Personal project
Svanetra
Correct one AI once, and every other AI you use already knows
7 AI tools that share your taste
Built with
TypeScriptNode.jsPostgreSQLNext.jsMCPFastifyChrome extension
- The problem
- Anyone who works with several AI assistants spends each new chat correcting the same things: lead with the answer, no emojis, always TypeScript. Each tool's memory lives inside that tool and is kept up by hand, so what you teach Claude Code never reaches ChatGPT, and the next session starts from zero.
- My role
- Sole builder: the product, the backend, the web app, the browser extension, and the helper that connects coding tools.
- Approach
- Svanetra doesn't wait for an AI to ask for your taste, because an agent can skip that step and your taste then silently goes missing. A small helper connects each coding tool through the hooks and files the tool already loads, and a browser extension does the same on the web apps. The cost is more to build and keep working: a helper on each computer, an extension, and a connection for every tool, rather than relying on one server the agents call.
- Outcome
- Live at svanetra.com and free, with the extension in the Chrome Web Store. Correct any connected AI once, and the next chat in every other one starts from it.
What I built
- One place for your taste. Svanetra learns how you like AI to work from the corrections you make and, with your yes, from your past chats on day one. You can see, change or undo any of it, and anything it isn’t sure of waits for your OK.
- Every AI you use. A browser extension for ChatGPT, Claude, Gemini and Grok, and a helper for Claude Code, Cursor and Codex, so the next chat in each of them starts from your taste.
- Your taste, never your chats. Past chats are read to learn from, never kept, and secrets are removed on your own device before anything leaves it.
- Sharing. Hand your taste to a friend, share it by link, or start from someone else’s.
Questions about this one? Email me, or ask my AI assistant.