The three-endpoint product extracts design systems, searches reference brands and scores agent output against a target identity.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [Taste Labs on X](https://x.com/tastelabs/status/2099913773951893761)
Why it matters #
AI agents can generate polished assets quickly, but brand consistency still requires structured context and evaluation. Taste Labs is turning that missing control layer into an API business.
Taste Labs founder Thais Castello Branco (@thaiscbranco_) launched a Brand API on September 15th that gives generative AI agents a structured way to extract, find and follow a visual identity. The San Francisco-based AI infrastructure developer introduced the product in a three-post thread on X, framing it as Taste Labs' first publicly available tool built specifically for agents.
https://x.com/tastelabs/status/2099913773951893761 The Brand API has three endpoints. The extractor turns a website into a design system covering elements such as logos, colors, typography, layout, components, motion, textures and shadows. The search tool accepts natural-language descriptions and retrieves reference brands from Taste Labs' curated index. The verifier compares generated work with a reference brand and returns a score, written criticism and structured instructions an agent can use in another generation loop.
Taste Labs is releasing the extraction endpoint in beta, while search and verification remain in alpha. Extraction and verification operate as asynchronous jobs, while search returns results directly. The output can include a brand profile in JSON alongside captured HTML, CSS and screenshots.
The product is also available through an MCP server, allowing developers to connect the same tools to Claude Code, Cursor, Codex and other compatible clients. Taste Labs says agents can request individual sections of an extracted design system instead of the entire result into context.
From model training to an agent product
Castello Branco started Taste Labs around the premise that generative models have made production cheap while leaving judgment scarce. She was an early employee at AI search developer Exa, where she worked on the Websets product, marketing and data partnerships before founding Taste Labs.
Her argument is that models trained to predict likely outputs tend to converge on familiar layouts, wording and visual motifs when the task lacks one objectively correct answer. In a June manifesto, Castello Branco wrote that agents need better context, inspiration and evaluation systems if they are going to create on behalf of users.
Taste Labs initially sold preference data, reasoning data, rubrics and evaluation environments to AI labs. The Brand API packages that research direction into a developer product for application builders. It also gives Taste Labs a usage-based software business alongside its work supplying training and evaluation data.
Taste Labs emerged from stealth on June 16th with an $18.5 million seed round co-led by CRV and Amplify Partners. The investors backed Castello Branco's effort to turn subjective judgment into signals that models and software can act on, starting with design.
That funding makes the Brand API an early test of whether Taste Labs can sell aesthetic judgment as infrastructure rather than keeping it inside bespoke data and evaluation projects. The target customer is an application developer whose agent can already generate a landing page, advertisement or sales deck, but cannot reliably make those assets look as though they came from the same organization.
One URL becomes the brand brief
Taste Labs' first published customer example is the General Intelligence Company of New York, which incorporated the API into Cofounder, an AI orchestration product that generates assets across marketing, sales, product and other functions.
General Intelligence co-founder Andrew Pignanelli said Cofounder had been building its own system to import a customer's existing visual identity. In Taste Labs' case study, Pignanelli described brand extraction as a harder infrastructure problem than expected. Cofounder now asks users for a website URL and uses Taste Labs to retrieve the associated design assets and tokens.
For customers without an established identity, Cofounder uses the search endpoint to retrieve a system matching descriptions such as a vintage or formal brand. Pignanelli said the integration improved Cofounder's output quality and consistency, though the case study is Taste Labs' own account of the deployment. The API is available with a free trial. Paid access starts at $99 a month for 600 credits, including all endpoints and MCP access. A $749 monthly plan includes 6,000 credits and onboarding support. Extraction and verification consume two credits per request, while search costs one credit for a light query and two for a deeper search.
Taste Labs is betting that model providers will continue improving raw generation faster than application developers can build the context and evaluation systems around it. The Brand API moves Castello Branco's pitch from teaching models what good design looks like to giving agents a brand book they can actually read - and a grader that can send them back to work.