How Fluent was built

Why I built Fluent

Good advice about clothes depends on what you already own, and a chat doesn't know that. Fluent keeps your closet in the AI you already use, so the assistant can answer "should I buy this?" and "what should I wear?" from your own clothes.

The product thesis

Fluent works inside ChatGPT, Claude, and any MCP-compatible app. The assistant handles the conversation and the recommendation; Fluent keeps the closet the user asks it to save.

It is designed around several questions:

• Is this fact confirmed, inferred, or out of date?
• Where did it come from?
• Should this AI app be allowed to use it?
• Can the assistant act on it?
• Did the change actually save?
• Can the user export or delete it?

Current product

• Closet: clothes and shoes with photos, details, fit notes, and product references the user confirms.
• Should I buy this? A purchase question brings in the owned items that matter for the comparison.
• What should I wear? Outfit and packing answers draw on exact saved items and their photos.

Architecture

Fluent uses one MCP interface across ChatGPT, Claude, and other compatible apps.

The main design choices are:

• typed shared and closet context
• source and certainty information for important details
• tenant-aware hosted storage
• approval before specific changes are saved
• verification that saved changes worked
• an optional visual closet that shows the same saved items
• original photos kept alongside any cleaned-up cover image

Fluent always works in text. Compatible apps can also show the visual Style Closet.

Availability

Fluent is a hosted service. Fluent is free.

What this project demonstrates

Fluent is an exercise in product architecture as much as implementation: deciding what an AI assistant should know, what it may change, what it must prove, and how those rules stay consistent across AI apps.