The Brand API extracts, searches, and verifies visual identities.

Taste Labs' Brand API extracts a website's visual identity, searches for stylistic references, and verifies whether an agent's creations respect the brand.

Give an agent a web address, then ask it to produce a presentation, an advertisement, or a page that matches the company. The process sounds simple, but it requires understanding the colors, fonts, spacing, hierarchy, components, motion, and visual logic connecting them all.

Taste Labs wants to bring this work together through a single interface. Launched on September 15, 2026, its Brand API gives AI applications three capabilities: extracting the visual system of an existing website, finding brands that match a creative direction, and verifying whether a new page remains consistent with a reference.

The company is not offering a new standalone creation tool. Its service instead forms an intermediate layer between an agent and whatever it has been asked to produce. It supplies context before generation and can then inspect the result afterward.

The first feature, Brand Extraction, receives a website address. Taste loads the page in a browser, analyzes the rendered content, and turns it into a structured system. The JSON response can include colors, typography, spacing, layouts, components, shadows, textures, animations, and a general description of the brand.

Taste also provides the material collected during the process, including the HTML, style sheets, and page screenshots. An agent can then consult all or part of this information when producing a new asset.

The extraction therefore goes beyond retrieving a logo and a few colors. It attempts to reconstruct the relationships among the elements: which typeface is used for headings, how sections are organized, which corners or spacing values recur, and what kind of motion accompanies interactions.

The result does not necessarily constitute the company’s official brand guidelines. A website expresses only part of an identity and may be shaped by technical constraints or choices specific to a campaign. It does not always include the rules intended for print, video, social media, or internal use.

The state of the website at the time of analysis also matters. A temporary page, localized experience, promotional banner, or interface test could be interpreted as a permanent part of the brand. The API automates the observation of an existing implementation; it cannot know intentions that are not visible on the page.

Extraction runs as an asynchronous task. A request creates a job that the application must poll until completion. Taste says an analysis generally takes a few minutes and longer when deep analysis is enabled. Some sections become available before the full process has finished.

Results are cached by URL. A page that has already been processed may therefore be returned immediately. A `force: true` option requests a fresh analysis when a website has changed. Without that refresh, an application may work from an older version.

If an extraction fails, the job status changes to `failed`, an explanation is provided, and the credits are automatically refunded. Taste has not yet published measurements covering extraction success rates, accuracy, or differences in performance depending on website complexity.

The second feature, Style Search, is designed for users who do not yet have a visual identity. An agent can submit a description such as “a vintage skincare brand” or “dark brutalist developer tools.” Taste then looks for references in its own index of visual systems.

A query can return between one and sixteen brands. Each result takes the form of a card containing a palette, typography information, descriptive tags, and a full-page screenshot. The agent can draw from a single reference or combine several.

The search can also use an existing extraction as its starting point to find visually similar brands. Filters covering industry, dominant hue, page type, or layout can be applied as mandatory constraints.

Two levels are available. Light search prioritizes speed and returns results within seconds. Deep search takes longer and promises better ranking. Taste does not disclose the size of its index, its selection criteria, its geographic coverage, or how frequently it is updated.

Search does not directly create an original identity. It supplies structured examples that the agent can use as context. The final result still depends on the generation model, the instructions it receives, and how the application uses those references.

This distinction matters when Taste presents the service as a way to combat the generic appearance often associated with AI-generated work. Adding more varied references may indeed broaden an agent’s visual vocabulary. It does not guarantee that the final proposal will be relevant, readable, accessible, or sufficiently distinct from its sources.

The option to request a brand that “feels like Apple” also raises questions about creative distance. Drawing inspiration from an atmosphere does not grant the right to reproduce distinctive elements, a protected composition, or a competitor’s identity. Taste provides a search tool, but teams remain responsible for the references they select and how they use them.

The third feature, Verify Adherence, comes into play after creation. It receives two URLs: a reference website that serves as the standard and the page to be evaluated. Taste extracts both, compares them, and returns a verdict, written recommendations, and a list of structured fixes.

The most serious issues appear first. The result can therefore serve as a task list for an agent: change a color, reconsider a typeface, adjust the hierarchy, or bring a component closer to the reference. A new version can then be generated and submitted for another review.

Taste presents this feature as a step that can support an autonomous loop. The agent generates a page, receives a critique, applies the corrections, and repeats the process until it reaches a level considered satisfactory.

This automation does not replace human review. Brand consistency is not limited to similarities between two pages. It also depends on the message, audience, medium,