Zero-log

A log is a record that a server writes about what it does: which address called, at what time, which page, which error. Logs exist for good reasons such as finding faults, stopping abuse and billing. The expression zero-log comes from the VPN world, where a provider promises not to record which sites you visit or which addresses you use. Applied to AI chat, the same idea asks whether the text of your conversations, and the habits it shows, are written down anywhere you do not control. The idea matters more for conversations than for browsing, because a chat can contain health questions, money worries, drafts of letters and half-formed thoughts that people would never post publicly.

That question has several layers, because an AI request crosses a chain: the app, its hosting, and the company that runs the model. Each link can write its own logs, and a promise from one says nothing about the next. A careful claim therefore says where it applies, separates the content of messages from metadata such as times and counters, and admits that the word zero is rarely literal: a payment needs a record, and a fair limit needs a counter. Auditors call this scoping: stating exactly which systems, which data and which period a promise covers.

To judge a zero-log claim, look for a precise list of what is kept and for how long, a way to check it yourself (the network tab of your browser shows what your page sends), outside audits where they exist, and honesty about what is left to others. Vague adjectives such as secure or anonymous are not a list. A short, specific, testable statement is worth more than a big promise. Be wary of claims that cannot be tested: a statement you can only take on trust is a marketing line, while one backed by a public table and a visible network trace is evidence.

NeverNo words its Zero-log badge with care. Your history stays in your browser, encrypted, and none of your message text is logged on our side: the text crosses our servers to reach the model and back, and the transparency page states that it is written neither in the database nor in the logs. What we do keep is operational data, listed there line by line: the account, the plan, the credits, usage counters, and for each call its time, the model and the credits charged.

The models are the other half. Private models are served by providers chosen because they do not keep requests, and the transparency page says plainly that this rests on their pledges, as NeverNo cannot inspect their machines. Anonymized models, such as Claude, GPT, Gemini or Grok, are read by their maker, who may keep the text for a while under its own policy. This keeps the promise narrow on purpose: it describes what our own servers write down, and it does not pretend to speak for companies whose machines we cannot see.