Lovable builds your frontend. Kumiko builds your backend. You describe it once and use it as a screen, an API and an AI tool.
A typical starting point
A B2B startup builds software for property managers. Six months in, there are four paying customers. The stack is Next.js, Supabase and Clerk, plus a marketing site generated with Lovable. Two months went into hand-rolling three things: keeping each customer’s data separate (still buggy), audit logs (homemade Postgres triggers) and user roles (Clerk plus custom claims).
What this company needs:
- Stop rebuilding customer separation, change history and roles for every new feature. Software that keeps each customer’s data apart is called multi-tenant.
- A customer-facing history dashboard, because an enterprise lead asked for one.
- A clean exit path for the day the company is sold, with no vendor lock-in.
What’s missing in the market
Lovable, v0 and Bolt only cover the frontend. The backend is still manual work, and the promise of shipping 80% quickly breaks at the API boundary. The generated TypeScript has no architecture behind it, so refactoring after five features hurts.
Supabase, Firebase and Convex are backend services, not generated backend code. You still write the queries, they don’t generate features for you, and keeping customers’ data apart is still your job.
Marten and Axon (frameworks that store every change as its own record) come from other languages, .NET and Java. The learning curve is steep and there is no AI integration.
Claude Projects and Cursor generate generic code without knowing your domain. There are no framework building blocks, so everything is built from scratch.
What Kumiko does differently
| Pain | Kumiko answer |
|---|---|
| “Backend takes longer than frontend” | You describe entities and handlers once in TypeScript. Kumiko builds the screens, the API and the tools for your AI assistant from that one definition |
| “Rebuilding customer separation on every feature” | Every entity is scoped to its customer by default. Access across customers is opt-in, not opt-out |
| “Audit logs are always an afterthought” | Every change is an event. History, time travel and rebuild come with the framework, and “who changed what, when” is one query |
| “I want to switch LLM providers” | One configuration setting switches between Anthropic, OpenAI and Llama. No code change |
| “Vendor lock-in scares my customers” | The framework is public under the MIT license and runs on Hetzner, Kubernetes or your own hardware. Self-hosting the platform comes with Enterprise |
| “Designer and AI are two different worlds” | Kumiko Studio is the one place to build: you describe the app, the AI proposes, you approve, and you change it in conversation |
Why it fits
- Each customer’s data is separate from day one. Every entity is scoped to its customer by default, which removes the most common bug class in B2B software at the framework level. You can see it live: publicstatus.eu serves multiple customers on a single instance, each with its own subdomain and branding.
- Hosting in Germany (Hetzner), and you bring your own LLM API key. Switching between Anthropic, OpenAI and a local Llama is one configuration setting.
Our own products run on Kumiko: publicstatus.eu, offlot.app and cashcolt.kumiko.rocks. Bugs hit us first.
Live today
- The framework with ready-made building blocks: login, change history, customer separation, realtime updates and more
- Showcase: publicstatus.eu in production
- Kumiko Studio, with the status of each step on its page
- The
kumiko createcommand-line tool, build and deploy pipeline
You can see it at publicstatus.eu and in a shared walk-through of your concrete use case.
Pricing
- Self-host: the framework is free to use under the MIT License
- Bring your own LLM key, no markup on AI usage
Details: Pricing.
Next step
The docs and the Show Pony tutorial take you from an empty app to a running application. Want the app built for you? Kumiko Studio.
Continue in the docs