Short, plain-English definitions of the terms that come up when people talk about private AI chat: the browser tech that keeps your history on your own device, and the provider-side terms that decide whether your conversations are stored, trained on, or covered by a compliance contract. Each links to a fuller explainer.
Terms
- OPFS (Origin Private File System)
- A sandboxed, origin-scoped file system built into modern browsers that lets a web app store files privately on your own device, isolated from other sites and from your regular files.
- Zero-Retention API
- An AI provider API mode in which your prompts and outputs are not persisted after the response is returned — nothing is kept server-side to log, review, or breach.
- BAA (Business Associate Agreement)
- A HIPAA contract that makes a vendor handling protected health information legally accountable for safeguarding it. Consumer AI plans do not sign one.
- Training Opt-Out
- A setting or contractual term that excludes your conversations from being used to train or fine-tune an AI provider’s models.
- AI Hallucination
- A confident, fluent answer from a language model that is simply wrong — produced by the same next-token prediction that produces correct answers, which is why tone is no guide to accuracy.
- Context Window
- The maximum amount of text a model can consider in one request — instructions, the whole conversation so far, attachments and the reply, counted together in tokens.
- System Prompt
- Standing instructions sent ahead of your question that shape how a model behaves. Influential, but text like any other — never a security boundary.
- Prompt Injection
- An attack in which text a model is asked to process contains instructions, and the model follows them — because instructions and data share one channel.