The answer depends on whose brand it is and how often you work with it. If you do not work with the company, an API that reads its public website gives you enough. It answers fast and needs only a domain. This site calls that observed context. For repeated work under your own brand or the brands of your customers, use context that the brand maintains. Rainbrand supplies the guidance and assets of the brand for each task, and you keep your model and workflow.
| Job | Pick | Why |
|---|---|---|
| Research a prospect or personalize outreach | Brandfetch Brand Context API | The only observed response with voice, audiences, products with prices, and competitors. It takes a work email |
| Make a demo or mock-ups in the brand of a prospect before they sign | Rainbrand: add their brand from their website | Guidance, rules, and files that a model can design from, not only a description. If they become a customer, their brand is already there |
| Show logos and colors in a list, a CRM (customer relationship management) tool, or a signup screen | Brandfetch Brand API, Context.dev Brand, or Logo.dev | Logo files for light and dark backgrounds from a domain. The Logo.dev links expire after about a day, so store the files |
| Build a page that looks like the site of a company | Context.dev Styleguide | Color roles, every heading size, spacing, and button and card styles from the CSS of the site |
| You already scrape pages with Firecrawl | Firecrawl branding format | Add branding to the formats of the scrape that you already make. It can sit next to markdown in the same request |
| Marketing work that your agents make under your own brand | Rainbrand: a Brief for each task | Guidance, claim limits, and assets chosen for the task from sources added for the brand, with the Brand Skill version that they came from |
| An AI product that makes work for the brands of its customers | An observed lookup at signup, then Rainbrand for each returning customer | The lookup fills in the first screen. Then you add the brand of each customer from their website and files, and every job requests a Brief |
| An agency that runs many client brands through Claude, ChatGPT, or Cursor | Rainbrand over MCP | Add each client as a brand. Agents request a Brief for each task, and nobody gives them the full guide again for every task |
For logos only, at volume, compare plans on The Brand API Register, which lists brand APIs product by product.
Why not an observed API for your own brand?
A reading of a public website is not the guidance of the brand. With Linear, on the same day, the responses showed three differences:
- The four observed vendors returned four different main colors. Only the Rainbrand Brand Skill showed which colors were official (Colors).
- The only things to avoid in any observed response were two tones. The Rainbrand Brand Skill returned naming rules and logo rules. Its Brief listed claims that the copy for the task must not make (Rules and claims).
- Observed responses carried a read date, a cache age, or nothing. The Brief carried the Brand Skill version that it came from. With it, you can tell which context a piece of work used (Dates and versions).
These results are not a fault in the observed APIs, because they describe what a site shows. Linear publishes its rules on its brand page at linear.app/brand, with the two official colors and the rules for the logo. None of the observed responses returned those rules. The Rainbrand brand for Linear, added from linear.app, returned them.
Why not Rainbrand for everything?
Rainbrand needs setup. Someone must add a brand before its first Brief, and Rainbrand prepares each Brief for its task. Our Brief took 42 seconds, but an observed lookup needs only a domain. Rainbrand returns nothing for a company that nobody added, so it cannot research a prospect. Rainbrand also supplies brand context, not the CSS of a site. For a page that looks like a site, the Context.dev Styleguide returns the numbers that a coding agent needs.
Using both
Many products use both kinds. They use an observed lookup when they have only a domain, and a maintained brand when a customer comes back for more work. Keep the two kinds apart in your data. If they disagree, the sources of the brand win. The customer brands recipe shows the change from one kind to the other in code.
Last updated 12 November 2026