krz/brand-bench

client-side brand identity generator. describe a project, get a brand package: positioning, tone, color palette, typography, logo concepts, taglines, usage examples. runs in the browser.

no backend. no accounts. state persists in localStorage.

• template generation, offline, tailored by category
• ai generation via a local ollama instance, cancellable mid-stream
• multiple packages in tabs, each persisted separately
• inline editing, click any field in the preview
• section locking across regenerations
• undo/redo per package (⌘Z / ⌘⇧Z)
• editable color swatches, native color picker
• font pairings per category, 8-level type scale
• export to markdown, json, or a self-contained html page

npm install
npm run dev

open http://localhost:5173.

• build:

npm run build      # dist/
npm run preview

1. install ollama, pull a model:

ollama pull llama3.2
ollama pull mistral
ollama pull gemma3

1. run it:

ollama serve

1. a non-default host or port needs OLLAMA_ORIGINS set for browser requests:

OLLAMA_ORIGINS="*" ollama serve

1. in the app: Settings, enable AI, set base url (http://localhost:11434), pick a model, Test connection.

bigger models give better-structured output. generation takes 15-60s depending on hardware.

Dockerfile and compose.yml deploy the app next to ollama.

docker compose up -d --build

app serves on port 8000. ollama's api is proxied through nginx at api, so the browser makes no cross-origin request. after first boot, pull a model:

docker compose exec ollama ollama pull llama3.2

set the ollama base url to http://your-server:8000. for an nvidia gpu, uncomment the deploy block in compose.yml (needs the nvidia container toolkit).

react 18, typescript, vite. no ui library, plain css.

0bsd. see LICENSE.