - separate_audio

Disabled tools are removed from the tools/list response and cannot be called. To re-enable, remove them from the list and restart Claude Code.

---

Config format

Augent tries ~/.augent/config.yaml first (requires PyYAML). If PyYAML isn’t installed, it falls back to ~/.augent/config.json:

{
  "model_size": "base",
  "disabled_tools": ["text_to_speech"]
}

---

Available tools

These are the tool names you can use in disabled_tools:

Tool	Description
download_audio	Download audio from URLs
transcribe_audio	Full transcription with timestamps
search_audio	Keyword search in audio
deep_search	Semantic search by meaning
take_notes	Download + transcribe + format notes
chapters	Auto-detect topic chapters
batch_search	Search multiple files in parallel
text_to_speech	Convert text to speech
search_proximity	Find keywords near each other
identify_speakers	Speaker diarization
list_files	List media files
list_memories	List stored transcriptions
memory_stats	Memory statistics
clear_memory	Clear transcription memory
search_memory	Search across all transcriptions
separate_audio	Vocal/stem separation
clip_export	Export video clips
highlights	Export best moments
tag	Tag transcriptions

Restart Claude Code after editing config. The MCP server reads the config once on startup.