Privacy & data flowchatRAR is designed so that your chat history never leaves your browser as a whole. Here is exactly what happens, step by step.Step 1 — File parsing (100% local)When you drop a .zip or conversations.json:The file is read by JavaScript running in your browser tab.The .zip is unpacked in-memory using JSZip (no network calls).Conversations are flattened along the canonical branch.Nothing is uploaded. You can verify in DevTools → Network: no request fires until you click Generate memory.Step 2 — Sample selection (still local)Sending hundreds of MB of chats to an AI would be slow, expensive, and unnecessary. So we sample:30 most recent conversations + 20 longest ones.From each, only your own messages (assistant replies are dropped — they don't describe you).Each message truncated to 600 characters; max 8 per chat.Whole sample hard-capped at 60,000 characters.The "Show what gets sent" button on the home page reveals the exact text that will be transmitted, before you click Generate.Step 3 — Generation request (the only network call)When you click Generate, exactly one POST request is sent:Endpoint: /api/generate-memory on this same domain.Body: { style, sample } — the sample is the truncated text above; style is one of about_me, working_on, everything.The server forwards your sample to an AI provider (Google Gemini via the Lovable AI Gateway) to produce the paragraph, then streams the result back.The server does not store, persist, log, or cache the sample or the generated paragraph.What IS logged (honest disclosure)We don't run analytics, set cookies, or fingerprint visitors. However, the platform that hosts this app (a serverless edge runtime) keeps standard access logs that include:Your IP addressTimestampHTTP status code and response sizeThese logs do not contain your sample text or output. They exist for abuse prevention and infrastructure debugging, and rotate automatically on the host's schedule.The AI provider (Google) processes the sample to generate the response. Per Google's API terms, content sent through paid API routes is not used to train models, but you should treat anything you send to any third-party AI as something a third party briefly saw.What is NEVER stored or sentThe full .zip / conversations.jsonAssistant replies from your chatsConversation titles you didn't include in the sampleAny account info — there are no accountsAny cookies or tracking pixelsQuestions or want to verify the source? About this tool.