Your project

Data, editing permission, and the terminal

What stays on your Mac, what goes to the model, how per-run editing works, and why stopping a run doesn't undo edits.

Reading code, asking the AI Reader a question, and running a terminal command each reach different things. This page covers what each one can see and change.

Where your data goes

What you do What happens
Browse, read, and search files Stays on your Mac
Go to Definition and other navigation Handled by the local language server for that language
Ask the AI Reader Your message and relevant code go to the model endpoint you set up
Let the AI Reader edit It can change text files inside the project you opened
Run a terminal command Runs as your macOS user, with that user’s permissions

A request can include code the AI Reader found while investigating, not only the file or selection you added. Adding a file sets where it starts; it doesn’t limit what it reads.

With a remote endpoint, the model provider processes that code outside your Mac. Before you send restricted code, check the provider’s terms and your organization’s rules.

When the app goes online

Yenpo has no telemetry or usage analytics.

Automatically check for updates is on by default. Yenpo checks when it launches or becomes active, with at least 24 hours between automatic checks. You can turn this off in Settings → General → Software Update and still check manually with Yenpo → Check for Updates….

The app uses the network in these situations:

What you do What it connects to
Check for updates automatically or manually The release list on yenpo.app, without project code, API keys, or device identifiers
Ask the AI Reader, sign in to a provider, or fetch the model list in Settings → Models The model endpoint you set up
Install rust-analyzer in Settings → Code Navigation GitHub, for the pinned official release, which is checked before it’s used
Click Retry when a Rust project is missing dependencies Runs cargo fetch --locked for that project, using your Cargo registries
Click a link Opens it in your browser
Run a command in the terminal Whatever that command reaches

Follow-up suggestions are a new development-preview feature: after an answer finishes, Yenpo may send that run’s questions and answer to the same model in an extra request, which uses your provider’s allowance or incurs charges. A failed suggestion request leaves the completed answer intact, and suggested questions are never sent automatically.

From version 0.1.293, language servers don’t fetch anything on their own: Cargo and Go run offline, TypeScript doesn’t download type packages, and crates or Go modules you already have keep working. Earlier versions already keep Cargo offline, but gopls and TypeScript Language Server can still download modules and type packages.

API keys and images

Enter API keys in Settings → Models. Provider API keys are currently stored in the app’s local preferences (UserDefaults), not in the macOS Keychain. Treat exports and backups of those preferences as sensitive, and don’t attach them to bug reports.

Image input is off by default for each provider. Turn it on only for a model and API that support it, and only attach images you’re fine sending to that endpoint. A screenshot taken for one small detail can still show private code or credentials elsewhere on the screen.

Reading by default, editing per run

The AI Reader only reads unless you say otherwise. In a normal conversation, you can turn on Allow Edits. The permission covers that one run; later runs and forked conversations start read-only again.

With editing allowed, the AI Reader can replace text in, create, or delete text files inside the project. It can’t run shell commands or change Git state. Git comparison conversations are always read-only, whatever the editing setting.

Before an editing run, know which uncommitted changes you already have, so you can tell the AI Reader’s edits apart in the diff afterwards. Keep using version control and backups; the permission switch can’t bring a file back.

Stopping doesn’t undo

Stop asks the run to cancel wherever it can. Changes already written to disk stay there, even if the model never finishes its answer.

After an interrupted editing run:

  1. Look at the affected files and the working-tree diff.
  2. Work out which changes were completed.
  3. Keep, adjust, or revert them with your usual tools.

Don’t assume nothing changed just because the answer stopped or ended in an error.

The terminal is separate

Yenpo’s terminal runs as your macOS user. It starts in the project folder, but it isn’t a sandbox: commands can reach anything your user account can.

The AI Reader can’t run commands in the terminal, and it doesn’t see terminal input, output, or history. If you paste terminal output into a message, that text becomes part of the conversation, so look it over before sending.

That also means a command or test in an AI answer is only a suggestion. It hasn’t run until you run it.

Website privacy

The website’s analytics, browser storage, and hosting are covered in Website privacy. None of that affects the app or your model provider.