Skip to main content

Store AI credentials

For the concept of providers and how keys are resolved, see Choose a translation provider; this recipe walks the desktop's AI Models settings.

Goal: save a provider's API key once, securely, so every translate, qa, or rewrite step that needs it just works, and the key never lands in the committed recipe.

Outdated recording

The interface has changed since this was recorded. The workflow it demonstrates still applies, but some screens and labels differ from the current release. The clip shows an earlier settings layout. Terms and Content Memory sit under Context, and the AI Models tab shows which model runs are using. The clip covers the whole settings area (appearance, providers, and plugins), so it ranges beyond credentials and does not follow a key through to a translate step.

Steps​

  1. You may not need a key. On first open, the Connect your AI card offers what it detects: a Claude Code sign-in (uses your Claude subscription) or a local Ollama model, one click, nothing stored. The steps below are for API keys.
  2. Add the credential. From Settings → AI Models, add a credential: pick the provider (Anthropic, OpenAI, Gemini, …) and paste its key.
  3. Stored in the keychain. The key is saved to the OS keychain and never written to the kapi.yaml recipe, so the recipe stays safe to commit and share.
  4. Use it. A translate step (in the flow editor) or any AI-backed tool resolves the credential automatically; no key on any command line.

Why the desktop​

Credential handling is safe by default: keys live in the OS keychain, the recipe carries only the provider name, and the app shows which providers are configured. The same keychain store backs the CLI's kapi credentials, so a key added here is available to command-line runs too.

Next​