Privacy¶
IntentumDiff runs entirely on your machine. Parsing, matching and classification are local. By default nothing leaves your computer, and there is no telemetry.
The optional LLM explainer¶
IntentumDiff can ask a language model to explain a change in prose. It is off by default, opt-in, and bring-your-own-key.
- We never bundle an API key and never run a paid proxy on your behalf
- Your key is stored in VS Code SecretStorage, never in
settings.json - A consent prompt precedes any send to a cloud endpoint
What is actually sent¶
To a cloud endpoint, IntentumDiff sends a privacy-safe fact sheet: counts, categories and boolean flags derived locally from the syntax tree.
| Sent | Never sent |
|---|---|
| Change categories | Your source code |
| Structural facts, e.g. "a guard clause was added" | Identifiers — names of functions, variables, files |
| Counts, e.g. "3 conditionals added" | Literals — strings, numbers, secrets |
The fact sheet is designed so that the original source cannot be reconstructed from it.
To a local endpoint — a model running on your own machine — verbatim source may be sent, because it does not leave the machine.
Token carriage¶
Where a fact could carry a source token, carriage is governed by an explicit policy that defaults to none and fails closed. You choose, per category, how much the model may see — from nothing at all up to full tokens. If the policy cannot be determined, nothing is carried.
Tests do no network¶
The test suites perform no network calls, so running them cannot send your code anywhere.
Reporting a concern¶
If you believe IntentumDiff sends something this page does not describe, please treat it as a security issue and report it privately through the repository's security policy rather than a public issue.