Choosing your AI

Trace doesn't lock you into one AI vendor. Every AI feature, chat, summaries, skills, and the navigation agent, runs through whichever provider you pick in Settings → AI. You can switch at any time, and different features can even use different models under the hood without you managing that yourself.

The six providers

Provider. Cost. Setup. Runs where

Self hosted (default). Included with Plus/Pro, limited on Free. None, works immediately. Breadcromb's Azure AI Foundry infrastructure

Ollama. Free. Install Ollama, pick a local model. Your own machine

OpenAI. Pay-per-use (your key). Paste an API key. OpenAI's servers

Anthropic. Pay-per-use (your key). Paste an API key. Anthropic's servers

Google Gemini. Pay-per-use (your key). Paste an API key. Google's servers

OpenRouter. Pay-per-use (your account). Connect via OAuth, no key to copy. Routed through OpenRouter

Trace defaults to Self hosted, so the app is fully functional the moment you finish onboarding. There's nothing wrong with staying on it, but it's worth understanding the alternatives.

Local and free: Ollama

If you'd rather nothing about your browsing or AI usage ever reaches Breadcromb's servers, install Ollama and switch the provider to Ollama in Settings → AI. Trace runs a local model (the default is qwen3:8b) entirely on your machine. It's free and private, at the cost of needing enough RAM and a one-time setup step outside the app.

Bring your own key

If you already have an OpenAI, Anthropic, or Gemini account, paste that provider's API key into Settings → AI and Trace uses it for every AI feature. You pay that provider directly, at their normal rates, with no markup from Breadcromb. OpenRouter works the same way but skips the key entirely: you connect your OpenRouter account once via OAuth and Trace routes requests through it.

Note:

A "Redact PII before sending to {provider}" toggle is available for every cloud provider except Ollama (which never leaves your machine, so there's nothing to redact). Turn it on in Settings → AI if you want obvious personal details stripped from a request before it's sent out.

Managed AI: Plus and Pro

Free tier users on Self hosted get a limited amount of usage; Ollama and bring-your-own-key have no such limit since you're providing the compute or paying the provider directly. Upgrading to Plus or Pro removes the Self hosted limit entirely and, on Pro, switches to a higher-quality model. See Plans & billing for pricing.

One more thing: per-feature models

Settings only controls which *provider* is active. If a provider exposes multiple models, picking exactly which model handles a given chat message is done from the model picker in the chat input bar itself, not in Settings.

Next steps