Configuration¶

MailTag uses two configuration sources:

config.toml¶

Main configuration file with all settings:

[general]
ollama_model = "ollama_chat/qwen3-vl:8b-instruct"
api_base = "http://localhost:11434"
use_imap_folders_for_classification = true

[classifier]
historical_confidence_threshold = 0.9
min_count = 5
ai_confidence_threshold = 0.85

[imap]
host = "imap.example.com"
user = "${IMAP_USER}"
password = "${IMAP_PASSWORD}"

[gmail]
credentials_file = "credentials.json"
token_file = "token.json"

[fast_parse]
batch_size = 500
folder_cache_ttl_hours = 24
unclassified_folder_name = "A Classer"
junk_folder_name = "Junk"

[mlx]
enabled = true
embedding_model = "nomic-ai/nomic-embed-text-v1.5"
llm_model = "mlx-community/gemma-4-e4b-it-OptiQ-4bit"
llm_confidence = 0.85
llm_max_tokens = 128
llm_temperature = 0.2

[logging]
level = "INFO"
file = "mailtag.log"

.env¶

Secrets and environment-specific values:

IMAP_USER=your-email@example.com
IMAP_PASSWORD=your-app-password

# Optional: cloud AI provider (overrides MLX for Signal 6)
# MODEL=gemini/gemini-2.5-flash
# GEMINI_API_KEY=your-key

Gmail OAuth Setup¶

Go to the Google Cloud Console
Create a project and enable the Gmail API
Create OAuth 2.0 Desktop credentials
Download as credentials.json in the project root
First run will prompt for browser authorization

Dynamic vs Static Classification¶

Controlled by general.use_imap_folders_for_classification:

Dynamic (default): Uses live IMAP folder structure as categories, refreshed at startup
Static: Uses fixed categories from data/classification_schema.yml