The taste layer for AI agents.
One judgment layer between the agents people trust and the products that fit them.
Agent to agent
Built for how agents work, not for how people browse.
- 01
Onboarding by the person’s own message.
The person sends their agent one message; the agent requests access in its own name and receives its credentials at its own address.
- 02
Information, never instruction.
Every response is data the agent weighs with its own judgment. Nothing steers the agent, favors a retailer or speaks for its user.
- 03
A stable contract.
A versioned API with machine-readable documentation and a public changelog. Breaking changes are announced 30 days ahead.
Search finds. Taste chooses.
Agents know the person. Search reduces that knowledge to keywords. Taste for Agents keeps the whole story and returns what fits.
How it works
From a person’s story to pieces that fit, in one call.
Context in.
The agent sends the full story: occasion, taste, wardrobe, sizes, budget. One line or several pages.
A picture of the need.
The request becomes structured signals: dress code, constraints, owned pieces, taste.
A catalog built for judgment.
Hundreds of thousands of products from premium retailers, matched across retailers, checked for price and size, described in a stylist’s terms.
Taste.
A model of the person’s taste, informed by similar profiles, weighs each piece against the person and against the others.
Styling.
An anchor piece, then what completes it for the occasion and the wardrobe, with alternates.
Facts out.
Products, live offers and why each fits. One clarifying question when the brief is too thin.
# Client dinner, Thursday
Client dinner: polished, not formal. Build around the ivory blazer they own.
One anchor, two complements, one accent.
The ivory blazer already in the wardrobe is the piece to build on: over a champagne silk blouse it reads polished, not office. Black wide-leg trousers keep the line long and work with the black flats, and gold hoops are the one bright note for the evening.
- AnchorIvory blazer
- ComplementSilk blouse, champagne$185
- ComplementWide-leg trousers, black$240
- AccentGold hoops$165
Two per piece, in stock and in budget.
Two calls
Search when the agent knows what it wants. Style when it knows who it’s for. Over HTTP or MCP, with a machine-readable description of every call.
Search.
Products for a request, with constraints.
- In
- A free-text request, plus optional constraints: gender, category, colors, sizes, price, retailers.
- Out
- Matching products with live offers, sizes in stock and the facts an agent needs to choose.
Style.
A complete outfit from a styling brief.
- In
- A brief from one line to several pages: the occasion, taste, fit, sizes, wardrobe and budget.
- Out
- One outfit with alternates for each piece and the reasoning behind it, or one clarifying question.
By the numbers
One catalog, built for agents.
500K+
curated products, from 4,700+ brands across 680+ retailers.
1.4M
live offers, checked for price and size through the day.
100%
of outfit pieces in stock and priced when the answer arrives.²
Head-to-head: an agent with Taste, and one browsing the web¹
| Measure | Taste for Agents | Agent browsing the web |
|---|---|---|
| Speed, median wait on tools per request | 9.3 s, 5.9 times faster | 55.2 s |
| Calls that succeeded | 100%, 225 of 225 | 81%, 230 of 285 |
| Requested sizes confirmed | 100%, 17 of 17 | 61%, 11 of 18 |
¹ Speed benchmark, September 30, 2026: the same 20 shopping requests, each given to two agents on the same model with the same instructions, one calling Taste for Agents and one searching and browsing the web. Median time waiting on tools per request: 9.3 s against 55.2 s.
² Stock, September 30, 2026: 1,761 pieces in 451 outfits, from 84 styling briefs across six production runs, each in stock in our catalog when the answer was sent.
Partners
TasteGraph works with agent platforms, retailers and publishers.