What Data Does an Inline AI Editor Actually Send?

Before you trust an AI editor with your text, you deserve a precise answer to one question: when I press the hotkey, what actually leaves my computer? Not a reassuring slogan — the literal list. This page gives you that, so you can judge any tool (including this one) on facts.

The short answer

When you select text and trigger a cloud-based inline edit, a typical tool sends:

1. The text you selected — the words you highlighted. This is the core payload; the model can’t rewrite what it can’t see.
2. The instruction (prompt) — what you asked for: “fix grammar,” “make professional,” “translate to French,” or your custom prompt.
3. Sometimes: a window of surrounding context — a bit of the text around your selection, so the rewrite matches the document’s tone and topic (more on this below).
4. Minimal technical metadata — request formatting, model parameters, authentication. Not the content of your work, but it travels with the request.

What it does not need to send: your whole document, your file, other open apps, your clipboard history, or your keystrokes. If a tool sends more than the list above, that’s a question worth asking.

The context-window nuance (the part most tools are quiet about)

This is where inline editors genuinely differ, and where the privacy story gets interesting.

A naive tool sends only your selection. That’s the most private option, but rewrites can come back tone-deaf — they don’t know what document they’re in.

A context-aware tool reads a small amount of the surrounding text so the rewrite fits — matching the voice of the email, the formality of the doc, the topic of the thread. The privacy trade-off is direct: better edits require sending a little more text. A well-designed tool:

• Sends a bounded window (a few surrounding sentences/paragraphs), never the whole document.
• Is transparent about it — you can know, and ideally control, how much context goes.
• Lets you turn it off for sensitive material, accepting slightly blunter rewrites in exchange for sending only the selection.

The right answer depends on your text. For a public blog draft, more context is free quality. For a contract, you may want selection-only.

What stays on your machine

A well-built inline editor keeps a lot local by design:

• Your document as a whole — only the selection (± bounded context) is sent.
• Your custom prompts and settings — stored locally; your configs are yours.
• Local edit history / undo buffer — kept on-device so you can recover the original. The history of what you edited shouldn’t be shipped anywhere.
• Your API key (if BYOK) — stored in the OS keychain, used only to talk to the provider, never transmitted elsewhere.

And with a local model, nothing in the list leaves at all — selection, context, and prompt are all processed on your machine.

Questions that reveal a tool’s real behavior

• Do you send only my selection, or surrounding context too? How much, and can I control it?
• Is my whole document ever transmitted? (It shouldn’t need to be.)
• Where’s my edit history stored — local or your servers?
• Is anything retained after the request? (See no logging or retention.)
• For BYOK, how is my key stored?

Transparency here is the single best signal of a trustworthy tool. Vagueness about what’s sent usually means more is sent than you’d like.

How EditSnappy handles it

EditSnappy is built so the answer to “what leaves my machine?” is small and knowable:

• It operates on your selection, not your whole document.
• It’s context-aware by design — quietly reading nearby text so the rewrite matches the document — which means a bounded amount of surrounding context can be part of the request. (Exact context size and on/off control: see flag below.)
• No logging or retention of what’s sent — processed and gone, not stored or trained on.
• Edit history for undo is local, so your original is recoverable without shipping your history anywhere.
• You see the result as a diff before it commits, so you always know what the AI produced before it touches your work.

[[MISSING: confirm the exact context-window behavior — how many surrounding tokens/sentences EditSnappy sends, and whether the user can cap or disable it. master-sales-copy §5 confirms “context-aware” as a feature but doesn’t specify the window size or a privacy toggle.]] [[MISSING: confirm exact no-logging/retention wording with Ken (homepage v1 line 66).]]

See the full trust stack on the Privacy, Security & BYOK hub, or try EditSnappy free — no credit card.