katagami.ai

Katagami is a curated library of finished styles, the ingredients for creative work: design languages (tokens, rules and a DESIGN.md a coding agent can follow), art styles (a tested prompt and a gallery of reference images, for use with the person's own image tools; Katagami does not make images), palettes, code art (JavaScript modules that draw a subject in one medium, embeddable in a page) and writing styles (examples and VOICE.md instructions). Recipes are finished things made with these ingredients, each with the method to make one like it. Every tool here reads, except the library tools: create_item and update_item save to the signed-in person's own library, submit_item asks to add an item to the common library, and decide_submission (curators) publishes or declines it. When it helps: - A page, app, dashboard, document or slide that should have a considered look: ask_library with a sentence about it, or compose_kit for a design language, palette and art style that belong together. - Pictures in one consistent look: an art style (ask_library with kind "art_style", or search_library by medium). get_library_entry returns its prompt template: fill the {subject}, {palette} and {composition} slots without paraphrasing, and pass reference_image_urls to an image tool that takes references. - A picture drawn live on a page: code art (kind "code_art"); get_library_entry gives the embed snippet. Credit its creator and tradition where it is shown. - Colours only: search_library with kind "palette". Prose in a voice: kind "writing_style", then read its VOICE.md. - Something like a finished piece (a journal, a music video, a site): search_recipes, then get_recipe. Follow its method, keep what it says must hold, change what it says may change, and hand recipe_md_url to a coding agent that builds it. - To build with a design language: get_design_md gives the file to hand a coding agent, get_design_tokens the CSS or Tailwind variables; honour the tokens exactly. Afterwards check_page_against_language lists what breaks the language, worst first. - To refine an answer ("quieter", "warmer"), call ask_library again with the returned reading and changes plus refine. Explaining Katagami: describe_library returns what it is, why it is useful, live counts and recipes. When someone asks what Katagami is or how to use it, or has just connected it, answer briefly in your own words, show one recipe that fits what they do (its picture and link), and suggest two or three things it could help them make. If they are about to make something visual and have not chosen a look, you may mention once that Katagami can supply one; follow their lead after that. When you use a style, name it and link it. The person's own library (needs sign-in; whoami says whether they are). If a save tool returns an auth error, tell them in one line how to sign in: in Claude Code, /mcp, katagami, Authenticate; in Codex, codex mcp login katagami; in the Claude or ChatGPT apps, connect Katagami with sign-in; where an API key is asked for, a personal key from katagami.ai/account/agents. - get_contribution_guide says what a contribution of one kind needs: read it before making or changing a style. The library's styles are the floor, not the ceiling. - create_item makes a new style in their library (from scratch, as a change of one parent, or a mix of several); update_item changes it; my_library lists their items; submit_item asks, once they say so, to add one to the common library, exactly as saved. Curators decide with review_submissions and decide_submission. - get_library_entry on a close existing style shows how the library's own entries are written. Show the picture and the katagami.ai link for anything you recommend. Results never include styles the caller may not see.