Terms

• Svanetra learns how you like your AI tools to work, as short rules, and gives the rules that fit each task to the tools you connect.
• It learns from the messages you send in those tools, from files and exports you import, and from your past chats: on a computer once a tool connects, in your browser only once you say yes.
• It keeps your preferences, never your chats, code or files. What it learns from imports and past chats waits for your yes.
• It works with other companies’ AI tools and sites, which have terms of their own that you are responsible for keeping to.
• It is free and provided as is. You can see, change and delete everything, and delete your account, at any time.

Svanetra is a free service made of four parts: this website, a helper you can run on your computer for AI coding tools, a browser extension for AI chat sites, and a server those tools can call (at api.svanetra.com/mcp). By using any of them you agree to these terms and to the privacy policy, which says in detail what is kept, for how long, and who helps us run the service.

You need to be at least 13, or the minimum age for online services where you live. You sign in with your email address, Google or GitHub, through WorkOS. One person per account. Keep your sign-in safe: you are responsible for what happens in your account, and for what the computers and browsers you connect to it do.

• The messages you send in the tools and sites you connect: your message, up to five of your earlier messages in that chat, and the end of the answer before it. Each is read once and deleted.
• Your past chats, once: in a coding tool, when you connect it on a computer; on ChatGPT, Claude and Grok, only after you say yes. Only your own messages from the past year are read, up to a limit, with code shortened and keys and passwords taken out.
• The rules and memory files your coding tools keep, and, on ChatGPT, your custom instructions and the memories ChatGPT saved, once you say yes.
• What you paste or import on the Tools page, such as a rules file or a chat app’s data export. An export is read in your browser: only your own messages from the past year leave it.

From these Svanetra finds your lasting preferences: how you like work done, written, built or shown. What it learns from imports and past chats waits in Review for your yes, and you can see, edit, undo or delete every rule. The privacy policy lists exactly what is kept and for how long.

• When you connect a computer, you run Svanetra’s helper on it. It runs in the background, checks in with Svanetra now and then, and updates itself from svanetra.com.
• For each AI coding tool you switch on (Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot or Windsurf), it installs Svanetra’s hooks in the tool’s settings and keeps your preferences in a marked block in the tool’s instruction file.
• Its hooks report each turn of a chat in that tool, as described above, and can give the tool the rules that fit your request. With Ask me first on, it may show a small window to choose which of your rules a chat uses.
• Switching a tool off in Svanetra takes its hooks and Svanetra’s block out again, leaving your own text as it was. Removing the computer in Svanetra takes Svanetra out of every tool on it.
• To keep a computer’s past chats out, run node ~/.svanetra/svanetra-hook.mjs chats off on it.

• It works on ChatGPT, Claude, Gemini and Grok, only on the sites you switch on in Svanetra, and never in their temporary or incognito chats.
• When you send a message there, it holds it for a moment and asks Svanetra which of your rules fit. With Ask me first on (the default), a card shows them, and the ones you tick go at the end of your message under “My preferences (from Svanetra):”. With it off, the rules that fit are added without asking. Then the extension sends the message for you, by pressing the site’s send button. If Svanetra doesn’t answer quickly, the message goes as you wrote it.
• It reports each message you send there, as described above, so Svanetra can learn from it.
• On ChatGPT, Claude and Grok, it asks before reading your past chats, and reads nothing until you say yes. Each site has a switch in the extension’s menu, to turn this on later or off again. To read them it uses the same calls the site’s own page makes, in your own signed-in page.
• It writes on a site only to add your rules to a message you are sending. It doesn’t change a site’s settings or read other sites.
• Removing the browser under Tools, or the extension from your browser, stops it. It keeps nothing of your chats in the browser, except what could not reach Svanetra yet, for a week at most.

Svanetra works with AI tools and sites that other companies run, such as ChatGPT (OpenAI), Claude and Claude Code (Anthropic), Gemini and Gemini CLI (Google), Grok (xAI), Codex, Cursor, GitHub Copilot and Windsurf. Svanetra isn’t made, endorsed or supported by any of them, and their names belong to them.

Your use of each stays under its own terms, and you are responsible for keeping to them. Some of those terms limit automated or programmatic access to the service, or taking data or output out of it. Svanetra’s helper and extension act only in your own session, when you ask them to, and only with your own chats and settings. But whether that is allowed for a service is for that company’s terms to say, not ours: read them, and switch the tool or site off in Svanetra if you aren’t sure.

These tools and sites change without notice. When one does, parts of Svanetra may stop working with it until we catch up.

Svanetra uses AI models, from OpenAI and TypeSafe, to read what your tools report and what you import, to find your preferences, and to check which rules fit a task. They can be wrong: a rule may miss what you meant, or a tool may be given one that doesn’t fit. So check what Svanetra learns, as Review asks you to, and change or delete what isn’t right. Your rules are not used to train AI models.

Your rules are yours. You let us store, process and show them so the service works, including to the people you share them with, and to give them to the tools you connect. Don’t put other people’s personal information, secrets, or anything unlawful or hateful in your rules or in a graph you share. Svanetra refuses a rule that contains code, a file path, a link, an email address, a username, a phone number or anything that looks like a key or password.

You choose who sees your graph. Anyone can see, star, follow and copy rules from a public graph, and rules you copy become yours. We may remove a shared graph that breaks these terms.

• Don’t overload or attack the service, scrape it, or get around its limits.
• Don’t use it to harm or deceive other people.
• Don’t use it to read chats or accounts that aren’t yours.
• We set rate limits, and may suspend accounts that misuse the service.

Svanetra gives your tools preferences about style and approach. It never authorizes them to act. Rules added to a message are sent to that site’s AI as part of your message. What your tools do stays your responsibility and theirs.

The service is free and provided as is, without warranties. It may change, pause or stop; before shutting it down we will try to give notice. To the extent the law allows, we aren’t liable for indirect or consequential losses, for what your AI tools do with your preferences, or for how other companies treat your use of their tools and sites.

You can switch any tool or site off, and remove a computer or browser, under Tools at any time. You can delete your account in Settings: the account and everything in it go right away. We may suspend or close accounts that break these terms.

We may update these terms; the date above shows when. For big changes we will tell you in the app or by email. Questions: hello@svanetra.com.