AI providers

Settings → AI is where you pick which AI provider powers every AI feature in Trace. For a decision-focused comparison of the six options, see Choosing your AI; this page covers the Settings tab itself in more depth.

Providers

• Self hosted (default): Breadcromb's managed Azure AI Foundry infrastructure. No key or endpoint needed, works immediately.
• OpenAI: paste an API key (starts with sk-). Trace verifies it live.
• Anthropic: paste an API key (starts with sk-ant-).
• Gemini: paste an API key (starts with AIza).
• OpenRouter: connect your account through OAuth. No key to copy or store.
• Ollama: local and self-hosted, no key needed. Runs a local model (qwen3:8b by default).

PII redaction

Every cloud provider except Ollama has a "Redact PII before sending to {provider}" toggle. Turning it on strips obvious personal details from a request before it leaves your device. It's hidden for Ollama since a fully local model never sends anything anywhere to redact from in the first place.

Embedding health

A live indicator shows whether the embedding model (used for semantic search and personal memory) is working correctly, with a manual re-check available if you ever suspect it's stuck.

Per-feature model selection

Settings controls the active *provider*. If that provider offers more than one model, choosing which specific model handles a given message is done in the chat input bar's own model picker, not here.

Next steps

• Choosing your AI for the plain-language comparison of all six options.
• Privacy & data for what does and doesn't get sent to whichever provider you choose.