👗 AI Clothing Consultant

Version 0.0.14

A CLI tool that suggests outfits from your wardrobe using either Google Sheets or local wardrobe files as a clothes database and a local LLM.

Features

• Reads your wardrobe from Google Sheets or local wardrobe files (tabs: Shoes, Tops, Pants)
• Fetches live weather forecast automatically based on your location
• Uses your personal style preferences as context
• Interactive multi-turn chat — ask follow-up questions naturally
• Runs fully locally via LM Studio or other local AI endpoint (no cloud AI costs)
• Saves the latest assistant suggestion to output/suggestions.md only for suggestion-style answers (not casual conversation)

Prerequisites

• Node.js v18+
• LM Studio with a model downloaded and the local server running
• If you use Google Sheets: a Google Cloud project with the Google Sheets API enabled and a Google Sheet with your wardrobe
• If you use local files: a wardrobe/ folder with CSV/XLS files or a wardrobe.xlsx workbook

Setup

1. Install dependencies

npm install

2. Choose Your Wardrobe Source

Option A: Google Sheets

Your sheet should have three tabs: Shoes, Tops, Pants, Accessories.
Each tab has a header row with columns: name | type | color | condition | fit | notes

Example:

name	type	color	condition	fit	notes
White Oxford Shirt	button-up	white	good	slim	slim fit
Navy Chinos	trousers	navy	okay	regular

Option B: Local Files

Set WARDROBE_SOURCE=local in .env to use local files instead of Google Sheets.

Supported layouts:

• A single wardrobe/wardrobe.xlsx or wardrobe/wardrobe.xls file with Shoes, Tops, Pants sheets
• Separate wardrobe/Shoes.csv or wardrobe/Shoes.xlsx, wardrobe/Tops.csv or wardrobe/Tops.xlsx, and wardrobe/Pants.csv or wardrobe/Pants.xlsx

Local files use the same columns as Google Sheets: name | type | color | condition | fit | notes

3. Option A: LM Studio

1. Download a model (e.g. gemma-4-e4b, llama-3.1-8b-instruct)
2. Go to the Developer tab → load the model → click Start Server
3. Note the model name shown in the UI

Any openai-compatible endpoint will do.

3. Option B: Google Service Account

1. Go to Google Cloud Console
2. Create a project → enable Google Sheets API
3. Create a Service Account → generate a JSON key
4. Save the key as credentials/service-account.json
5. Share your Google Sheet with the service account email (Viewer access)

Only local or only external endpoint can be used.

4. Configure environment

cp .env.example .env

Edit .env. LM Studio is the default — no model name needed:

# Local AI (default — LM Studio, Ollama, llama.cpp, etc.)
LOCAL_AI_BASE_URL=http://localhost:1234/v1
# LOCAL_AI_BASE_URLS=http://192.168.0.36:1234/v1,http://localhost:1234/v1
# LOCAL_AI_MODEL=google/gemma-4-e4b   # optional: pin a specific model
# AI_AVAILABILITY_TIMEOUT_MS=12000      # model availability checks (startup + ask)
# AI_RESPONSE_TIMEOUT_MS=45000          # full model response timeout
# AI_TIMEOUT_MS=45000                   # legacy fallback for both values

# External endpoint (optional fallback after local endpoints)
# EXTERNAL_AI_BASE_URL=https://models.inference.ai.azure.com
# EXTERNAL_AI_BASE_URLS=https://models1.example.com/v1,https://models2.example.com/v1
# EXTERNAL_AI_API_KEY=your_api_key_here
# EXTERNAL_AI_MODEL=gpt-4o

SHEET_ID=your_google_sheet_id_here
GOOGLE_CREDENTIALS_PATH=./credentials/service-account.json

When EXTERNAL_AI_BASE_URL is set, the app still tries local endpoints first and then falls back to the external ones if local requests fail. Works with any OpenAI-compatible API: OpenAI, GitHub Models, Ollama, etc.

5. Style preferences

Copy style-preferences.template.md to style-preferences.md and edit the copy to describe your personal style — the AI uses this as context for every suggestion. The local style-preferences.md file is gitignored so your personal notes stay private.

Usage

Make sure LM Studio's local server is running, then:

npm start

For a one-shot request (single command, no interactive prompt):

npm start -- "what should I wear in Gdansk tomorrow?"

On startup the app will:

• Load your wardrobe from the configured source
• Fetch today's and tomorrow's weather forecast
• Start an interactive chat session

After suggestion-style assistant responses, the app overwrites output/suggestions.md with the latest suggestion and logs where it was saved.

Example prompts:

• what should I wear today?
• suggest something for a casual dinner tonight
• what about tomorrow? it's a job interview
• something warmer, it's cold outside

Type exit or press Ctrl+C to quit.

Project Structure

clothes/
├── src/
│   ├── index.ts              # Entry point & chat loop
│   ├── sheets.ts             # Google Sheets wardrobe reader
│   ├── local-reader.ts       # Local wardrobe file reader
│   ├── ai.ts                 # LM Studio client
│   ├── prompts.ts            # System prompt builder
│   └── weather.ts            # Weather forecast fetcher
├── style-preferences.template.md # Starter template for your personal style notes
├── style-preferences.md      # Your local personal style notes (gitignored)
├── credentials/              # service-account.json (gitignored)
├── .env                      # Your config (gitignored)
├── .env.example              # Config template
├── package.json
└── tsconfig.json