Models & AI providers

Choosing a reasoning model

Under Settings → Reasoning model, pick the model that writes this workspace's answers. The default is Auto, which names exactly which model it resolves to today - it's a sensible default, not a hidden different thing. Below it, models are grouped by provider (Anthropic, OpenAI, Google) with a plain-language description of what each is good at.

A model you can't route yet - for example, Gemini before you've added a Google key - isn't hidden: it shows an Add your key badge that takes you to the provider-keys section, and unlocks the moment a working key is saved.

Bring your own keys

By default, answers run on Rationale's own provider accounts. Under Settings → Model providers you can instead paste your organization's API key for a provider - the key is checked against the provider when you save it, stored encrypted, and from then on that provider's AI traffic for this workspace runs under your own agreement. Claude and GPT models are available either way; Gemini requires your own Google key.

• Usage caps. Each key has a rolling daily token cap, so a runaway job can't quietly spend against your account.
• Fail closed, visibly. If your provider rejects the key (revoked, out of credit), Rationale pauses it and the affected work stops with a clear notice - it never silently falls back to a different account.
• Removable anytime. Delete a key and it takes effect immediately; the workspace falls back to what it can serve without it.

One honest limit: the search index that powers retrieval (embeddings) always runs on Rationale's account - your keys are used only for the reasoning model that writes answers.

What stays true regardless of provider

Whichever model writes the answer, your content is sent to that provider only to produce the answer, and is never used by Rationale to train AI models. Every answer still names the model that wrote it, and still cites its sources - see How Rationale reads your artifacts.