Choose how assistance runs
First-run setup and Settings → AI Assistant let you choose a provider, model, endpoint, and commit style. Options include hosted providers, a local Ollama server, and offline rule-based commit drafts.
Save & test saves the settings and sends a built-in example to the configured model, without reading repository content. Normal AI features send selected change content only when you explicitly request assistance.
Review a message before applying it
Recompose commit message with AI… generates a draft for a selected commit. Edit it and choose explicitly whether to apply it; generation does not automatically publish or push anything.
Recomposition requires a clean local branch with at most 500 linear commits. A restore point protects the old history. File trees and author metadata are preserved, while rewritten hashes and cryptographic signatures change.
Explain a branch from its base
Explain branch changes… compares selected base and tip revisions from their common ancestor. The result reports the resolved revisions, commit and file counts, and any large-diff truncation.
Use an explanation as context alongside the actual patch and your own review. The configured provider receives the requested input; a local Ollama endpoint keeps that model request on your machine.
Keep credentials separate from interface preferences
API keys are encrypted through Electron’s operating-system keychain or keyring integration. Model and provider preferences are stored separately in the stable app data directory and survive normal updates under the same OS account.
A blank key field keeps a saved key. Removing a key is explicit, and switching providers preserves other saved credentials. Resetting interface preferences does not delete AI keys or restart onboarding.